Angry_Games Posted January 26, 2006 Posted January 26, 2006 Ok...this has been the subject of MUCH discussion ever since the A64's first started clocking and we found out that dividers really don't mean diddly squat, and really neither does CAS latency, nor memory timings, nor anything... You've all heard me preach OVER and OVER again about how cpu Mhz is the king...cpu Mhz is the king...cpu Mhz is the important thing, forget about damn divider penalties and this or that I've just entered two entries into the 939 Overclocking Database. they are the same machine, overclocked to 275x10, with two different sets of memory, BH-5 on a divider, and TCCD running 1:1 (no divider). first the machine specs: * * GEIL ONE BH-5 @ 2-2-5-2, DDR180 Divider DFI LANPARTY nF4 SLI-DR - 7/04-2BTA BigToe BIOS AMD Athlon64 3200+ Venice 2x512MB GEIL ONE BH-5 (GOW1GB3200DC) XFX Geforce 7800GTX PVT70FUNF7, 489/1314 78.03 drivers Maxtor 160GB 8mb cache IDE hdd ASUS CD-S520/A5 Black IDE CD-ROM Drive OCZ PowerStream OCZ600ADJ ATX 600W Power Supply - Retail * * the second machine, is EXACTLY the same as this, except with Centon Advance 3200LL TCCD @ DDR200: DFI LANPARTY nF4 SLI-DR - 7/04-2BTA BigToe BIOS AMD Athlon64 3200+ Venice 2x512MB Centon Advance PC3200LL (TCCD) XFX Geforce 7800GTX PVT70FUNF7, 489/1314 78.03 drivers Maxtor 160GB 8mb cache IDE hdd ASUS CD-S520/A5 Black IDE CD-ROM Drive OCZ PowerStream OCZ600ADJ ATX 600W Power Supply - Retail * * same BIOS, same everything...different memory (and of course with different memory, come different timings) * * the differences side by side? This table shows initial settings in the GEIL BH-5 column for 275x10 @ DDR180. On the rest of the BH-5 w/dividers, the settings stayed EXACTLY the same except for dropping the dividers down * * So both are running at 2750Mhz...both got there by slightly different paths Now for some interesting comparisons....benchmark scores (from the Overclocking Database) UPDATE - DDR200, DDR180, DDR166, DDR140 comparison tables, and now with 300x9 to compare to @ 2.5-4-8-4 The only thing that 1:1 275x10 beats divider (9:10) 275x10 in is memory bandwidth...which sorta proves my point that memory bandwidth is useless, and that it is all about cpu mhz, as well as proving that dividers are not always a performance penalty (unless, for some reason, memory bandwidth is that important to you...) Now sure, if I could magically get my 275x10 BH-5 to do 1:1, you'd see a bigger increase in memory bandwidth, as it (BH-5) would trump because of same actual memory speeds but tighter timings. Now, I am priming the same machine @ 280x10 (DDR180) just to see if it will prime long enough to get into the Overclocking Database at a higher speed, then I will again bump it up to 285x10 and test again and just keep going until I find the max of this GEIL ONE BH-5 (last GEIL Ultra-X BH-5 that came through my hands did 260 (DDR520) without a hitch, and it belonged to a customer, so didn't have time to maximize it) as well as this 3200+ Venice. I'll leave this stuck to the top for a while for you guys to discuss and debunk what I have done, while I continue testing this rig...and when I am done maxing it out, I will then drop back to 250x10 and do straight 1:1 comparing since I know the BH-5 will do 250x10 1:1 easily (already Primed it at those speeds, and it is doing 250Mhz on the DDR180 divider at 275x10). The TCCD will of course do it...the differences then will be purely in what the different timings will net you (which won't be much difference either way, again making timings a moot issue unless you just HAVE to have the best benchmarks on the planet). I will also run this 2750Mhz at DDR166 divider and show you the performance differences (2-2-5-2 on BH-5) at a slightly lower divider...which will again show you very little difference. Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 ok, everyone wanted me to do this, and I got a bit of it done. This sorta follows the same path as the ever-popular "tight timings vs dividers" thread. here's the skinny on the setups involved: NF4 SLI-DR 7/04-2BTA bios AMD Sempron 3200+ 939 cpu (Winchester core, 1800Mhz stock speed, 256k L2) OCZ PC4000EB 2x1024, 3-3-7-2 (all tests, all DRAM timings were the same) Sapphire X800GTO2 (modded to 16-pipe) OCZ Powerstream 520w psu Maxtor 160GB IDE hdd Asus 52x CD-ROM on the first block of NF4 tests, the first set were run @ stock cpu speed, 1T, with the gpu @ default of 400/490 the second set were run @ stock cpu speed, 2T, with the gpu @ default 400/490 the third set were run @ stock cpu speed, 1T, with the gpu @ X850XT Platinum Edition speeds (540/590) the fourth set were run @ stock cpu speed, 2T, with the gpu @ X850XT Platinum Edition speeds (540/590) on the second block of NF4 tests, the first set were run @ 250x9 (2250Mhz), 1T, with gpu @ default 400/490 the second set were run @ 250x9 (2250Mhz), 2T, with gpu @ default 400/490 the third set were run @ 250x9 (2250Mhz), 1T, with gpu @ X850XT Platinum Edition speeds (540/590) the fourth set were run @ 250x9 (2250Mhz), 2T, with gpu @ X850XT Platinum Edition speeds (540/590) notice in the chart below that one set is missing Everest read/write/latency scores...I must have forgotten to write them in, so I just left them as n/a and hope you guys realize that that particular set that I missed writing them down really wasn't all that important (only the gpu speed had changed) specs for the NF3 single-channel testing: NF3 250Gb socket754 Athlon64 Turion ML-32 (1800Mhz, 512k L2) 2x1024MB OCZ PC4000 EL @ 3-3-8-3 Sapphire Radeon 9500NP (modded to full 9700Pro, 300/300) WD Raptor 36GB SATA hdd Asus 52x CD-ROM Enermax 600w psu the first set of NF3 tests were run using stock cpu speed, 1T, 300/300 video card the second set of NF3 tests were run using stock cpu speed, 2T, 300/300 video card the third set of NF3 tests were run using 275x8 (2200Mhz), 1T, DDR166 DRAM divider, 300/300 video card the fourth set of NF3 test were run using 275x8 (2200Mhz), 2T, DDR166 DRAM divider, 300/300 video card now for the chart: as you can see, there is very little difference in any of the scores from 1T to 2T. I've heard many claim that dropping to 2T killed their performance, or that they scored 4,000 less points on 3dmark2001, etc, and the only thing I can think is that they did something incorrectly...as this took me a good bit of time to do this testing, and the only thing I ever changed from test to test really was 