AMD Phenom II AM3 review thread

[Rammsteiner]

Is it just me, or does AMD's chips seem to perform better at higher res/graphics? I'm not talking about bottleneck margins here but at least 5 - 6 FPS better than Intel counterparts. Can anyone explain why 'cause I always thought that lower res/graphics would show the maximum potential of CPUs the most as it eliminates graphics card bottlenecking.

Ive been wondering the same.

Although it's fine if they review at 800x600, on the otherhand it starts to become like a 3DMark result. It doesnt tell us anything, no one plays at that resolution. That wouldnt be a problem if at higher resolutions the results would stay the same, but indeed PhII scales better on higher resolutions, making those low-res results instantly useless. They'd better review just for the normal used resolutions for 17, 19, 22, 24 screens.

Why PhII would scale better, Ive no clue. Some comparison done like a year ago between a Ph and a Yorkfield both at 3.4Ghz shown that the Yorkfield scored a higher CPU score but the PhII increased the GPU scores so much the end score was like the same

[ether.real]

Ive been wondering the same.

Although it's fine if they review at 800x600, on the otherhand it starts to become like a 3DMark result. It doesnt tell us anything, no one plays at that resolution. That wouldnt be a problem if at higher resolutions the results would stay the same, but indeed PhII scales better on higher resolutions, making those low-res results instantly useless. They'd better review just for the normal used resolutions for 17, 19, 22, 24 screens.

Why PhII would scale better, Ive no clue. Some comparison done like a year ago between a Ph and a Yorkfield both at 3.4Ghz shown that the Yorkfield scored a higher CPU score but the PhII increased the GPU scores so much the end score was like the same

PhII performs better at high settings? How do you figure?

http://www.cluboverclocker.com/revie...40Black/p4.asp


Just like it should be, high settings = GPU bottleneck, silly.

[Glow9]

Nevermind, I just woke up when I saw Glow's post and somehow I read something else, like completely missing it was about performance/value instead of just value. Now things makes sense

Forgive me now

[ether.real]

http://www.tomshardware.com/reviews/...om,2148-4.html

"Modding And Overclocking–Doable?
Our first thought after hearing that AMD wouldn't be launching a Phenom II X4 940/920 equivalent for the AM3 platform was, "what about all of the enthusiasts who've been eying those high-end models and still want to experiment with DDR3 memory?"


We knew we had a Phenom II X4 940 in the house that didn't scale very well and would likely be replaced soon by another chip that would ideally have more headroom built into it, so we decided to try "creating" a 938-pin AM3 chip out of our 940-pin AM2+ sample. After all, the silicon under its proverbial hood was the same–the only difference was its interface.


Zoom

So, using a mechanical pencil, we bent the two offending pins back and forth until they snapped off. The chip now fit into our AM3 test platform, albeit not flush due to the metal nubs where each pin broke. Unfortunately, the modded processor would not POST at all, forcing us to conclude that the task wouldn't be as easy as popping off pins. It's truly a shame that enthusiasts can't get access to AMD's AM2 pinout, which would describe the exact role of each pin rather than force us to guess. The most recent tech doc publicly-available relates to the old-school Socket 940 interface.


Wondering if we'd just nuked a perfectly good CPU, we moved the Phenom II X4 940 back to its AM2+ board. Lo and behold, it still ran fine, without any immediately apparent issues.
"

How in the does anyone take these guys seriously? What's next? Trying to cram DDR3 modules in DDR2 sockets?

[msimax]

call me crazy but i was wondering the same thing but he saved me the task

[Enoc]

what AMD missed was to send the reviewers a PII 910 or PII 925...I really wanted to see how the L3 6mb quads scale with DDR3...

[ether.real]

Did anyone else notice that not many sites really tried pushing the X4 810 much? I got 3.51GHz, I think two sites pushed harder than that, and everyone else crapped out at 3GHz-3.2GHz....

[G0ldBr1ck]

http://www.tomshardware.com/reviews/...om,2148-4.html






How in the does anyone take these guys seriously? What's next? Trying to cram DDR3 modules in DDR2 sockets?

Good to know. I had wondered the same.

