9700 pro - the limiting factor?

[Rambler]

i've read a review of p4 3.06 procesor at http://www.guru3d.com/review/intel/pentium4-3060/ and something caught my attention.

the guy there says:

Before we start off with the gaming benchmarks I need to explain something to you to understand what is happening to the test-results. Logic assumes that you need to take a look at the highest resolution tested versus the score. This however is not the case.

The CPU's used are faster than a graphics card can handle. So in the highest resolutions you'll notice not much difference in the results. This is the videocards bottleneck as it can't push any harder. What you need to monitor are the lowest resolutions as the videocard is not limited in raw power or memory bandwidth. I know this may sound a bit weird but the Radeon 9700 Pro that we used is actually a limiting factor for the CPU in the highest resolutions. Bare that in mind as we try to demonstrate the power of the CPU and not the graphics card

Well i was always under the impression that current cpus are holding the 9700 pro cards back not vice versa.
Plus my another impression was that for demonstrating pure cpu speed low resolutions are usually used.
But i have to admit, that then i don't know how to explain that he get's same scores at 16x12 on p4 2.4 and 3.06.

Is the guy right about 9700 pro being the limiting factor or has he done some basic mistakes in that review?

 

[demalion]

I'd say within the range of CPUs he is testing, and with the resolution, settings, and games utilized, and with the criteria for "bottleneck" (i.e., any change in performance as EDIT: resolution increases) provided, yes, the point seems valid given the results.

I mean, if fps goes down between 1280x1024 and 1600x1200, it is the graphics card that is responsible for the slowdown, isn't it? Whether dropping from, for example, 120 to 100, warrants the label "bottleneck" simply depends on your perspective.

 

[Althornin]

I'd say within the range of CPUs he is testing, and with the resolution, settings, and games utilized, and with the criteria for "bottleneck" (i.e., any change in performance as EDIT: resolution increases) provided, yes, the point seems valid given the results.

I mean, if fps goes down between 1280x1024 and 1600x1200, it is the graphics card that is responsible for the slowdown, isn't it? Whether dropping from, for example, 120 to 100, warrants the label "bottleneck" simply depends on your perspective.

every current video card is "the" bottleneck at really high resolutions with AA and aniso.

 

[Doomtrooper]

Other factors need to be taken into account here also, CPU is just one area, also important is BUS speed, memory timings and AGP interface..so platform limited would be my preference.

The other main factor is games Today utilize very little of the 9700's capabilities that would increase its peformance even more, almost all games today are still wtitten for the DX7 pipeline and doesn't even look at these fancy card features of the 9700..i.e vertex and fragment level processing...

So in essence the 9700 is bottlenecked by the software and hardware and speed we are seeing today done by the 9700 is just brute force (faster Ram, better z-rejection and faster VPU/GPU's)

And I don't see developers jumping on the Dx9 bandwagon yet, heck Dx8 is yet to come.

 

[Colourless]

There can also be multiple bottlenecks in any given benchmark. Some parts of the benchmark might be CPU limited and other parts are GPU limited. Who is to say that a faster CPU wouldn't still up the 1600x1200 score?

 

[Rambler]

So when you take a look at those scores - for example in ut2k3 10x7.6 -> 12x10 the frame rate drops 29fps on 2.4 and 47fps on 3.06, then 12x10 -> 16x12 2.4 drops 45fps while 3.06 49fps, Q3 12x10 -> 16x12 drops 47fps on 2.4 and whopping 63fps on 3.06, is it safe to assume that in this scenario it's limited by gfx card other than by cpu?

Also can anyone explain to me why in CodeCreatures he gets the EXACT same scores in all res. on both cpus? Does that mean it's cpu independent? Or the benchmark shows something completely different? Because i find it strange that it's slows down when increasing the resolution, but there's no gain when adding additional 600MHz??

 

[Althornin]

Also can anyone explain to me why in CodeCreatures he gets the EXACT same scores in all res. on both cpus? Does that mean it's cpu independent? Or the benchmark shows something completely different? Because i find it strange that it's slows down when increasing the resolution, but there's no gain when adding additional 600MHz??

My guess would be the game could be shader limited at low res, and fillrate limited at high res, so the CPU is effectively removed as a bottleneck.

 

[Crusher]

The other main factor is games Today utilize very little of the 9700's capabilities that would increase its peformance even more, almost all games today are still wtitten for the DX7 pipeline and doesn't even look at these fancy card features of the 9700..i.e vertex and fragment level processing...

So in essence the 9700 is bottlenecked by the software and hardware and speed we are seeing today done by the 9700 is just brute force (faster Ram, better z-rejection and faster VPU/GPU's)

You're saying the framerate would be faster if you gave the video card a bigger workload? That makes no sense. The card is still going to be the limiting factor at 1600x1200.

 

[Joe DeFuria]

You're saying the framerate would be faster if you gave the video card a bigger workload? That makes no sense. The card is still going to be the limiting factor at 1600x1200.

I think he's actually saying that it's not a bigger workload...but a more efficient workload. In other words, if certain games were coded with an DX9, 8 pipeline architecture in mind instead of a DX7 architecture, the engine could be more efficient and doing the same thing.

Doom3 Example: The NV1x path should be able to look just as good as the NV2x path. Both paths should run on NV2x hardware. However, I'd bet that the NV2xpath runs faster on the NV2x, because it takes advantage of the increased flexibility of NV2x shaders and pipelines...

 

[Crusher]

Yes, but the games tested weren't doing the same things Doom 3 does. If they were using applications that did DOT 3 register combines for per-pixel lighting, then I could see where his argument might almost make sense. As it stands, using DX9 to do the same things that these applications do would not make the cards any faster. And the card would still be the bottleneck at 1600x1200, regardless of what DX9 features they used. It's not system memory bandwidth, AGP bandwidth, or the fact that it's using DX7 multitexturing and T&L that's causing a 2.4 GHz processor to get the same framerate as a 3GHz processor at 1600x1200.

 

[Joe DeFuria]

Yes, but the games tested weren't doing the same things Doom 3 does.

Right, which is why Doom said: "The other main factor is games Today utilize very little of the 9700's capabilities that would increase its peformance even more, "

It's not system memory bandwidth, AGP bandwidth, or the fact that it's using DX7 multitexturing and T&L that's causing a 2.4 GHz processor to get the same framerate as a 3GHz processor at 1600x1200.

Certainly in agreement!