AMD vs. Intel Architectural Debate

[Verran]

Anyway, I feel this discussion is veering away from the original topic of the thread somewhat.

I you want to talk about processor architecture further and how a majority of testing methods may be flawed in assessing performance, the feel free to start a new thread, I think it would be interesting.

Done! Topic split from review thread for further discussion!

 

Personally, I agree that bio is being rather dismissive towards the "standard" testing that's found on enthusiast boards like these. Like it or not, gaming is what people do. There is no better standard to test the needs of the enthusiast community. I'm not saying they all game exclusively or anything like that, but gamers are BY FAR the largest demographic.

 

The applications reviewers test with are chosen based on the most common needs of their readers. It's a very simple and helpful design. I don't really care if AMD quads are a little faster at some obscure I/O benchmark. That does me no good. If that helps you sleep at night as an Phenom owner then that's great, but I'd rather test my chip with applications I actually use.

[jammin]

Cheers for splitting the discussion off Verran.

 

I think it's an interesting subject to talk about, and I do think that with Intel still basking in the limelight at the moment, that people forget that AMD are making pretty capable chips themselves.

[Crazy_Nate]

AMD has always had the balls to take on the 800 pound gorilla, Intel. Major props to them. They keep Intel from doing whatever they want with prices.

 

I'm not sure if you guys have ever taken a look at: http://www.top500.org/

 

They keep track of the monster supercomputers. Lots of interesting information. Data is available for every month...

 

Here's an interesting chart: http://www.top500.org/charts/list/31/procfam

 

Here's the top 10 for this month: http://www.top500.org/lists/2008/06

 

The top is a hybrid system, Opty dual cores and cell processors

 

The Bluegene's are PPC

 

Ranger is quad core Opteron, as is Jaguar.

 

 

 

However, I'll point you back to the processor family chart. It's very, very hard to argue with the 71.2% + 3.2% that Intel holds.

 

 

 

Oh...and for a sidenote 75.60 % LINUX. Eat that M$.

[Waco]

Do you know why AMD is strong on servers?

Because they are very cheap in bulk for both cpus and motherboards, low power consumption, and because servers don't get changed out nearly as often as desktop machines. A few years ago Athlon64s/Opterons were the top dog.

 

There are very few times when an AMD system will outperform an Intel Core 2 system in reality. Sure, for those few cases, AMD may be a slightly better choice (if that's the only thing you do and you care about the couple percent that they are faster) but for EVERYTHING else Intel is the way to go by a huge margin.

 

Few people will be able to tell if the sit on a AMD X2 4850e, AMD phenom, C2D or C2Q

Au contraire, mon amie. It's pretty damn easy to tell if you start up any kind of game that's fairly well parallelized.

 

 

Crazy_Nate - remember that those lists only hold the public supercomputers. There are many organizations with highly parallel machines set up that lay that list to waste.

[dr_bowtie]

Do you know why AMD is strong on servers?

 

 

AMD may be strong on servers but at the current time Intel own 75% of the server market share...

 

 

it DOES NOT matter that AMD has strong IO through put or that it's cache is faster...the fact is as a whole it's NOT faster than the current Intel offerings...

 

Sure Intel may have slower cache or less through put on the FSB and non-intergrated memory controller....but what does that tell you...?

 

Tells me they dont need the speed of the things AMD does to get the job done....in fact they way they do it currently to me is more efficient....

 

is does NOT matter what parts of the chip are faster or slower than each other the only thing that matters is the END RESULT....

 

 

Intels clock higher than AMD not because they have to, it's because they can...that means there is more room left in the chip than AMD offerings....

 

Dont get me wrong if you care to look at my sig you'll see 1 Intel rig and some of the many AMD rigs I own...

 

the poorest performing chip in Intels line up is the only one that AMD even come close to in the way of competing to....

 

the E2140/E2160/E2180 from Intel....if you take for instance the E2160 and run it clock for clock with a AM2 dual core you get about the same performance....granted Super Pi is faster on the Intel....

 

If you take the E2160 and clock it to 3.4ghz (which is about the max) and put it up against say a 5000 Black edition clocked at 3.4ghz and run video encoding test they are very similar in terms of times....but those are the ONLY chips that score low for Intel....any other chip would be farther ahead....

 

I think being an AMD person myself this is a losing arguement and far as anything compared in terms of speed or performance....

 

things that look good and sound good on paper dont always go that way in real world as is the case with AMD specs...

[Hienrich Jager]

I think everyone has missed the point of OP's topic. At least what I got from it seems to be different.