1T or 2T on the NF4 testing I changed the gpu/mem speeds on the vid cards at both 1T and 2T just to see what happened on the NF3, I left the poor AGP card at 300/300 through all the tests I also thought maybe overclocking the cpu itself would show a big difference, but the only difference it really showed was what we expected it to show, an increase in overall performance (ie faster cpu mhz is going to trump slower cpu mhz, which is just plain common sense lol) so I know a few of you will argue about my timings, or my settings, or something (some of you just like to argue). Most of you I hope will take a look at the chart, the differences, and realize that just like dividers etc in the original thread, there's very very little difference to be had between 1T and 2T. so anyone wishing to go from 2x512 or 2x1024 to 4x512 or 4x1024, you can expect a little hit dropping to 2T, but what everyone seems to forget, is that dropping to 2T can sometimes (usually more noticeable when only running 2x dimms instead of 4x dimms as 4x dimms makes the memory controller really work harder) give you a nice big fat increase in overall memory speed ability, which can easily translate into cpu mhz example 1T i might be limited to 275x9 @ 2475Mhz but if i drop to 2T, i might be able to get 302x9 @ 2718Mhz the increase in cpu mhz is going to negate ANY penalty incurred from dropping to 2T I hope everyone finds this informative, and if you wish me to drop some more testing in on different setups (ie: BH-5 which i didn't cover on this, I only used 2GB memory kits), then let me know and I'll try to arrange some extra testing on a station, then add it into this post ps: i guess I forgot to mention that the real important benchmark numbers, for me, and I assume for most of us here, is the 3dMark2003 and 2005 scores, as those are very reflective of modern games that use DirectX9 fully, and are more limited to gpu performance than they are to cpu performance pss: my ML-32 Turion mobile 754 is a big piece of crap....I have never been so disappointed in a cpu before. I plan to eventually remove it from the NF3 250Gb and drop my trusty Sempron 3100+ or 2600+ in as I am well aware of just how well the Semprons clock...but the ML-32 was BARELY able to pass 275x8, and only on a divider...and even then I had to fight it quite a bit... Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 placeholder #1 Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 placeholder #2 Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 placeholder #3 Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 once again I am having to close this thread because certain know-it-alls have come to argue that I don't know what I am talking about, yet they show NO PROOF WHATSOEVER to back up their claims that I am an idiot. Once again, I've run out of patience for anyone who cannot show proof otherwise that their claims are true (I think I have shown more than enough proof...don't you?) If you wish to argue them, then you had better have appropriate proof to back them up (all of the testing from the Overclocking Database requirements etc). (when I open this thread up for discussion again that is...) I didn't go through all that testing just to have someone claim without proof that I am a retarded idiot who doesn't know what I am talking about when I clearly have shown you with my results and testing that the numbers don't lie. More as I get more testing in. If the demand is great, I will once again open this thread for discussion. Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 Thanks for the interesting study, AG. It's got some new ideas poppin' up for me here! Definitely gonna have a lot more testing to do..... (Update 11-25 with testing results) I did some testing with Everest at various speeds with the following results: 1. No real difference in "Read" 2. -10-12% in "Write", seems to increase with HTT speed. 3. +5-10% in "Latency", but difference seems to decrease as speed increases. From: 1T vs 2T discussion thread Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 @ Angry; so what you are saying is that you would rather use memory running 3-3-3-8 @ 2T at DDR600, than memory running 2-2-2-5 @1T at DDR400? I only ask because I don't quite understand your first statements at the top of the page and I am trying to understand better. I am sorry if this is a stupid question, I am still quite new to all of this. basically what I am saying is...it really doesn't matter how you get there...as long as you get there memory bandwidth is irrelevant on an A64 because the memory controller is built into the cpu...you can never really give enough memory bandwidth to the processor because it accesses memory directly...there's no Northbridge (memory controller hub) in between them. and memory bandwidth doesn't really mean much in real applications like games and photoshop and this or that. what matters is the cpu Mhz. Look at my numbers...I got to 2750Mhz from 2000Mhz stock in a few different ways... 275x10 (DDR550, no divider)? No problem with TCCD memory and this cpu. It is loser timings, but those timings are easily able to let all of my TCCD memory do 275 (DDR550), without a divider. 275x10 (DDR180 divider)? No problem, as my BH-5 that can do tight timings can easily get me there at DDR180 and DDR166 and DDR150 dividers...with the tight timings, performance loss is practically nil when running a divider 300x9 (DDR600, no divider)? A little harder to get to, but with the right TCCD it is a little easier, but still a very real challenge to get 300HTT (DDR600) out of any TCCD...with it fully stable (not suicide screenshots to show off some benchmarks). The 300x9 I showed was fairly stable, but I never got to put it in the OC Database as it started taking too much of my time getting 300x9 fully stable =/ and look at 300x9...it is 2700Mhz then look at 275x10...I can get this (as my cpu is easily capable of it apparently lol) with about 10 or more different kits of RAM I have. I can even get this easily from my OCZ PC4000EB 2x1024MB modules. Basically...as long as my RAM can do 275, my cpu can do 2750. And with 10+ kits of RAM able to get me there....I've got choices on how I want to get there (tight timings, but with higher voltage with BH-5? looser timings, lower voltages, but no divider with TCCD?) compare the benches for 300x9 and 275x10 in the chart at the top...pretty much the same scores, with 300x9 