[Mads321]

Did anyone else notice that not many sites really tried pushing the X4 810 much? I got 3.51GHz, I think two sites pushed harder than that, and everyone else crapped out at 3GHz-3.2GHz....

I've got 37xx MHz out of the sample I've got here now - still working on the OC section of that review, which I suspect will get public sometime soon

[Mav451]

I've got 37xx MHz out of the sample I've got here now - still working on the OC section of that review, which I suspect will get public sometime soon

I suspect that the gimped cache is probably reducing people's interest in it. The 720, OTOH, has more cache but 1 less core + unlocked multi. I dunno, *shrug*

[Glow9]

what AMD missed was to send the reviewers a PII 910 or PII 925...I really wanted to see how the L3 6mb quads scale with DDR3...

No kidding why would you want a Quad with 4mb of cache I thought AMD would be done shorting us on the cache. I'm real curious about their highend but leary about the cost cause the mass amount of crap they are puttin in midrange will prolly push the cost of the highend up. Really hope its not much more than what 940 since thats their current highend price

[Rammsteiner]

PhII performs better at high settings? How do you figure?

http://www.cluboverclocker.com/revie...40Black/p4.asp


Just like it should be, high settings = GPU bottleneck, silly.

There's this weird difference between resolutions and actual settings No way any decent GPU, which you should use anyway to make the possible bottlenecks as less as possible, would become a bottleneck at the before mentioned resolutions

Also, what a review that is, only four games where two of them are already known to be highly GPU bound!? What's wrong with UT3 etc

[jmke]

you're talking to the author Ramm what's the point if a CPU can run UT3 at 250fps or 270fps anyways? give me meaningful numbers over those anyday of the week



very nice test, thanks! SupCom is known for being multithreaded but here an Intel dual core is more powerful ; at what setting did you bench this? Did you use the included -perf map test etherreal?

[MsB]

Thats all Nice~~ but I want 945 and I want it now

[G.Foyle]

Did anyone else notice that not many sites really tried pushing the X4 810 much? I got 3.51GHz, I think two sites pushed harder than that, and everyone else crapped out at 3GHz-3.2GHz....

We did push, got 3770 MHz fully stable on crappy cooler

[Frontl1ne]

PhII performs better at high settings? How do you figure?

http://www.cluboverclocker.com/revie...40Black/p4.asp


Just like it should be, high settings = GPU bottleneck, silly.

Look at this score: http://www.guru3d.com/article/amd-ph...-review-am3/23

If you follow the margin between the E8400 and the X4 810, you'll notice it's a 1 frame difference at 1024 res, 1 frame difference at 1280 res and then a 6 frame difference at 1600 res. Also, the i7 drops below the PII's at 1600 res, the X4 920 gains 2 FPS moving to 1600 res, and the X3 720 stays the same moving from 1280 to 1600 res.

http://www.guru3d.com/article/amd-ph...-review-am3/25

The X4 810 has a 3 FPS lead to the i7 in 1600 after trailing in the 2 lower resolutions.

This is where it starts looking suss: http://www.firingsquad.com/hardware/...ance/page7.asp

X3 720 trails the Q6600 by about 6 FPS in 800 res. At 1600 res with same settings, the margin is only 3 FPS. Turn on AA and all graphics to high, the X3 720 leads the Q6600 by 5 FPS. Even the X4 810, which was trailing the Q6600 by 10 frames in 800 res, leads it at 1600 res by 2 frames. Also, in that Far Cry 2 test, top four in 800 res is Intel > Intel > AMD > Intel. Top four in 1600 res is Intel > Intel > AMD > Intel. Top four in 1600 res with eye candy is Intel > AMD > AMD > AMD.

This is interesting, particularly the Q6600 and the two Phenoms: http://www.firingsquad.com/hardware/...ance/page9.asp

Well, that's all I've got. And it's interesting because through forums and people I know, I've been hearing bits and pieces of the same thing, that Phenoms perform better at higher resolutions. Why does the Phenoms score higher than Intel chips in higher resolutions? It's more than just bottlenecks. Not saying it happens in all games though or that the lead is massive in high resolution. But the lead is there

Sorry about the lack of data, not many people seem to test at both low and high resolutions and others don't test gaming performance at all

Edit: Just checked out some 920/940 reviews and found some interesting stuff:

http://www.legitreviews.com/article/860/11/ <- Fallout 3 results
http://www.trustedreviews.com/cpu-me...m-II-X4-940/p3 <- Quake Wars, PII 940 loses 2 FPS shifting to 1920 res whereas Intel loses 8 - 20 FPS.