 

I think that bioduken brings up an interesting point about the basic mechanics of modern processors. On the whole we (as the collective of computer enthusiasts) have learned that raw processing power does trump all in terms of performance. And on the whole that is correct solely because there are very very few applications that severely choke the bandwidth of modern FSBs. In your average video game or movie render/conversion most of the arguments about one processor over the other are moot, because as someone pointed out a lot of those tasks are farmed out to the video card etc. I don't think that the OP's intent was to convince anyone to use/buy AMD over Intel or other, rather what I understood from his arguments was that there is an inherent weakness in Intel's current line of processors that may eventually slow things down for the regular consumer if applications are written more and more multi-threaded. This was an attempt to discuss the first article that he posted and comment on why he thinks that the Integrated memory Controller is the way to move forward for the big chip makers.

 

Additionally, I was really appalled at how most of the responses read. I really had expected better from the OCC community. Posts challenging OP to find his best AMD chip and you would compete with your Intel XXXX is absolutely RETARDED!! I COMPLETELY AGREE that the Intel will win every time in applications that are single threaded. Just to give an example, I work at Georgia Institute of Technology as an assistant researcher in the Computer Molecular Modeling division. As a daily part of my work I have to submit 3D systems of 10k or more atoms to one of two computers that we utilize for our research. Computer "A" is a rack of 36 Phenom X4 [email protected]; Computer "B" is a rack of 52 Intel Xenon's [email protected] and each node has 2 proc's on the mobo. (for housekeeping details we are using the LAMMPS software to split the jobs into more manageable segments). Now I have seen in my own research that a job that is split up among a single CPU (4 cores) the Intel chip will complete the work unit roughly 15% faster than the AMD quad core. This is most likely due to the Intel's faster processing speed and the fact that the FSB is not capped on bandwidth, however when I split a work unit among 2 CPU's (8 cores){note that this is still remaining withing a single node and all 8 cores are going through a single FSB} the AMD based chip will complete the work unit roughly 20% faster. My theory is exactly that the FSB is being flooded and can't handle the load placed on it. Of an additional side note a previous poster's comment about using less electricity and needing less cooling is completely accurate when 8 or more cores are utilized for the AMD system.

 

NB:Just in case anyone was wondering when I assign more than 8 cores and thus necessitate the work unit going outside the single motherborad, even though we are using Infiniband the performance drops through the floor.

 

Just as a recap to a reply that turned out way longer than I had originally been planning.... I agree with the OP that the IMC will be the way of the future, just imagine a C2Q with the IMC architecture! But currently AMD's chips suck when compared to Intel's in pure performance and I don't see any applications available to the average (or even the enthusiast) user which would cause them to notice any benefits of AMD's chips. Oh, and about the gaming performance on AMD Vs. Intel... I have absolutely no idea about that topic so I will not comment. I DO think that reviewers should give some more raw data about the chips that they are testing though.

 

Hienrich Jager

 

EDIT: Fixed a typo

[Waco]

I agree with the OP that the IMC will be the way of the future, just imagine a C2Q with the IMC architecture!

Oh, you mean like Nehalem?

[Verran]

Oh, you mean like Nehalem?

 

I was thinking the same thing. The point seems moot because anyone who has researched enough to know all these intricacies has got to know that Intel is already working on this for the relatively near future.

 

Unless AMD has a major trick up its sleeve then Nehalem will probably only extend the dominance of the Intel lineup. A quick look at the rigs in my sig (1 Intel, 4 AMD) will show that I'm not particularly happy about this, but that seems to be the way it's headed.

[hardnrg]

The problem with showing raw data for processors is that it doesn't really carry much meaning. If you look at the suite of synthetic and real-world apps and benchmarks used in a CPU review at OCC, you can see there is a range of general-use apps, scientific/engineering apps, and 3D games/apps.

 

Showing the CPU performance in practice in this range of different applications gives a clear view of the performance of a processor. If you have suggestions for additional apps/benchmarks that you would be interested in, then let bosco and/or ccokeman know, and they may consider adding it to the review suite.

 

Simply having a number for I/O performance isn't going to mean anything to almost everyone reading a review.

 

I'm not sure that comparing the performance of enterprise/server class systems applies to this thread, it seems as though the OP is making a point about a single consumer processor in terms of 3D performance for advanced graphics including Ray Traced implementations by Pohl (I've already read his papers on his Quake mods as part of my MSc Computer Games Technology course)...

 

SpecView (Catia, Maya, ProE), CineBench and ScienceMark seem quite relevant to the performance being debated in this thread, and when comparing the Phenom 9600 to Q6600 (here and here), you can see that the Phenom loses out 3 tests out of 9... granted the Phenom is 2.3GHz vs the Q6600 2.4GHz, but that's about as close a clock comparison as you can get... I'll leave the overclocking headroom argument out of this...

