phenom in company of heroes

[Fox5]

Originally posted by: taltamir
Fox5, interesting theory, by that theory using a 1GB vram video card would cause the performance ratios to flip...

They would be as normal, Intel cpus would lead.
Well, unless the game was programmed to push an awful lot of data over the pci express bus anyway. The PCI Express bus is bi directional so it's possible that a developer might want to write to vram, though with how advanced shaders are now I doubt it.

 

[SlowSpyder]

Originally posted by: Fox5

Originally posted by: SlowSpyder
It is rather odd that in that game the Phenom, which is considered to be so much slower ends up being very much faster. Some of those benchmarks show odd results, look at the GPU test in Crysis. As the resolution goes up the Phenom gains fps in some situations, and ends up very slightly faster then the Intel offerings. WIC shows the Phenom to be a lot better then most people here would tell you it is too. If you take the results shown on that website as truth, then if you game on a 22" monitor or higher, all I take away from it is that the Phenom is as good as any other option. Not bad for a Phenom 9600, which is middle of the Phenom range.

My guess:
The video card's memory bandwidth/memory size is being stressed to its limits, forcing more data over the PCI-Express bus. At this point, AMD's superior memory controller and bandwidth comes into play.

Heh, I kind of want to see a low end shoot out now. Of course, even low end cards have 128MB to 256MB worth of ram, but it'd be interesting to see if cards with lower amounts of ram could be stressed and give an edge to phenom.

So if you game on a 22" monitor or higher with a 512MB video card or lower, the Phenom might be better then most people thought. It would be nice to see this looked into further, the 8800GT is available in 256, 512, and 1GB versions... it'd be perfect for testing.

 

[uribag]

I saw something like this at Phoronix.
Look the results at 1680x1050.
http://www.phoronix.com/scan.p...article&item=961&num=4

 

[heyheybooboo]

Originally posted by: SlowSpyder
It is rather odd that in that game the Phenom, which is considered to be so much slower ends up being very much faster. Some of those benchmarks show odd results, look at the GPU test in Crysis. As the resolution goes up the Phenom gains fps in some situations, and ends up very slightly faster then the Intel offerings. WIC shows the Phenom to be a lot better then most people here would tell you it is too. If you take the results shown on that website as truth, then if you game on a 22" monitor or higher, all I take away from it is that the Phenom is as good as any other option. Not bad for a Phenom 9600, which is middle of the Phenom range.

I'd be careful drawing any conclusions. I agree it's very interesting but clearly that was not the point of the article.

It would have been more interesting if he had even tried to OC the Phenom 9600BE on that Asus mobo. That's a crime

 

[SlowSpyder]

Originally posted by: heyheybooboo

Originally posted by: SlowSpyder
It is rather odd that in that game the Phenom, which is considered to be so much slower ends up being very much faster. Some of those benchmarks show odd results, look at the GPU test in Crysis. As the resolution goes up the Phenom gains fps in some situations, and ends up very slightly faster then the Intel offerings. WIC shows the Phenom to be a lot better then most people here would tell you it is too. If you take the results shown on that website as truth, then if you game on a 22" monitor or higher, all I take away from it is that the Phenom is as good as any other option. Not bad for a Phenom 9600, which is middle of the Phenom range.

I'd be careful drawing any conclusions. I agree it's very interesting but clearly that was not the point of the article.

It would have been more interesting if he had even tried to OC the Phenom 9600BE on that Asus mobo. That's a crime

Like I said, that's only if you take everything on that review as gospel truth... I'd certainly say that there needs to be more investigating done. Somehow I don't see C2D owners rushing to replace their current rig with Phenom 9600's.

 

[heyheybooboo]

Originally posted by: uribag
I saw something like this at Phoronix.
Look the results at 1680x1050.
http://www.phoronix.com/scan.p...article&item=961&num=4

That is interesting. Maybe we are seeing a slight trend. I wonder if the drivers had anything to do with that? Maybe they were optimized for that resolution in that version of Quake?

It looks like the fps scaled pretty well with an OC. Not quite 1:1 but pretty close ...

Originally posted by: SlowSpyder ~~snip~~

Like I said, that's only if you take everything on that review as gospel truth... I'd certainly say that there needs to be more investigating done. Somehow I don't see C2D owners rushing to replace their current rig with Phenom 9600's.