winning a couple, but the 275x10 winning what I consider the more important ones for gamers (3dMark03 and 05). But even so, there's very little difference in them...and only 50Mhz difference in raw cpu speed. then compare the BH-5 benches with 2-2-5-2 timings...even on a DDR140 divider, they are still as good as 275x10 and 300x9 TCCD...better with the higher divider (closer to 1:1) you use. but and this is the most important part I have tried to get you guys to get into your minds... but look at the difference between all of those different ways to get to 2700/2750Mhz on the cpu. You got advantages and disadvantages depending on which route you go all for the sake of a couple percentage points difference. to me, the important thing isn't the biggest number (to some, I guess they live and die by this stuff...but I pay money for good hardware to ENJOY it after I overclock the snot out of it). the most important thing is the flexibility, knowing that no matter HOW I get to a clock...that I am guaranteed almost identical results, with up to 3% variation in real performance, depending on which way I choose. A lot of users still think that high HTT (or FSB as we used to call it lol) is so damned important that they just have to hit 300x whatever. And they get bullheaded and sometimes downright nasty if they cannot achieve 300x whatever, as if it were their God-given right to do so just because they bought this or that, or a review said they would. What this thread is about, is letting people know something important. It really doesn't matter how you get there, as long as you get there. Try going 5 or 6 different routes to work for a while. Once you maximize 5 different routes to work, you sorta realize that there is no 'shortest' route...it takes you within 1-5 minutes no matter which route you take to arrive from your house to work. Some routes might be really long and out of the way. Some might be really short but pretty hazardous (lots of roadwork? bad area of town where they hate nerds like you?) Overclocking is something that you have to generally work very hard for to feel really justified, and truly proud of what you have done. Some of you feel overwhelmed with joy just hitting 250Mhz overclock, because you never did such a thing before, and no amount of reading and asking really made you any less nervous the very first time you rebooted the machine after overclocking that 250Mhz to see if anything would burn up, blow up, spark up, bust up, or even turn back on. Some of you only feel truly rewarded after squeezing out a 1000Mhz overclock from some poor Sempron or Mobile or 3000+ Venice, the whole time listening to it's cries like a damsel tied to a railroad and the 3 o'clock Eastern Pacific rumbling the tracks close by. What you can be comforted in, is that if you have the right parts to do some overclocking, and can get overclocked, it really doesn't matter how you get overclocked...as long as you get overclocked. its getting that cpu clocked that is important. it doesn't really matter how you get there...as long as you get there (unless obsessed with having to have the bestest most absolutest score to brag to your clan members about while getting your yak beat down by some South Korean Nat'l Team BF2 clan members to show them how tough you and your machine be lol) SOMETHING IMPORTANT: this does not exonerate you from purchasing the right parts to start and finish this journey. no sir the first guy that comes back and says 'so this means I can buy some cheap-butt VALUE memory and blammo I am home free?'...I'm gonna boot him over the pigpen fence, landing hopefully face first. This does not mean that VALUE memory won't work. I've seen plenty of users with it...and I have to say a majority of overclocking problems stem from this need to purchase the most complex, best overclocking motherboard on the planet, and then grab some VALUE memory that couldn't even make it as 'regular' memory (whining mostly). I explained it a little harshly before I closed the other thread...but I am going to put it to you in clear, plain English, one more time, in hopes that you understand. What I am going to tell you cannot be argued with, because there is nothing to argue. There is only the truth. There is no other explanation, because there is only the truth. No amount of arguing or even believing, no matter what I say, is going to change the truth of what I am going to tell you. This board (Lanparty NF4 and NF3 250Gb), was never designed for running VALUE memory. It never entered the designer's head to test for VALUE memory compatibility and overclocking ability. It never once occurred to him to try overclocking with VALUE memory. And I will tell you why. Because these boards were DESIGNED, from the initial idea, to the final production, to be overclocking boards that nerdy little boys had wet dreams about. These boards, when the designer dreamt them up, had never had anything of the low grade like VALUE, even enter the ionosphere of the dream. These NF4's were out at the end of January of this year. It was not until Oskar added the "DRAM Response Time" option in the 6/23 BIOS that many VALUE memories finally were able to work at least at stock speeds. That's almost 6 months...why so long? Because these boards were never designed to run VALUE memory of a low grade that couldn't even cut it as 'regular' or 'decent', or even 'fairly well, but not really even our signature nor ultimate line...not even our normal line really'. The designer used PERFORMANCE memory, because we all know, you use PERFORMANCE items to get maximum PERFORMANCE. Notice I did not say 'the most expensive memory'. Because the OCZ BH-5 2x512's that were $119 at one point on Newegg for the 1GB kit, would do those 2-2-5-2 timings @ 250, or 275 on a divider, just like teh GEIL ONE BH-5's, the GEIL Ultra-X BH-5's, the OCZ 4000VX, Mushkin Redline...as were the $155 OCZ Plat rev2 TCC5/TCCD modules...just at looser timings with lower voltages and no divider. Expensive memory was not performing as well as good solid TCCD and BH-5. Some of you bought the most expensive BH-5's or TCCD's that you could find, others waited a little and bought yours at a stellar deal and got more than your money's worth. But either way, you didn't buy the most expensive memory. You bought stuff that others were using and showing off as being great for them. Those of you that bought VALUE memory, have found out the hard