I've also noticed after looking through a couple of game tests that the i7 seems to take a bit of a performance hit at high resolution gaming. It crops up quite often:

http://www.hwbox.gr/showthread.php?t=3189&garpg=23
http://www.overclockersclub.com/revi...omii940/14.htm

Maybe that's why people feel Phenoms offer a smoother gameplay experience.

[Shintai]

Look at this score: http://www.guru3d.com/article/amd-ph...-review-am3/23

If you follow the margin between the E8400 and the X4 810, you'll notice it's a 1 frame difference at 1024 res, 1 frame difference at 1280 res and then a 6 frame difference at 1600 res. Also, the i7 drops below the PII's at 1600 res, the X4 920 gains 2 FPS moving to 1600 res, and the X3 720 stays the same moving from 1280 to 1600 res.

http://www.guru3d.com/article/amd-ph...-review-am3/25

The X4 810 has a 3 FPS lead to the i7 in 1600 after trailing in the 2 lower resolutions.

This is where it starts looking suss: http://www.firingsquad.com/hardware/...ance/page7.asp

X3 720 trails the Q6600 by about 6 FPS in 800 res. At 1600 res with same settings, the margin is only 3 FPS. Turn on AA and all graphics to high, the X3 720 leads the Q6600 by 5 FPS. Even the X4 810, which was trailing the Q6600 by 10 frames in 800 res, leads it at 1600 res by 2 frames. Also, in that Far Cry 2 test, top four in 800 res is Intel > Intel > AMD > Intel. Top four in 1600 res is Intel > Intel > AMD > Intel. Top four in 1600 res with eye candy is Intel > AMD > AMD > AMD.

This is interesting, particularly the Q6600 and the two Phenoms: http://www.firingsquad.com/hardware/...ance/page9.asp

Well, that's all I've got. And it's interesting because through forums and people I know, I've been hearing bits and pieces of the same thing, that Phenoms perform better at higher resolutions. Why does the Phenoms score higher than Intel chips in higher resolutions? It's more than just bottlenecks. Not saying it happens in all games though or that the lead is massive in high resolution. But the lead is there

Sorry about the lack of data, not many people seem to test at both low and high resolutions and others don't test gaming performance at all

Edit: Just checked out some 920/940 reviews and found some interesting stuff:

http://www.legitreviews.com/article/860/11/ <- Fallout 3 results
http://www.trustedreviews.com/cpu-me...m-II-X4-940/p3 <- Quake Wars, PII 940 loses 2 FPS shifting to 1920 res whereas Intel loses 8 - 20 FPS.

I've also noticed after looking through a couple of game tests that the i7 seems to take a bit of a performance hit at high resolution gaming. It crops up quite often:

http://www.hwbox.gr/showthread.php?t=3189&garpg=23
http://www.overclockersclub.com/revi...omii940/14.htm

Maybe that's why people feel Phenoms offer a smoother gameplay experience.

You basicly point out GPU limitations. It doesnt matter if you put a 500Ghz i7, Core 2 or Ph2 infront of an "old" game. It will still perform around roughly to the lowest max CPU. So its natural that the faster CPU will lose more "FPS" when resolution goes up. Hence why slower CPUs tend to shine the slower the graphic card(s) is. Overclockers.com uses a single 4850 as an example. Firingsquad uses a 295GTX. However no Q9550 or i7. Legit used a 260GTX and so on.

Phenoms smoother? No...more stable perhaps since the variation is lower. But ranging in say 10-50FPS aint better than 10-200FPS.

Also most of not all reviews use fraps. Fraps is however far from the optimal way of measuring. So you could easily have a 5% or more in differation. Plus you have the standard test variations.

As many tests shows the exact opposite of yours. Its just that you handpicked those you wanted to see and base your conclusion on.

[ether.real]

There's this weird difference between resolutions and actual settings No way any decent GPU, which you should use anyway to make the possible bottlenecks as less as possible, would become a bottleneck at the before mentioned resolutions

This makes absolutely no sense. Care to elaborate?