Don't stand in front of the door - - you may get trampled - lol

 

[Fox5]

Originally posted by: SlowSpyder

Originally posted by: Fox5

Originally posted by: SlowSpyder
It is rather odd that in that game the Phenom, which is considered to be so much slower ends up being very much faster. Some of those benchmarks show odd results, look at the GPU test in Crysis. As the resolution goes up the Phenom gains fps in some situations, and ends up very slightly faster then the Intel offerings. WIC shows the Phenom to be a lot better then most people here would tell you it is too. If you take the results shown on that website as truth, then if you game on a 22" monitor or higher, all I take away from it is that the Phenom is as good as any other option. Not bad for a Phenom 9600, which is middle of the Phenom range.

My guess:
The video card's memory bandwidth/memory size is being stressed to its limits, forcing more data over the PCI-Express bus. At this point, AMD's superior memory controller and bandwidth comes into play.

Heh, I kind of want to see a low end shoot out now. Of course, even low end cards have 128MB to 256MB worth of ram, but it'd be interesting to see if cards with lower amounts of ram could be stressed and give an edge to phenom.

So if you game on a 22" monitor or higher with a 512MB video card or lower, the Phenom might be better then most people thought. It would be nice to see this looked into further, the 8800GT is available in 256, 512, and 1GB versions... it'd be perfect for testing.

I don't agree that a 512MB card would be enough to show the limits.
Maybe a 256MB card.
And even then, raising resolution alone may not be enough to cause a lot of swapping.

It looks like the fps scaled pretty well with an OC. Not quite 1:1 but pretty close ...

The hyper transport bus and memory gets faster with the overclock, maybe?

 

[taltamir]

I found it, they specify that they use an ASUS 8800GT TOP, which is a 512MB model with a 3rd party heat sink and is OCed.

This explains why only two games did it, my bet is that those two games indeed required more then 512MB of vram and that is why the phenom won there.

Now if only we could test it... I simply don't have the hardware variety needed to perform such a test. Maybe an anadtech mod...

This will explain how a such a variety of intel processors, from extreme editions OCed to 4ghz to stock dual and quad cores at 2.4 could get the exact same result and the phenom could dominate them all with almost double that. The intels are thus all limited by the motherboard's memory controller, while the phenom is limited by its own memory controller, which is superior.
Getting a 1gb ram video card should eliminate that dependancy, or maybe it is inherant in that game and cannot be eliminated, worth testing.

 

[Phynaz]

Originally posted by: Martimus

Originally posted by: Phynaz
It makes perfect sense. The cpu involvement of sending texture to a video card is for all practicle purposes zero.

The loading and usage of texture by the video card is what takes all the time.

Except the latency is about 3 times as much for an Q9450 than a Phenom. So access times will be slower in the Intel system. The CPU/Northbridge still has to do all of the accessing of the system memory. The GPU does not directly access the system memory. Here is a more indepth article on how a GPU will access system memory.

One more time:

It doesn't matter. The slowdown is in the video card. It doesn't matter that a Phenom can access memory a few ns faster, when it takes the video card 10's of ms to do texture management.

 

[Martimus]

Originally posted by: Phynaz

One more time:

It doesn't matter. The slowdown is in the video card. It doesn't matter that a Phenom can access memory a few ns faster, when it takes the video card 10's of ms to do texture management.

If that is the case, why have memory on the video card at all? Why not do all of the tasks using system memory if texture management is the real bottle neck and not memory access time?

 

[SlowSpyder]

Originally posted by: Phynaz

Originally posted by: Martimus

Originally posted by: Phynaz
It makes perfect sense. The cpu involvement of sending texture to a video card is for all practicle purposes zero.

The loading and usage of texture by the video card is what takes all the time.

Except the latency is about 3 times as much for an Q9450 than a Phenom. So access times will be slower in the Intel system. The CPU/Northbridge still has to do all of the accessing of the system memory. The GPU does not directly access the system memory. Here is a more indepth article on how a GPU will access system memory.

One more time:

It doesn't matter. The slowdown is in the video card. It doesn't matter that a Phenom can access memory a few ns faster, when it takes the video card 10's of ms to do texture management.