way...even getting it to run at stock speeds. A very small few of you were even able to get your VALUE memory to overclock decently. A couple of you showed some other guy in another forum using VALUE memory claiming that it was just as good as the more spendy PERFORMANCE memory and the even more expensive PREMIUM memory. If you broke down our Lanparty board to any other board...there are some significant aspects that probably 99% of you don't know, that makes a huge difference in why they overclock so well, but are so picky about hardware. Most of you know that Hynix modules don't work so well, and others have real mixed luck with 1GB kits of Ballistix or Infineon. This is because the designer, at the time of him dreaming up this board that everyone wanted, wanted to make a board that was so good as to surpass all expectations in the overclocking community for it's ability to overclock. To do that, you have to make some choices. One of those choices was to optimize the board for the two most popular types of memory...BH-5, the old standby that was making a comeback, and this new TCCD, that was incredibly flexible at low voltages. To make TCCD run as good as it does, it required modifying the design to where hynix would not do well at all for overclocking. This was a sacrifice that, in my opinion, was done correctly. This and other decisions all added up to one thing...the board was a freakin monster when you overclocked it with some decent TCCD or BH-5 (or even CH-5). I don't think once, one time ever, did the designer even have any thought of VALUE memory. In fact, I remember when mentioning to it about the complaints from users trying to use VALUE memory, he was as much at a loss as to why anyone would put VALUE memory in a PERFORMANCE board to OVERCLOCK and MAXIMIZE, as I was in having to tell him lol. It was like two old women sitting around complaining to each other "well why the hell would you put VALUE memory in this board? You bought this board to OVERCLOCK! That is what it was DESIGNED for!" Finally, after we finally decided that we'd lost the battle (too many people whining that 'boo hoo these boards won't run my cheapo VALUE memory or psu') of trying to explain that these boards were never DESIGNED for VALUE anything...their ONLY DESIGN was to overclock the crap out of good TCCD and BH-5 and possibly some other decent memory...we got the designer to figure out how to make this cheap VALUE stuff run correctly....when it was never designed to run such memory. The moral of the story is.... you might be able to get there in a Honda Civic...but your Honda Civic really wasn't designed to do Baja 1000 runs at top speeds. now please people.....I am just as big of a dorky nerd as all of you (whether you admit it or not lol) And I surely like to see myself get crazy good good scores on my benches when I am overclocking and fooling around or building a personal machine. But in doing this every single day for the last 3 years with multiple machines and kits of memory and psu's and this and that and the other...there's one thing I think I have got a lock on.... and that is with the Athlon64 cpu, and some decent memory, a decent power supply, and of course a decent vid card...not the most expensive crap you can find either...you can make it outperform even the most expensive crap you can find lol. And if you show me 275x10 with 10% higher benchmarks....am I gonna go 'oh no i got to go fix my low weak girly benchmarks'? No, I am going to continue playing F.E.A.R. and Call of Duty 2, and Quake4, and Age of Empires III etc, because my rig works just fine. But your 275x10, with the same hardware I am using, will be within 3% of mine, with it really being closer to .2%-.5% This obsession with the greatest, biggest, giantest (? lol) benchmark scores is sorta worthless. For some, if that is what they want, then great. But most of us don't have a lot of money to keep buying parts until we get the 'premium stepping' etc. Most of us just want to dabble with it and get to a certain level that we are comfortable with, then get the hell on with our installing of games and porno and music and movies and applications. If you got to have the most most most out of it all the time....then this whole thread is not really for you, because you aren't interested in seeing results that show 2750Mhz is 2750Mhz just about any way you slice it. You want to argue it to death that your way is the better way and my way is the weak girly way and that isn't true for the majority of us. And we aren't really interested in hearing how you can tighten your timings down to 1-1-1-5 at 438x6 and blow all of our scores out of the water and that it is the most important thing on earth is to show superior benches. For the majority of us that are interested in this thread...we want to know...do I really have to be able to do 300x9? Can I do 275x10 and still get good performance widening my choices in what to buy or expect? Is there really that much of a difference in real performance that I am going to notice? Me... I just want you guys to overclock something, and enjoy it I like to enjoy the fruits of my hard labors my fruit at the moment is F.E.A.R. with 4x1024 @ 245x11 with an X2 4400 @ 2700Mhz, and a 7800GTX. I worked for almost 11 days putting that mother together and tweaking it to the 2700Mhz plateau with all 4GB of memory in it...but most of the time was spent getting 2 dual-120mm radiators in a Stacker lol. I'm down with 2700Mhz. I want to go play now I could probably get 2750, or possibly even 2800-ish...but that could take me a whole month or two...it took me 11 days to get from 2200 to 2700 exactly how I like it (and plenty of RAID-0 testing for the thread stuck at the top of the Off Topic section about RAID/hdd's)...for another 50-150Mhz I don't even want to think about it. It doesn't matter how you get there really....as long as you get there...and have the right equipment to make the journey. and most importantly enjoy your stuff....you bought it to enjoy it...so take a break once in a while and enjoy it Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 I have done some test to see dif combos in my TCCDs. Here you have the results... The best configuration (by far) for my system is without doubt the 300ish FSB salu2. From: Tight timings vs Loose Timings vs Dividers thread Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 Ok now I tried some things and see what I got: Config 1: DFI LANPARTY nF4 SLI-DR - 7/04-2BTA BigToe BIOS AMD Athlon64 3200+ Venice@260x10 @1.4Vx103% 1 x 512MB OCZ Gold EL PC3200 BH-5 @ 2-2-5-2-7-16, DDR180 Divider (DDR468) @ 3.0V Gainward Geforce 7800GT, 495/1250 81.95 drivers WD Raptor 36Gb Levicon 420W (My OCZ PowerStream 420W died yesterday) 3dmark05 score: 8234 Config 1: DFI LANPARTY nF4 SLI-DR - 7/04-2BTA BigToe BIOS AMD Athlon64 3200+ Venice@290x9 @1.525Vx113% 1 x 512MB OCZ Gold EL PC3200 BH-5 @ 2-2-5-2-7-16, DDR166 Divider (DDR474) @3.5V Gainward Geforce 7800GT, 495/1250 81.95 drivers WD Raptor 36Gb Levicon 420W 3dmark05 score: 8244 Difference: 10, that can be from any slight misconfiguration or so. That proves HTT does nothing. And that raising: Vcore from 1.4Vx103% to 1.525Vx113% LDT to 1.2V 1.4V Chipset from 1.5V to 1.8V Vdimm to from 3.0V 3.5V and Idle temp from 29 to 35ºC As I don't have a proper PSU I can't try to get harder in DRAM speeds, but I don't think theres too much difference in 3dk05... BTW I have screens, but they are 1280x1024 and I don't think thats ok to post in the forums. How do you ppl capture a screenshot with a single focused window? or you do the paint job and crop it? :nod: From: tight timings vs loose timings vs dividers thread Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 right... but what about DDR400 @ 3-4-4-10 2T -vs- DDR500 @ 2-2-2-6 1T ...shouldn't that just be a "major" performance difference? (this might be the part where I have to eat my tin foil hat) you answered your own question and it again goes along with the point of this thread...look at your scores from your 4 different benchmarks...how much difference is there? run a bunch of different benchmarks and measure them all like I did with different timings and dividers now take a look at hte difference in scores not much is there? (if there is...there must be something odd about how your setup is configured or such because we've repeated these results over and over and over...which is why this thread is here) In summary this thread is basically saying that when it comes to overclocking your computer' date=' don't skimp on CPU performance to gain on RAM performance. After you max your CPU, go for the highest speeds and tightest timings you can manage on your RAM-including the use of dividers if necessary (the original argument was that 1:1 is better for some reason). But if you can't get 1:1 while keeping your max CPU speed, a divider doesn't hurt performance (in reality). I hope that summed it up ok, obviously all RAM is not equal so it is difficult to compare between two different brands with two different chips. Obviously DDR500+ will certainly give better gaming performance than DDR400, but if you can't afford everything at once its a viable alternative to use what you have the best way you can.[/quote'] dan is exactly right...when overclocking your computer (heck when running at stock speeds also), comparing cpu Mhz to cpu Mhz while using different memory speeds and timings is going to show you what this thread has already shown you...there is so little difference that it is a moot point...and by loosening up the timings you might achieve more cpu mhz which will almost always improve whatever score you are comparing. 192.5mhz/2-2-2-5(slow running RAM with fast timings) ~performs the same as~ 275mhz/2.5-4-4-8 (fast running RAM with slow timings) this is like say'n a VW bug on nitro runs as fast as a Porsche with four flat tires (right?...that could be true) at the same cpu overall mhz...as this thread has shown you at the beginning...there's very little difference...look back at the first post... plenty of 2-2-5-2 tests...and a few 275 1:1 tests at 2.5-4-8-4 etc....notice the differences in scores? sure you do...in Everest...but not much else...a couple percentage points here and there...some live and die by that 1.252% difference....but the majority of us want to get it clocked and stable and then enjoy it because there's no real professional overclocking league to impress....but there's plenty of games and applications that people want to use on the computer they just spent a good chunk of money on to enjoy to use for playing games and using applications... Ok...this has been the subject of MUCH discussion ever since the A64's first started clocking and we found out that dividers really don't mean diddly squat' date=' and really neither does CAS latency, nor memory timings, nor anything...[/quote'] there is nothing unclear about this statement...I've shown (as have others in this very thread) that my statement is very true...and the only persons who really have trouble accepting this are those that consider themselves so elite that they have to have the tip-top performance so they can brag at forums and to their buddies that they scored this or that in some benchmark. the point of my thread is to let people know that dividers and loose timings are a moot point with the integrated memory controller on teh A64/Sempron cpu...and as of this very moment, no one has been able to disprove this, unless, again, you are one that is so concerned with 2% difference in performance (in a particular benchmark, but my results show different results across the board...tighter timings gives sometimes better 3d scores, but looser timings with higher memory mhz gives bigger memory bandwidth scores...). And lets be honest...gamers might crow about that 1% or 2.32% difference in framerates, but most (and I mean overwhelming majority) only care that they can now use 1280x1024 with 2xAA/8xAF and all the graphics options on maximum since they overclocked their cpu from 2Ghz to 2750Mhz. Or that Divx might encode their porno in 2 minutes and 9 seconds instead of previous 2 minutes and 19 seconds... to keep the RAM at lower speeds with tight timings...as with the higher speeds and 1GB sticks you would need to use much higher timings than with the 512mb sticks..and thus lose more performance than you gainedso the divider is almost a must with 1GB sticks again...memory speeds are pretty well moot points...it is the cpu speed that is important. only when getting into very weak dividers like DDR140, DDR133, DDR100, do you see a performance penalty that is so great that not even 200 or 400Mhz of cpu increase can make up for it. I've tested this also where I could hit 2660Mhz on a Sempron using a DDR150 divider, and my scores would stay near the rest of my scores at 2660Mhz, but dropping down to lower dividers, and