Also, what a review that is, only four games where two of them are already known to be highly GPU bound!? What's wrong with UT3 etc

Why would I bench a game that no current hardware has issues running? Benching UT3 is pointless. The last time I benched UT3, the highest settings/res ran at over 120FPS. Its a waste of time.

[ether.real]

you're talking to the author Ramm what's the point if a CPU can run UT3 at 250fps or 270fps anyways? give me meaningful numbers over those anyday of the week



very nice test, thanks! SupCom is known for being multithreaded but here an Intel dual core is more powerful ; at what setting did you bench this? Did you use the included -perf map test etherreal?

Yes, I used the -perf option, maxed out, 1920x1200.

[ether.real]

You basicly point out GPU limitations. It doesnt matter if you put a 500Ghz i7, Core 2 or Ph2 infront of an "old" game. It will still perform around roughly to the lowest max CPU. So its natural that the faster CPU will lose more "FPS" when resolution goes up. Hence why slower CPUs tend to shine the slower the graphic card(s) is. Overclockers.com uses a single 4850 as an example. Firingsquad uses a 295GTX. However no Q9550 or i7. Legit used a 260GTX and so on.

Phenoms smoother? No...more stable perhaps since the variation is lower. But ranging in say 10-50FPS aint better than 10-200FPS.

Also most of not all reviews use fraps. Fraps is however far from the optimal way of measuring. So you could easily have a 5% or more in differation. Plus you have the standard test variations.
As many tests shows the exact opposite of yours. Its just that you handpicked those you wanted to see and base your conclusion on.

Sounds like margin of error to me. Most of the testing shows 2-5FPS difference, which is hardly substantial. If this rumor persists, I guess the NBOC crew is going to have to mythbust this.

[gosh]

About performing on higher resolutions.
The workload for the CPU in games is not constant, everybody knows this because there is min, max values.
If you are walking on a road in the game and there isn't much action, then the CPU doesn't need to calculate that much for the picture and there is probably less calculation done for physics, AI etc.
This situation favors intel and all the tricks that it has for speedups. Memory is probably less fragmented and threads doesn't need to synchronize that often.
When there is a fight or something else that need much more calculations done for the picture workload changes. It is probably more synchronizations done if the game uses threads, more memory is used and it is more fragmented.
On low resolutions the GPU doesn't limit performance and the calculation might look like this (VERY SIMPLIFIED).
Most of the time spent in games isn't spent in heavy action situations, if 95% is spent on the road and 5% is spent in action situations.
Low resolution (gpu doesn't limit the fps):
Walking on a road 200 FPS for i7 and 35 FPS in a fight.
200 * 0,95 + 35 * 0,05 = 191,75

Phenom may get 130 FPS and 60 FPS when there is a fight.
130 * 0,95 + 60 * 0,05 = 126,5

High resolution (gpu can only draw 60 frames per second):
i7: 60 * 0,95 + 35 * 0,05 = 58,75
Phenom: 60 * 0,95 + 60 * 0,05 = 60

These numbers are just to show the outcome when the GPU starts to limit cpu.

The L3 cache on phenom is 48-way set associative, i7 has 16-way set associative L3 cache. Phenom has 64 KB L1 cache for data and code, i7 has 32 KB cache for data and code, L2 cache is 512 for phenom and 256 for i7.
I think that phenom handle complex situations better, it is built based on decreasing bottlenecks. Core 2 and i7 may have a different focus, they may have the focus on maximize performance and when it works it works good, when it doesn't work the slowdown is larger compared to phenom.

EDIT: There is some who thinks this is a chipset problem, but if it was the chipset then Phenom would always be better on high resolutions when the gpu limits the cpu. Maybe the difference sometimes is so small that is hard to spot but it would be there.