[cirro]

 

it doesnt take a genius to tell that the intel warrior can obviously hold the weapon better making him the more efficent killer

 

 

[Hienrich Jager]

@Thewacokid & Verran

Yes. That is exactly what I was thinking of when I typed that. I just didn't want to bring in another Processor name to add fuel to the fire.

 

@hardnrg

Yes most reviews these days do use a wide range of applications to get a rough gauge of how well a processor would perform in most day-to-day use. Possibly the only thing that I can think of that might be added to any OCC reviews is an average time that a SETI, F@H, or WCG work unit finishes. Don't get me wrong I know full well that it would be near impossible to get a standardized work unit that could be used for benchmarks/testing but I like the concept.

 

Additionally, I don't know about everyone else, but I know that I never turn off my WCG processing. This therefore means that when I fire up COD4 or Crysis I am still chugging along and processing data. Perhaps a test where all cores are doing a WU and then do a time-demo on a game for FPS difference? idk.

 

Yes I realize that a simple number of how many calculations a second will not mean much to the average review-reader, but I had simply said that I would like it *chuckles*. I usually end up reading the white-pages on the proc anyways, but it is always nice to get a rough idea about real-world performance as well.

 

Again I was just stating my impressions about OP's comments. And I guess i need to state that I didn't read the thread that this one was split from. I guess I might want to go do that. *looks for other thread to read*.

 

Jager

[dr_bowtie]

I think everyone has missed the point of OP's topic. At least what I got from it seems to be different.

 

I think that bioduken brings up an interesting point about the basic mechanics of modern processors. On the whole we (as the collective of computer enthusiasts) have learned that raw processing power does trump all in terms of performance. And on the whole that is correct solely because there are very very few applications that severely choke the bandwidth of modern FSBs. In your average video game or movie render/conversion most of the arguments about one processor over the other are moot, because as someone pointed out a lot of those tasks are farmed out to the video card etc. I don't think that the OP's intent was to convince anyone to use/buy AMD over Intel or other, rather what I understood from his arguments was that there is an inherent weakness in Intel's current line of processors that may eventually slow things down for the regular consumer if applications are written more and more multi-threaded. This was an attempt to discuss the first article that he posted and comment on why he thinks that the Integrated memory Controller is the way to move forward for the big chip makers.

 

Additionally, I was really appalled at how most of the responses read. I really had expected better from the OCC community. Posts challenging OP to find his best AMD chip and you would compete with your Intel XXXX is absolutely RETARDED!! I COMPLETELY AGREE that the Intel will win every time in applications that are single threaded. Just to give an example, I work at Georgia Institute of Technology as an assistant researcher in the Computer Molecular Modeling division. As a daily part of my work I have to submit 3D systems of 10k or more atoms to one of two computers that we utilize for our research. Computer "A" is a rack of 36 Phenom X3 [email protected]; Computer "B" is a rack of 52 Intel Xenon's [email protected] and each node has 2 proc's on the mobo. (for housekeeping details we are using the LAMMPS software to split the jobs into more manageable segments). Now I have seen in my own research that a job that is split up among a single CPU (4 cores) the Intel chip will complete the work unit roughly 15% faster than the AMD quad core. This is most likely due to the Intel's faster processing speed and the fact that the FSB is not capped on bandwidth, however when I split a work unit among 2 CPU's (8 cores){note that this is still remaining withing a single node and all 8 cores are going through a single FSB} the AMD based chip will complete the work unit roughly 20% faster. My theory is exactly that the FSB is being flooded and can't handle the load placed on it. Of an additional side note a previous poster's comment about using less electricity and needing less cooling is completely accurate when 8 or more cores are utilized for the AMD system.

 

NB:Just in case anyone was wondering when I assign more than 8 cores and thus necessitate the work unit going outside the single motherborad, even though we are using Infiniband the performance drops through the floor.

 

Just as a recap to a reply that turned out way longer than I had originally been planning.... I agree with the OP that the IMC will be the way of the future, just imagine a C2Q with the IMC architecture! But currently AMD's chips suck when compared to Intel's in pure performance and I don't see any applications available to the average (or even the enthusiast) user which would cause them to notice any benefits of AMD's chips. Oh, and about the gaming performance on AMD Vs. Intel... I have absolutely no idea about that topic so I will not comment. I DO think that reviewers should give some more raw data about the chips that they are testing though.

 

Hienrich Jager

 

 

are you refering the performance of AMD quads vs Intel quads in this post....because the Phenon X3 is a tripple core and not a quad....definitely not apples to apples comparison...