then even being able to get my Sempron to 2940Mhz showed a big 20% or more drop in performance... this is going to be true of anything really....there's a limit on how far down you can drop the divider before the penalty incurred cannot be corrected by the raw cpu speed being gained. the problem with this is there is some talk about the dual core needing more memory bandwidth provided by running the RAM faster...so this is where you start to make a small sacrifice of performance to run these sticks at CAS 3 and DDR500+ lol talk talk is something a lot of people do showing is something we do here its not meant to be a belittling statement its just the truth when someone shows me that dual cores need this hypothetical increase in bandwidth to perform better...then I might believe it...but probably not since I have been testing dual core AMD's since before the public could buy one and there's no difference in the amount of performance gained or lost because of memory bandwidth in them any more than in a normal single core AMD...after all, the X2 AMD is simply two cores on a single die and BOTH cores share the same system memory...and with integrated memory controllers...this again renders the memory bandwidth and timings issue a moot point. the next point of interest in memory speed-vs-timings and their relationship is with the long rumored coming sockets and CPUs from AMD...as this is where DDR2 can shine...not only are 1GB DDR2 sticks less expensive, they also have better speeds, and with the same or better timings as a 1GB DDR1 stick would have if it could run at those speeds...the DDR2 I just bought to go in an Intel system was rated at DDR2-533(266MHz) @ 3-3-3-8 timings...and was able to hold those timings all the way to 280MHz..now this wasn't a super high priced RAM I was using..as it was only ~$90 per gig....I don't know of any DDR1 that I could buy at that price that could reach those speeds...and 1GB sticks of DDR1 would be straining to run at 266MHz at any price...and still need to use CAS 3, and most likely worse timings to be running in that 250MHz to 280MHz range(if they can)but thats just where DDR2 starts, as AMD wont even be using DDR2 that slow...they are starting with DDR2-667(333MHz) at its slowest, even the highest performing 512MB sticks are are only thinking about reaching up to those speeds, no matter what the timings well, we'll find all of this out when we actually have some real items to look at and test...until then, its all speculation. but higher memory bandwidth at same cpu mhz is beneficial (we've never denied that)...but only in some instances. Some applications or programs will show this to be true, some will show this not to be true....it will end up pretty much like the tests I've done in the first post...the application you are testing has a certain dependency...if it is on cpu speed, then higher cpu speed is going to make the difference regardless of memory timings/bandwidth...if memory bandwidth is what the application depends most on, then that will be the difference maker... I feel that this is where AMDs move to the new sockets will pay off...as the speed/timings will be better with DDR2 at the higher speeds...and any performance lost from using 1GB sticks will be gone ah but there is no performance loss from using 1GB modules instead of 512MB modules...unless you are worrying only about benchmarking (and are worried about a 1% to 3% loss in performance from reduced timings). but the benefits from moving up to 1GB modules far outweigh the slight (1% to 3%) penalty from reduced timings and taxation of the cpu's memory controller take, for example, Joe Blow playing Battlefield2 Joe Blow went for 2-2-5-2 BH-5 modules in 1GB kit...and he has a nice gpu so he wants to play at 1024x768 or 1280x1024 with all the AA and AF enabled to max as well as all the eye candy maxed out... so now Joe Blow has a gig of tight memory...but his game stutters like crazy because it is constantly caching off the hdd since BF2 with all the eye candy turned on easily runs out of that 1GB system memory...so when loading levels and playing, the 1GB of memory is forcing the rig to constantly page to and from the pagefile... so now Joe Blow says "hrmm...these guys recommend two 1gb modules for BF2, Quake4, FEAR, etc" but he looks at the timings...and they are now 3-3-8-3 or even 3-4-9-4 timings...not nearly as good as his BH-5 at 2-2-5-2 but guess what? integrated memory controller...you guessed it...so now maybe at same speeds his rig might perform 1% to 3% less... but you guessed it again no more caching off the hdd/pagefile...because his system now has enough memory to run the games fully in memory...no paging, so no hitching, hiccupping, snorting, spitting, no nothing...smooth as glass. and ask Joe Blow..."hey man, aren't those 3-4-8-4 timings crappy compared to your BH-5???" and Joe Blow's reply is going to be "hey man...it might bench 2% lower but all of my games are actually performing better and running much more smoothly" (you can insert me for Joe Blow because I, like you, and many others, believed from the Intel and AthlonXP days that memory timings and bandwidth were the holy grail...only after realizing that the separate memory controllers that Intel and AthlonXP cpu's required were actually bottlenecking and that an integrated, on-die memory controller on the cpu was the busting-of-the-myth did I start to spread the word that we must get away from this hard-held belief that memory timings and bandwidth mean everything...because they don't...and most of us just didn't understand AMD's architectural trumping of previous designs as well as Intel's designs) but if dual/multi cores are going to make use of large amounts of RAM, and need to access a lot of it quickly we will see that the faster speeds(no divider), better CAS latency and memory timings do show performance advantages when the software demands bandwidth at low latency with 1T and 2GB(having it all with no sacrifices) surely if we have games that need to hold in RAM 2GB(2X 1GB or 4X 512MB) of data and process large amounts all at once with multi threading(bandwidth) with fast CPUs that will need this data quickly(speed/low latency) we will see more of a benefit from having all of these things working together not particularly. most people don't realize there's a difference in true SMP and these 'dual core' cpu's. just the same as they don't realize the huge