[Zucker2k]

About performing on higher resolutions.
The workload for the CPU in games is not constant, everybody knows this because there is min, max values.
If you are walking on a road in the game and there isn't much action, then the CPU doesn't need to calculate that much for the picture and there is probably less calculation done for physics, AI etc.
This situation favors intel and all the tricks that it has for speedups. Memory is probably less fragmented and threads doesn't need to synchronize that often.
When there is a fight or something else that need much more calculations done for the picture workload changes. It is probably more synchronizations done if the game uses threads, more memory is used and it is more fragmented.
On low resolutions the GPU doesn't limit performance and the calculation might look like this (VERY SIMPLIFIED).
Most of the time spent in games isn't spent in heavy action situations, if 95% is spent on the road and 5% is spent in action situations.
Low resolution (gpu doesn't limit the fps):
Walking on a road 200 FPS for i7 and 35 FPS in a fight.
200 * 0,95 + 35 * 0,05 = 191,75

Phenom may get 130 FPS and 60 FPS when there is a fight.
130 * 0,95 + 60 * 0,05 = 126,5

High resolution (gpu can only draw 60 frames per second):
i7: 60 * 0,95 + 35 * 0,05 = 58,75
Phenom: 60 * 0,95 + 60 * 0,05 = 60

These numbers are just to show the outcome when the GPU starts to limit cpu.

The L3 cache on phenom is 48-way set associative, i7 has 16-way set associative L3 cache. Phenom has 64 KB L1 cache for data and code, i7 has 32 KB cache for data and code, L2 cache is 512 for phenom and 256 for i7.
I think that phenom handle complex situations better, it is built based on decreasing bottlenecks. Core 2 and i7 may have a different focus, they may have the focus on maximize performance and when it works it works good, when it doesn't work the slowdown is larger compared to phenom.

EDIT: There is some who thinks this is a chipset problem, but if it was the chipset then Phenom would always be better on high resolutions when the gpu limits the cpu. Maybe the difference sometimes is so small that is hard to spot but it would be there.

Gosh it's simple; a gpu must work in tandem with a cpu to render images on your screen; the stronger of two cpus in this instance is the one that processes information and spits it back fastest to the gpu. This assumes that the gpu is processing a very light load; meaning the cpu has to work more to keep up. As soon as you shift the load to the gpu, then the weaknesses of the weakest cpu becomes diminished because the gpu is not as demanding anymore since it is dealing with a heavy load itself. This is what happens when most people run high res on mid-range gpus; it allows weaker cpus to appear shine because they operate at lower cycles (load). If you're attributing these anomalies to the architecture of the PHII, then how come the PHII still loses in many other games (both low res and high res)? Even a dualcore becomes a contender in high res situations as the link above shows.

Anyway, I went to trouble of playing Grid (my favorite) at a low res of 640x480 to show you a visual image of the loads on cpu, and gpu cores.

Edit: Q9650 @ 4.3Ghz 9x478.

[gosh]

Gosh it's simple; a gpu must work in tandem with a cpu to render images on your screen; the stronger of two cpus in this instance is the one that processes information and spits it back fastest to the gpu. This assumes that the gpu is processing a very light load; meaning the cpu has to work more to keep up.

If you mean that the gpu lets the processor "spin" faster thats correct.

As soon as you shift the load to the gpu, then the weaknesses of the weakest cpu becomes diminished because the gpu is not as demanding anymore since it is dealing with a heavy load itself.

But if the load for one frame is on the gpu and then there is something that happens and the CPU need to fetch memory, the load is for the CPU rendering that frame. A larger cache miss could mean that that speed for that frame depends on the cpu, next frame and data is in the cache and speed is decided by the GPU. It can shift a lot.
This is also one situation that is hard to spot on graphs that times frames because the gpu runs asynchronously to the cpu. It is always at least one frame behind. Sometimes almost two.

If you're attributing these anomalies to the architecture of the PHII, then how come the PHII still loses in many other games (both low res and high res)? Even a dualcore becomes a contender in high res situations as the link above shows.

It depends on how the code is done, if the code doesn't hit bottlenecks it will scale well on intel. I think you will see much more of this phenomena when games start to use DX11, that phenom runs better on high resolutions compared to intel for processors today.
EDIT: i7 may like DX11 games too

[rk7p5]

Gosh's theory sounds quite plausible. What ever the cause, phenom 2 is a pretty amazing little cpu. The platform cost makes it even more enticing.

[largon]

Soo, where's the beef? I mean bandwidth.
Looks like DDR3-1600 yields only around 15% more b/w for AM3 CPUs compared to DDR2-1066 on AM2+.