My 8800 GTS & 7900 GTX CPU Scaling Tests

[BFG10K]

Some preliminary numbers while we wait for nVidia to add back the xS modes in order for me to do a proper comparison of the 7900 GTX vs 8800 GTS.

Click.

 

[Matt2]

Interesting read.

In your opinion is the 8800GTS is held back by the CPU than the 7900GTX is as many continue to claim?

 

[BFG10K]

In your opinion is the 8800GTS is held back by the CPU than the 7900GTX is as many continue to claim?

The faster the video card, the more it's potentially bottlenecked by the CPU. This is a standard rule that applies to any video card.

 

[Elfear]

Thanks for taking the time to do that BFG. It's interesting that even at pretty high resolutions and eye-candy the processor still made a difference.

 

[josh6079]

Nice finds! This makes me look even more forward to moving to the E6600 from my 4400+ X2.

Not to trail off of the CPU attention, but how much does the DDR2 RAM influence such scores? When do we know whether or not the frame increases are from the more efficient CPU, higher-bandwidth RAM, or both?

 

[BFG10K]

When do we know whether or not the frame increases are from the more efficient CPU, higher-bandwidth RAM, or both?

An interesting question and the answer is that we don't. The only way would be to clock the RAM down to 400 MHz but that opens up other issues such as different memory controllers and the fact that the Athlon has an on-die memory controller.

It's probably best just to wrap the whole system package under E6600 and leave it at that.

 

[CP5670]

Do you have a way of measuring the maximum and minimum framerates over the benchmarks? It would be nice to see how much those are affected by the processor on an 8800. I did some tests with that about a year ago (with an Opteron 146 at 2ghz and 3ghz) and found that in most games I tried, the maximums greatly increased with the overclock while the minimums stayed about the same.

 

[CKXP]

Originally posted by: CP5670
Do you have a way of measuring the maximum and minimum framerates over the benchmarks? It would be nice to see how much those are affected by the processor on an 8800. I did some tests with that about a year ago (with an Opteron 146 at 2ghz and 3ghz) and found that in most games I tried, the maximums greatly increased with the overclock while the minimums stayed about the same.

Really, I always thought in most cases a higher clocked CPU will have more of an impact on min. fps compared to max. fps.

I remember FARCRY was a good example of this. back when I had a A64 3200 Newcastle, I noticed a big improvment in my min. fps (and overall) when OC'd to 2.6ghz when compared to stock 2.2ghz

 

[allies]

Hmm... couldn't there be a difference there since you're using a single core Athlon vs. a dual core C2D? Aren't nvidia's drivers "smp enhanced" (even if only a couple %, the results would be closer IMO).

 

[BFG10K]

Do you have a way of measuring the maximum and minimum framerates over the benchmarks?

Yes but I didn't take any this time. Sorry.

Really, I always thought in most cases a higher clocked CPU will have more of an impact on min. fps compared to max. fps.

If you're CPU bound a faster CPU will usually improve minimum framerate.
If you're GPU bound a faster CPU will usually improve maximum framerate.

Hmm... couldn't there be a difference there since you're using a single core Athlon vs. a dual core C2D?

Some of the difference is undoubtedly coming from the second core but again that's part of the total package, just like the faster RAM.

 

[Conky]

Thanks for this.

I think an additional set of tests with the E6600 overclocked might show the scaling in a more comparable way and remove the various platform difference questions.

It would sure be interesting anyway. *hint hint*

 

[josh6079]

It's probably best just to wrap the whole system package under E6600 and leave it at that.

That or for $57.00 we could find out for sure.

I believe though that there will still be a good amount of increases by the more efficient Intel core. I'm specifically interested in one for the fact that MMORPG's are becoming more of what I play. Do you have any game in that genre that could be used in a similar test? MMORPG's can really stress the CPU.

 

[Noema]

Originally posted by: josh6079

It's probably best just to wrap the whole system package under E6600 and leave it at that.

That or for $57.00 we could find out for sure.

I believe though that there will still be a good amount of increases by the more efficient Intel core. I'm specifically interested in one for the fact that MMORPG's are becoming more of what I play. Do you have any game in that genre that could be used in a similar test? MMORPG's can really stress the CPU.