difference in an integrated memory controller versus a separate memory controller hub you just cant really over-saturate the memory-to-cpu bus on a cpu with an integrated memory controller... and again...it will always depend on the software YOU are using to determine the difference. so you use DivX6.2 and with tight timings and no divider at same cpu speed it is 11 seconds faster at encoding...or even 32 seconds faster... but yet the difference in a video game, or photo editing, or some other application that is designed and compiled a little different...shows zero increase, or such a small minute increase that it is laughable. then you begin your overclock...now you find with a divider and loose timings you get 400Mhz more cpu speed than your highest overclock without at divider and tightest memory timings. now you see a difference...DivX is now slower because the cpu Mhz is lower...games are lower performing because cpu mhz is lower...but yet a memory bandwidth benchmark shows a difference because the tighter timings without a divider is beneficial to that particular benchmark and how it benches. and then keep in mind games... games are not created equal Unreal engine games demand high cpu time...every Unreal engine game is limited by cpu over gpu but other games, most particularly Quake3 and those based on it's engine (as well as a few other newer games I am sure), are what we term gpu-limited...meaning that while ramping up the cpu speed will of course increase performance, it does not have nearly the effect that ramping up the gpu/gpu-memory speeds do. Q3 engine is a classic example of a gpu-limited game. Games and applications are like this...each is a bit different (unless based on a specific code engine like Q3 games are, and Unreal engine games are etc). there is no one right or one best setup. My tight timings on a divider rig might crush yours in a certain game...but your loose timings with mega memory bandwidth (again assuming we are both running same cpu speed) might crush mine in a different game... but what about all the talk about multi threaded games and software?...shouldn't they theoretically place twice(or multi times) the demand on not only our CPUs(dual/multi core)....but also demand more load/speed/amount from the memory(bandwidth/latency/speed/quantity) take a look at two current games on the market right now that support multi-threading... CoD2 performs rather awful on multi-threaded (SMP type) setups when SMP is enabled yet Quake4 on the other hand...performs in a much much much better manner latency on the memory has nothing to do with this an integrated memory controller again moots timings and dividers because there is no separate memory controller hub to have to navigate through on an A64 platform...the cpu feeds data directly in and out of system memory without having to detour through a controller hub the difference is how the programs are coded...Q4 is a much more mature code base for SMP, while CoD2's engine is not...while both support SMP, both are not equal! This is quite true of all SMP enabled software...some perform only slightly better when in an SMP environment...and some are built solely for SMP environments and will always perform better (3d Studio Max for example is built on code that pretty much demands multiple cpu's either in the same workstation or over a network). Since SMP enabled software is very very new to the consumer market (only servers really utilized this for databases and high-end graphics creation), there is going to be a large disparity in performance across the different software titles. But SMP and dual core etc are here to stay...as without a major revolution in fab process, we've pretty much hit the Mhz barrier for now...so it is very safe (and efficient) to assume that instead of a single 3Ghz effort, 2x2Ghz working together will be much more efficient. In theory of course, this is true....in reality... we know that this is not always true...it is how the code is implemented, and how well the platform it is implemented for can handle the tasks presented to it. So while crybabies like Gabe Newell from Valve go around crying about SMP and how hard it is going to be blah blah blah, others like Carmack from Id are already working hard to make sure their next creations take advantage of SMP since by the time the next generation game engines are available, SMP/dual-core systems will probably be very much in circulation...yet they will still have to cripple their engines to run on a single core...meaning less physics maybe, less rendering...less AI...who can say but crippling code to run on a single core instead of multiple cores is exactly that...crippling...especially if it is designed for multiple core use (which is why Photoshop and 3d Studio Max and MySQL etc all will perform decent on a single core setup, but really start to shine on multiple cpu systems). *ps I got nothing against Gabe Newell...I love HL2 as much as the next guy...but he's one of those guys who can't let go of 'today'...and while I don't care for Carmack's engines/games so much, I like him a lot more because he innovates and realizes that consoles as well as PC's are moving to multiple cores and thats just the way it is so stfu and start making the best you can* see aren't you glad I didn't have to eat my tinfoil hat in that last post..you would have missed all of that heh no one eats tinfoil hats as much as I do (I'm wearing mine right now so your mental prowess cannot overcome my own thought processes!!!!) this thread isn't to shame anyone, nor is it to talk down to anyone (I hope you nor anyone else feels like this is what is being done). this thread is here for exactly this type of discussion...and we always want to discuss why things are the way they are....as old habits and myths are extremely hard to break (there are still people out there who thing being homosexual is just a mental illness, as well as those who think video games cause humans to become violent and shoot up shopping malls, as well as those who still think tight timings and massive memory bandwidth are the be-all/end-all of cpu overclocking) :tooth: Share this post Link to post Share on other sites More sharing options...
