17 replies to this topic

[Gr4vitas]

So this is the best I've accomplished: http://3dmark.com/3dm11/2782526

First of all, since then I've brought my ram up to 1941 (was at 1866 for the score above) and brought my graphics to 1000 / 1500. Since I did that I consistently score ~200 lower than the score above. I did have to loosen the timings just a little (about 2 across the board) to get 1941. (my ram is 1600 stock btw ) would that small change in timings really hurt me more than the gain in mghz im getting?

Second question is, I was looking at the scores from people with the same hardware, there are a bunch of guys at 7k+ scores. So I was looking through them and most of them have there hardware running at lower stats than me (or very similar nobody really above) and there pulling off better cores by 700 points, hell I even have my 2500k at 5.2ghz and just about all the people that are 7k+ are at 5ghz.

Also it seems that 3dmark doesnt grab your ram speeds correctly, it shows everyone at 667 per stick. So what im thinking is all these people with these high scores are running 2000mghz+ memory while im running 1941 (with probably slower timings too). Would memory speeds really make up for that much?

The only other difference is I only have 4 gigs of ram (upgrading to 8 soon), but I really don't see that making any difference.

[MercuryDoun]

Since i already did a bunch of testing on memory speeds and timings for 3DMark11. Maybe this will help you too

http://forums.overcl...dpost&p=1870001

In short. Timings > MHz in memory, the best of both worlds increase of both MHz and lower timings is best, but yeah. I have since achieved FAR greater memory overclocks into the 2200MHz range with same i7 and RAM as in this link, but back then i had a 5770 and i have a GTX570 now. But either way that doesn't change the drastic amount of points in the physics department that memory speeds and timings play in 3dMark11.

[MercuryDoun]

Also You may have pushed the memory on your video card too high and its throwing errors now, thus reducing performance.

Make sure you look at, and compare, the split scores that make up the P-Score, these are the Graphics Score, the CPU Physics score, and the Combined Score. Using this split scores, you will be able to tell which of your recent overclocks had a negative impact on your score!

[Speedway]

To add to Merc's mem findings - There ARE sweet spots with all 3dMark benches I have found. Sometimes I will get a better score with lower GPU clocks, especially if you are really pushing the card(s). Recently I pushed my 6970's to 1050 core and 1550 mem, but found my 3dMark 11 scores were lower than when I ran 1030/1500. I even ran the test 3 times thinking it was just a fluke. But over the last 2 years and benching the crap out of FutureMark products, I have def noticed that there is a "sweet spot" on your GPU clocks, and going over it can reduce the scores. Also, is you CPU running non-turbo 5.2GHz? If not, then it WILL throttle up and down during the test and this can also affect the scores. I run my OC with turbo for 24/7 use, but when I'm benching I turn turbo OFF and run the CPU at a constant OC, so that throttling does not factor into the score

edit: Sorry, my best scores were with OCs of 1050/1550 on the GPUs. When I pushed the cards even higher than that, although the test ran through and didn't crash, my scores dropped significantly.

[Gr4vitas]

To add to Merc's mem findings - There ARE sweet spots with all 3dMark benches I have found. Sometimes I will get a better score with lower GPU clocks, especially if you are really pushing the card(s). Recently I pushed my 6970's to 1050 core and 1550 mem, but found my 3dMark 11 scores were lower than when I ran 1030/1500. I even ran the test 3 times thinking it was just a fluke. But over the last 2 years and benching the crap out of FutureMark products, I have def noticed that there is a "sweet spot" on your GPU clocks, and going over it can reduce the scores. Also, is you CPU running non-turbo 5.2GHz? If not, then it WILL throttle up and down during the test and this can also affect the scores. I run my OC with turbo for 24/7 use, but when I'm benching I turn turbo OFF and run the CPU at a constant OC, so that throttling does not factor into the score

edit: Sorry, my best scores were with OCs of 1050/1550 on the GPUs. When I pushed the cards even higher than that, although the test ran through and didn't crash, my scores dropped significantly.

Interesting, I am indeed running with turbo on for general use, didn't think about it throttling down while loading good point there, I'll have to turn it off for benchmarking.

[MercuryDoun]

Any luck getting it all figured out?

[Gr4vitas]

Any luck getting it all figured out?

Na I couldn't figure out how to turn off the throttling crap in the bios. Still getting scores 700 points under other people online with the same hardware, while my hardware is clocked significantly higher than theres.

[MercuryDoun]

Na I couldn't figure out how to turn off the throttling crap in the bios. Still getting scores 700 points under other people online with the same hardware, while my hardware is clocked significantly higher than theres.

Have you compared the subscores(CPU/Graphics/Combined) of your highest score to your other lower scores? Im pretty sure you'll pin point what went wrong when you do that. Also doesn't hurt to compared your Subscores to that of the other similar spec'd hardware people either.

If your CPU score is lower then you know its probably the memory overclock that hurt and if your graphics score is lower then you'll know if it is your video card overclock.

[Gr4vitas]

Have you compared the subscores(CPU/Graphics/Combined) of your highest score to your other lower scores? Im pretty sure you'll pin point what went wrong when you do that. Also doesn't hurt to compared your Subscores to that of the other similar spec'd hardware people either.

If your CPU score is lower then you know its probably the memory overclock that hurt and if your graphics score is lower then you'll know if it is your video card overclock.

Yea there website seems to be having issues atm, but I do remember comparing initially and if I remember correctly it was there GPU scores that were higher, my CPU score was actually higher than the highest scoring person. I'll have to take a look again though when there website clears up.