You keep saying what it isn't, any idea what you think it is? Looks like in at least a little bench the quote in your sig is correct.

 

[Phynaz]

Originally posted by: Martimus

Originally posted by: Phynaz

One more time:

It doesn't matter. The slowdown is in the video card. It doesn't matter that a Phenom can access memory a few ns faster, when it takes the video card 10's of ms to do texture management.

If that is the case, why have memory on the video card at all? Why not do all of the tasks using system memory if texture management is the real bottle neck and not memory access time?

Because video cards do much more than texture management?

 

[Phynaz]

Originally posted by: SlowSpyder

Originally posted by: Phynaz

Originally posted by: Martimus

Originally posted by: Phynaz
It makes perfect sense. The cpu involvement of sending texture to a video card is for all practicle purposes zero.

The loading and usage of texture by the video card is what takes all the time.

Except the latency is about 3 times as much for an Q9450 than a Phenom. So access times will be slower in the Intel system. The CPU/Northbridge still has to do all of the accessing of the system memory. The GPU does not directly access the system memory. Here is a more indepth article on how a GPU will access system memory.

One more time:

It doesn't matter. The slowdown is in the video card. It doesn't matter that a Phenom can access memory a few ns faster, when it takes the video card 10's of ms to do texture management.

You keep saying what it isn't, any idea what you think it is? Looks like in at least a little bench the quote in your sig is correct.

I have no idea, I haven't even read the article. But I can tell you that cpu memory access time for video card texture loads isn't it.


Anybody with a Phenom could test it, just overload your video card so you are texturing from system ram instead of local video memory. Notice the 80% drop in frame rates? There's nothing on the planet a cpu could do about that.

 

[OldPueblo]

I see this all over the net, people are starting to recognize that maybe we should be looking at our platforms in a different light. We don't spend all day running single monolithic tasks, the future is parallel processing just like we are already moving to with our video cards. Platform plays a very important part here and Intel has obviously recognized this with their move to the same type of platform that AMD has already (Nehalem). Sure AMD's processor is a little slower, but we spend more time out of game then we do ingame after all, what benefits do the Phenom platform have that we are missing because we aren't "benchmarking" it properly? I've used this analogy before:

A car that goes incredibly fast in a straight line but sucks at turning is only going to win a race where they are going in a straight line. If the track has many turns, it's going to take second place to a car that's only a little slower but corners very well.

The straight line track effectively seems to be the benchmarking methodology we are using today. I'd rather have excellent thread switching and move smoothly back and forth between lots of apps I multitask then get 10fps more in a game I play here or there. And for the record, I don't own a Phenom I have an e4400 c2d clocked to 3.2. It's just something I've noticed in all the forums I move around in. It's enough to make me want to buy a Phenom to try it out. For me, absolute high-end speed isn't the most important thing anymore, the actual computing experience is.

 

[Gikaseixas]

Originally posted by: OldPueblo
I see this all over the net, people are starting to recognize that maybe we should be looking at our platforms in a different light. We don't spend all day running single monolithic tasks, the future is parallel processing just like we are already moving to with our video cards. Platform plays a very important part here and Intel has obviously recognized this with their move to the same type of platform that AMD has already (Nehalem). Sure AMD's processor is a little slower, but we spend more time out of game then we do ingame after all, what benefits do the Phenom platform have that we are missing because we aren't "benchmarking" it properly? I've used this analogy before:

A car that goes incredibly fast in a straight line but sucks at turning is only going to win a race where they are going in a straight line. If the track has many turns, it's going to take second place to a car that's only a little slower but corners very well.

The straight line track effectively seems to be the benchmarking methodology we are using today. I'd rather have excellent thread switching and move smoothly back and forth between lots of apps I multitask then get 10fps more in a game I play here or there. And for the record, I don't own a Phenom I have an e4400 c2d clocked to 3.2. It's just something I've noticed in all the forums I move around in. It's enough to make me want to buy a Phenom to try it out. For me, absolute high-end speed isn't the most important thing anymore, the actual computing experience is.

my main rig is a phenom for that exact reason, i love how that system manages work loads thrown to it. My E8500 is awesome in gaming and benches better than the quad in a few scenarios but overall i rate the Phenom a little bit better.