Angry_Games Posted January 26, 2006 Posted January 26, 2006 ...its just splitting hairs really...but that is what benchmarking is about this is exactly the point you are missing unless you are a professional benchmark contestant....benchmarks are useless other than giving you a general idea of system performance...but benchmarks don't give you a 100% accurate picture of how your system will perform in a given application...but we don't buy rigs to join the NBL (National Benchmark League) on a $5,000,000 contract with endorsements from Reebok and ATI we buy rigs to enjoy software that is actually useful...and benchmarks are absolutely useless for using... as for your theory of spending an extra $20 to get 3% increase or such...it just doesn't wash. Because you cannot fit in any dollar limit to show any hard performance gain (or decrease). It just doesn't work that way. Again, the point of this, is that people want to know that they didn't buy the $300 RAM, yet with $210 RAM they can STILL get the same performance (and please, no boo hoo'ing about 1% or 3% performance in a benchmark or even real world application...99% of us would give up 1% to 3% to spend $90 less on one part so they could put it to another part that would actually show an increase like moving from a 6800GT to 7800GT would which puts you up at a very good performance increase...while RAM on an A64 just isn't really going to do more than 1% to 5% maximum) it is clear that you are saying its not worth gaining 3% performance to be running better RAM settings to lose out on 25% more performance from a higher CPU overclock or being able to run more than 1GB of RAM so that BF2 doesn't need to use the page file....that information is made clear from your testing on page 1...and I'm not disputing that good this is the entire point of this thread and the 1T vs 2T thread I'm just trying to clarify the line between where that price|performance is...and what the effects we can see from bad RAM settings -vs- good settings 1% to possibly 5%, but the average being about 2% or 3% and price doesn't always have much to do with anything...what if I buy some 3200 TCCD that hits DDR604 for $137.50 from newegg...and you buy some 4400 TCCD that can only do about 280-ish and you paid $262 from newegg? (and before you scoff at this...it is well known in overclocking circles that this has happened PLENTY of times as a good many of us have fallen for this trap ourselves lol) no, this is not ALWAYS the case...but lets say you can do DDR600 1:1 and you paid $199 for your RAM and lets say I can do 300x whatever but I have to use a divider to get there, say DDR166, and I paid $179 yet the difference between your scores and my scores are 1% to 5% it might be important to you that you spent the extra $20 and got 1% to 5% more performance but what most of us don't care...we only care that we could drop 300x whatever on our cpu even with a divider and still hit the same cpu Mhz as some guy like you who can do it 1:1 and we only lost 1% to 5% performance and lo and behold...what if we, since we are using a divider, could crank up the cpu Mhz even more and go above and beyond your 300x @ 1:1...and now we are 1% to 5% above your max peformance? (would that still not make you drop to a divider to possibly lose 1% to 5% on some benchmark to possibly gain however many % points in a real application that needed that boost in cpu speed???) and what the effects we can see from bad RAM settings -vs- good settings typically 2% to 3% average, high of maybe 5%, low of less than 1%....on an Athlon64, there really is no such thing as 'bad ram settings' because of the integrated memory controller and the superior architecture of the A64 cpu... sure that is a thin line (so thin that you call it "moot")...but you can't deny that the line is there I don't deny it is there and some drive themselves mad trying to get over this thin line but it is "moot" because majority of us only care that using dividers, 2T, and loose timings to get nice cpu clocks don't hurt in the least. those anal-retentive types that claw their eyes out trying to get 1% or 5% gains from tightest timings with the 1:1 ratios etc...those are not really realistic users (but I do wish them well in the National Benchmark League 2006-2007 rookie draft as there are some good teams that need quality players!) I laugh at people who say this too...though I have had some cheapo ram that was decent...I never trust cheapo value junk (boy's ram) to do a man's overclock there's a big difference in cheapo RAM and "VALUE" RAM and anyone that has visited these forums for more than a day has somewhere, somehow, come across me and possibly other Street cronies preaching the evils of "VALUE" anything on a motherboard/setup that was designed from teh ground up to be nothing but a pure overclocking monster. I've got some cheapo RAM as I said that is ultra-fantastic but it is not VALUE RAM I wouldn't put 84 octane gas in my new $100,000 racecar but i would try out some 107 octane generic racing gas but surely there is nothing wrong with try'n to eliminate as many of the disadvantages as you can...as long as you are not creating greater disadvantages in other areas(like CPU MHz) or spending moneys that would show greater advantages if spent in other areas agreed but you are still missing the point with this seemingly endless argument MOST people (ie majority of users who spend their cash to buy a rig to play the latest games and use the latest applications etc) want performance but MOST people are not going to rely completely on benchmarks and spend every waking second for the next 78 days trying to squeeze out that 1% to 5% so they can maybe get themselves drafted higher in the National Benchmark League rookie draft MOST only care that the dang rig works and gets them 70+ fps in CoD2, FEAR, etc. while you are grease-monkey'n around trying to squeeze out that extra 1% to 5%, they are fragging their buddies and hated enemies on the Western Front or on a no-gravity airless moon or some troll-infested WoW raid etc thats the point there's nothing wrong with wanting the best and trying for the best this thread and the others stuck to the top of this section....these threads are NOT for you these threads are for ordinary joe's who want to know that they can still get top performance within a couple percentage points of you superstars who make big bucks in the NBL without spending as much money and without wasting as much time Share this post Link to post Share on other sites More sharing options...
Recommended Posts
Please sign in to comment
You will be able to leave a comment after signing in
Sign In Now