That's a nice board josh. I could upgrade to Conroe with it for less than $300, and wait for the G80 refresh while keeping my still semi-decent 7800GS

 

[josh6079]

That's a nice board josh. I could upgrade to Conroe with it for less than $300, and wait for the G80 refresh while keeping my still semi-decent 7800GS

Indeed it is a good way to go for one who wants a cheaper upgrade. It's always nice to have companies like them provide easy, alternative upgrade paths. Not to mention for a test such as the one BFG10K was doing it would be more ideal and more frugal than taking the other route with an AM2 setup with DDR2 memory.

Perhaps when you purchase one you could run some numbers for us and add to this thread?

 

[Conky]

Originally posted by: Noema

Originally posted by: josh6079

It's probably best just to wrap the whole system package under E6600 and leave it at that.

That or for $57.00 we could find out for sure.

I believe though that there will still be a good amount of increases by the more efficient Intel core. I'm specifically interested in one for the fact that MMORPG's are becoming more of what I play. Do you have any game in that genre that could be used in a similar test? MMORPG's can really stress the CPU.

That's a nice board josh. I could upgrade to Conroe with it for less than $300, and wait for the G80 refresh while keeping my still semi-decent 7800GS

That's exactly what I did with this board. I used it as a cheap upgrade path to C2D last August since I was stuck with DDR and AGP. And I've upgraded the RAM and videocard to DDR2 and PCI-E(7900GS) and will make the jump to a "regular" DS3 motherboard later this month.

 

[jim1976]

Thanks for taking the time to do the tests BFG10K.. :thumbsup:

 

[Smartazz]

nice tests. Not a big enough diffrence to make me go core 2 duo, so I'm relieved.

 

[Smartazz]

Do you think this diffrence is because this is a single core vs. a good dual core or because the Core 2 Duo is just that much faster. We need a comparison of the X2 vs. the Core 2 Duo. If I get an 8800GTS, I'll do some tests to find out.

 

[Tanclearas]

I was thinking the same. The biggest differences exist in older games where the frame rate is already pretty high anyway. Think I'll be holding off on a CPU upgrade for a while longer.

 

[secretanchitman]

im curious, how did you get descent 3 to work under xp?

 

[BFG10K]

you have any game in that genre that could be used in a similar test?

No, sorry. I have a couple of old flight sims but that's about it. The only RTSes I play are Starcraft and Warcraft 2.

Not a big enough diffrence to make me go core 2 duo, so I'm relieved.

Keep in mind that when you are CPU limited the difference will be quite large. For example in Vampire Bloodlines my framerate almost doubled when I'm outdoors and that's at 1920x1440 with 8xS.

im curious, how did you get descent 3 to work under xp?

Just patch it to 1.4/1.5 and it'll work fine under Direct3D or OpenGL.

 

[anandtechrocks]

Nice work as always BFG10K. Thanks for sharing.

 

[LittleNemoNES]

wow different reaction here than from the ars-technica site.

I have to say that although the gains are always in favor of BFGs hypothesis, the driver differences can't be ignored. Nice pre-liminary results, IMO. Once the 8 and 7 series can share a common driver I think it will be more appropriate to make an apples to apples comparison. As always, I could be wrong

A low clocked x6800 vs an OCed x6800 would be ideal, I tank'.

 

[Tegeril]

I'm AMAZED at how ridiculously misdirected the responses are on the ars forums. They are just arguing for the sake of arguing and it's stupid.

And gersson, this is not a GPU comparison, so the graphics cards don't matter. They are a constant so that the CPU can be changed. It wouldn't matter if it were an x800 and a quad sli 7950 because what he aims to show is that if you have a given GPU, you will see x% difference if you change CPUs.

I don't get why this is so hard for the ars people to get.

 

[BFG10K]

I don't get why this is so hard for the ars people to get.

You and me both Tegeril.

I expected a debate but some posters there are trying to see something in the numbers that isn't there. As you pointed out correctly it's not a GPU comparison.