[Gr4vitas]

So I just looked, it looks like there physics score / GPU scores are WAY higher, I have no idea how, I'm going to try bringing down the clocks on my memory / core and see if I get better results. I did the same thing for the furmark benchmark underclocked from 1000 core and 1500 memory to 740 core and got much better scores, though I noticed the memory clocks didnt matter in furmark (tried with like 600mghz and got the same results)

[MercuryDoun]

So I just looked, it looks like there physics score / GPU scores are WAY higher, I have no idea how, I'm going to try bringing down the clocks on my memory / core and see if I get better results. I did the same thing for the furmark benchmark underclocked from 1000 core and 1500 memory to 740 core and got much better scores, though I noticed the memory clocks didnt matter in furmark (tried with like 600mghz and got the same results)

You're on the right track, but i'd basically start from scratch at this point. Bring it back down to stock, and start over. Benching every step of the way, until you hit a spot where you scores are either not increasing or maybe even decreasing. Then you'll find that sweet spot for your CPU and GPU. The GPU is hard to say, so much as more likely the memory just got too high and started throwing errors so while not bad enough to crash the Bench, but bad enough to have a negative impact on the score. If its not the GPU Memory, then i would be looking else where, such as your PSU. How many watts is it? But more importantly, how many amps? If you have a multi rail PSU, do you have one of each of the PCI-e power connectors that plug into your 6970 come from a different rail each? This would give your GPU access to significantly more amps over all, since it could draw amps from 2(Or even 3 rails depending on PSU spec and if the ATX power rail is also separate from the PCI-e rails)

Basically there is a possibility if you have say a 4 12v rail PSU, and both of the PCI-e connectors to your 6970 AND the 8-pin CPU power are on the same rail(by pure unlucky chance) those 3 sources could easily drain all the amps of one rail and leave all your components starving for power while 2 or your rails remain relatively idle. Of course all of this means nothing if you have a single rail PSU, just giving you possible directions to look into. Maybe your overclocks are perfect, just under powered.

Have to try and find a decent combo of MHz on and timing on your memory. I ended up being able to take my DDR3-1600 9-9-9-27 1T memory anywhere from 6-8-6-20 1T at 1600-1750MHz to 10-11-10-27 2T at 2250-2300MHz. I ended up landing on 2100~ish area with 7-10-7-27 2T at 4.4GHz on my i7 and ran some of the best benchs i ever did!

Now i run it 24/7 at 1924MHz 7-10-7-24 2T and get fantastic results in every day use and 4.0GHz on my i7. Basically i spent 2 months benching my memory one dozens of different tests before i found the right spots. I was getting lower memory scores at 2300MHz(And high timings) than i was at 2000 with slightly tigher timings. To be it seems like the gains of MHz are exponentially decreasing. With the first few hundred making a huge difference(So long as timings aren't horribly mangled) but the next few hundred show little improvement. While the gains from timings are pretty linear in that you get fairly equal gains as you continuously find tighter stable timings.

[Gr4vitas]

You're on the right track, but i'd basically start from scratch at this point. Bring it back down to stock, and start over. Benching every step of the way, until you hit a spot where you scores are either not increasing or maybe even decreasing. Then you'll find that sweet spot for your CPU and GPU. The GPU is hard to say, so much as more likely the memory just got too high and started throwing errors so while not bad enough to crash the Bench, but bad enough to have a negative impact on the score. If its not the GPU Memory, then i would be looking else where, such as your PSU. How many watts is it? But more importantly, how many amps? If you have a multi rail PSU, do you have one of each of the PCI-e power connectors that plug into your 6970 come from a different rail each? This would give your GPU access to significantly more amps over all, since it could draw amps from 2(Or even 3 rails depending on PSU spec and if the ATX power rail is also separate from the PCI-e rails)

Basically there is a possibility if you have say a 4 12v rail PSU, and both of the PCI-e connectors to your 6970 AND the 8-pin CPU power are on the same rail(by pure unlucky chance) those 3 sources could easily drain all the amps of one rail and leave all your components starving for power while 2 or your rails remain relatively idle. Of course all of this means nothing if you have a single rail PSU, just giving you possible directions to look into. Maybe your overclocks are perfect, just under powered.

Have to try and find a decent combo of MHz on and timing on your memory. I ended up being able to take my DDR3-1600 9-9-9-27 1T memory anywhere from 6-8-6-20 1T at 1600-1750MHz to 10-11-10-27 2T at 2250-2300MHz. I ended up landing on 2100~ish area with 7-10-7-27 2T at 4.4GHz on my i7 and ran some of the best benchs i ever did!

Now i run it 24/7 at 1924MHz 7-10-7-24 2T and get fantastic results in every day use and 4.0GHz on my i7. Basically i spent 2 months benching my memory one dozens of different tests before i found the right spots. I was getting lower memory scores at 2300MHz(And high timings) than i was at 2000 with slightly tigher timings. To be it seems like the gains of MHz are exponentially decreasing. With the first few hundred making a huge difference(So long as timings aren't horribly mangled) but the next few hundred show little improvement. While the gains from timings are pretty linear in that you get fairly equal gains as you continuously find tighter stable timings.

Thanks, I actually can't seem to find any information on my power supply as apparently its not manufactured anymore. It's an OCZ 1000w, and it isn't modular. So there are only two cables for PCI express power and I would assume they are on different rails. Though since you brought that up, I wonder if maybe just by chance I happened to hook up everything to one rail (ie: I just happen to use a molex and sata power cable from the same rail as the PCI express power etc) because my graphics card makes this really annoying whine sound when under heavy load. It sounds like a noisy cap or something (not really concerned about it but its really annoying) its so loud that I can hear it over all 6 fans in my case at 100% though they are pretty quiet, my graphics card is easily the loudest thing in my case between its fan and the wining / gritty electrical current noise.

I really wish I could just sell it to someone for near new value and spend $30 to buy a new one that doesn't make so much damn noise.