edit: i'm getting Nehalem for sure, it looks sexy (no FSB)

 

[Markfw]

Originally posted by: OldPueblo
I see this all over the net, people are starting to recognize that maybe we should be looking at our platforms in a different light. We don't spend all day running single monolithic tasks, the future is parallel processing just like we are already moving to with our video cards. Platform plays a very important part here and Intel has obviously recognized this with their move to the same type of platform that AMD has already (Nehalem). Sure AMD's processor is a little slower, but we spend more time out of game then we do ingame after all, what benefits do the Phenom platform have that we are missing because we aren't "benchmarking" it properly? I've used this analogy before:

A car that goes incredibly fast in a straight line but sucks at turning is only going to win a race where they are going in a straight line. If the track has many turns, it's going to take second place to a car that's only a little slower but corners very well.

The straight line track effectively seems to be the benchmarking methodology we are using today. I'd rather have excellent thread switching and move smoothly back and forth between lots of apps I multitask then get 10fps more in a game I play here or there. And for the record, I don't own a Phenom I have an e4400 c2d clocked to 3.2. It's just something I've noticed in all the forums I move around in. It's enough to make me want to buy a Phenom to try it out. For me, absolute high-end speed isn't the most important thing anymore, the actual computing experience is.

Well, for me, its F@H. My Intel quads run 5000 ppd, and my last X2@2.5 was 1000 ppd, so even doubling that to 2k to account for more cores wouldn't touch my Intel boxes.

 

[taltamir]

the analogy kinda breaks down... what we have here is two cars.. one wins 19 race tracks. the other wins 1 race track by finishing it in half the time.
This is a very much WTF moment, and has nothing to do with platforms.

The qeuestion asked is "what is it with that 20th track that makes car B actually win in it, and by such a margin...

And those aren't artificial tests (straight line). Those are all REAL tests, like FPS in games and encode times in various apps.

 

[OCGuy]

Originally posted by: taltamir
the analogy kinda breaks down... what we have here is two cars.. one wins 19 race tracks. the other wins 1 race track by finishing it in half the time.
This is a very much WTF moment, and has nothing to do with platforms.

The qeuestion asked is "what is it with that 20th track that makes car B actually win in it, and by such a margin...

And those aren't artificial tests (straight line). Those are all REAL tests, like FPS in games and encode times in various apps.

a?nom?a?ly

NOUN:
pl. a?nom?a?lies
1. Deviation or departure from the normal or common order, form, or rule.

2. One that is peculiar, irregular, abnormal, or difficult to classify

 

[Martimus]

Originally posted by: Ocguy31

Originally posted by: taltamir
the analogy kinda breaks down... what we have here is two cars.. one wins 19 race tracks. the other wins 1 race track by finishing it in half the time.
This is a very much WTF moment, and has nothing to do with platforms.

The qeuestion asked is "what is it with that 20th track that makes car B actually win in it, and by such a margin...

And those aren't artificial tests (straight line). Those are all REAL tests, like FPS in games and encode times in various apps.

a?nom?a?ly

NOUN:
pl. a?nom?a?lies
1. Deviation or departure from the normal or common order, form, or rule.

2. One that is peculiar, irregular, abnormal, or difficult to classify

Anomalies happen for a reason, and we are just trying to figure out why. When I test a component and it does something it isn't supposed to do, you better believe that I will track down why it does it.

 

[jones377]

Can we simply not track down two people, one with an Intel rig and one with Phenom and have them test this?! COH should be a sufficiently popular game

 

[corsa]

words getting around....

AMD Phenom 8750 Triple-Core vs Core Duo E8400 in "Real Game Resolution "@ Extremetech.com

2.4Ghz x3 on Gigabyte GA-MA78GM-S2H vs 3.0Ghz Duo on ASUS Blitz Formula

Phenomenal :Q ..Its like a nightmare isn't it Intel

To Markfw900...use Firefox with AdBlock Plus, very affective :laugh:

 

[taltamir]

mmm, no this is a triple vs dual core setup. we were talking about a case where quad vs quad the phenom won. The performance of the triple core there is not very inspiring considering it has a whole extra core, it is however similarly priced and faster before OC. So it is a case where AMD is actually competative, and offers a better deal.