Why run a graphic test on a CPU?

[jeffa123]

For years now Futuremark - Madonion, have used a cpu mark in with their graphics tests and to be quite honest it shows haow little the CPU is used during graphics applications with the exception of Photoshop programs which can be both cpu and graphic intensive. But I don't see a CPU photo-intensive test done on 3dmark06. In fact the whole test seems, for all intensive purposes, to be an Ego test and not a fully professional test. What I would like to see coming from Futuremark is an explanation, even if it is on screen, for each test and what it is meant to achieve. So far it is rare that anyone bothers to watch the test because they are either off watching the TV, washing up etc. But since our graphic experience is dictated so much by one uttility shouldn't their tests be brought into the everyday market. Everyone whom owns a pc these day almost always owns a digital camera and plays games and I'll lay odds there are more. Gamers Benchmark should be updated to everyday graphic intensive bechmark.

 

[PinkPig]

I don't really get what you're getting at at all. 3DMark is made as a "gamer's benchmark", there are plenty of other benchmarks around for other experiences.

"But since our graphic experience is dictated so much by one uttility"

Huh? Are you saying that the public's PC experience is dictated by the 3DMark benchmark? If not, what are you saying? If so then I disagree completely, very few people care in the slightest.

Modern games make big use of both the CPU and the graphics card, and 3DMark06 is the first 3DMark benchmark that specfiically includes a CPU test in calculating a score.

If you're saying that 3DMark should be a general benchmark rather than a gaming benchmark then find some other software - Futuremark's own PCMark, for example, or a whole string of other programs around that test general PC performance.

 

[jeffa123]

But how much CPU time it used by current games? The only edict for this is DirectX, Opengl uses far less cpu clocks because it uses internal drivers not controlled by DirectX. But games themselves are only using the DirectX libraries as reference. In simple Laymans terms a sound instruction is sent via DX to a play.DLL etc. etc. DirectX is an input output map. How many people play games in DIrextX's software mode? Is 3Dmark saying something to these users because it is not telling me anything about my CPU. A CPU, Intel or Amd has to carry out a lot more instructions other than dealing with graphics. In some principal I do not disagree with you but would it be better if the tests were carried out as a combination instead of sitting through some rendition of jerk wars. Updating a screen is not the CPU's job its the graphics cards. If you can agree that you purchased a graphics card for graphics and ease your CPU useage then you can see my point.

 

[welshtom]

try a p3 500mhz with a x1900 and see what score you get in 3dmark 05 (i know this is impossible).

then come back and tell us how much a cpu is used


Even a A64 3000 holds back an X1900. the card isnt used to its full potential because the cpu isnt processing what it has to do fast enough

 

[TheMightyTen]

But how much CPU time it used by current games? The only edict for this is DirectX, Opengl uses far less cpu clocks because it uses internal drivers not controlled by DirectX. But games themselves are only using the DirectX libraries as reference. In simple Laymans terms a sound instruction is sent via DX to a play.DLL etc. etc. DirectX is an input output map. How many people play games in DIrextX's software mode? Is 3Dmark saying something to these users because it is not telling me anything about my CPU. A CPU, Intel or Amd has to carry out a lot more instructions other than dealing with graphics. In some principal I do not disagree with you but would it be better if the tests were carried out as a combination instead of sitting through some rendition of jerk wars. Updating a screen is not the CPU's job its the graphics cards. If you can agree that you purchased a graphics card for graphics and ease your CPU useage then you can see my point.

So when playing a game you just wamt to run around pretty looking levels with nothing else going on. You don't want enemies adapting to you, physics we can do away with that as well. Just because a game pushes the edge in terms of graphics and GC capabilities doesn't make the cpu irrelevant, it just frees up the CPU to do all the things that make the game worth while.

Most bench marks are synthetic and to that extent an inaccurate representation of real world performance. Get a copy of Fraps and play a few games to get a real world view.

 

[smids]

Because you can benchmark the CPU for comparison. A gfx card a stock with say a 2.0ghz cpu will get say x score but clock the cpu up 200mhz and you get y clock. The CPU mark allows people to compare graphics score on a comparable system and see whether it is the graphics card giving the score or cpu also. CPU's are heavily used in games anyway when calculating physics etc or enemy bots etc.

 

[semi-pro waster]

jeffa123 you appear to know this but don't really seem to count it - the CPU is still responsible for calculating the physics(otherwise why would a separate PhysX chip be possible), the AI, loading maps etc. This means that yes updating the screen is the graphics cards job but without a strong CPU to back it up then it is never going to give you a decent gaming experience.

Rendering a screen solely from CPU power even without having to determine physics or AI is still incredibly intensive(hence the jerk-o-vision) which gives a reasonable indication of how powerful the CPU is and how ably it will support the graphics card.

 

[james.miller]

I think you guys are missing his point. The only difference between say.....the gpu accelerated flying ship scene in '05 and the fully cpu driven scene is that the gpu is handling the acceleration. *** cpu is still doing the same physics calculations and everything else that's required in both tests.


a year ago i would have it would have made more sence for the cpu tests to be pure physics tests, but with the impending release of the physics accelerating cards, i dunno what the situation will be

I personally have never seen the point of those particular cpu tests either, but yes cpu tests are still needed, just maybe something a little more suitable. Current cpu tests in futuremark programs consistes of turning all gpu acceleration off. nobody does that.

 

[jeffa123]

A lot of you have picked up on my point because your all saying that both the CPU and Gpu control the hardware. So why switch off the thing your using the benchmark to test. Yes I gree try a PII or PIII and you would get bad results but thats just the same as buying an AMD chip and not installing a GART control driver. You end up with jerks. Since we agree that both CPU and GPU handle the graphics why not include the cpu score into the GPU accelleration tests. Or is that too intelligent. Or are there users out there still trying to play QUAKE 4 on an Intel 386 PC. I am sorry but the CPU score test in its present format makes no sense to me wether I have one of my old ATI 9700pro boards, Nvidia 5900 ultra, 6800 GE, 6800 Ultra or my SLI 7800GTX installed. It makes the test laborious and time consuming and pretty pointless.
If I found a logical reason for it still to be there then I would have no real arguement but logically it doesn't deserve to be part of the test. Surely Futuremark have better things to do.

 

[smids]

Have you ever done some experiments in science? When testing things, you have to keep everything else constant so that the test actually means something. Not everyone has the same CPU. CPU's of different speed will give different scores in both tests (CPU and GPU). To make a fair comparison, people can search the orb for people with everything else the same including CPU score, thus allowing people to make a comparison between cards alone, and not cards with CPU. Hence the test.

 

[jeffa123]

I accept your reply Smids but you are not taking into account that benchmarking as well as science is not a constant. Why would I like to compare my projects with others which mean absolutely nothing to me. e.g. Someone may have the same setup as me and might overclock theirs to an ungainly amount which may give them an advantage in speed over my system but that does not tell me what they are doing to get that benchmark and why. To me a bench mark is there for a reference point and if you are a true scientist you must accept the Chaos Theory. What may be constant one minute is not the next. If you were to run the same benchmark over and over again for 3 or 4 hours will it tell you what your PC can do or will it just give you a reference point for discussion. Out of all the tests you would have to take a reference point between the highest benchmark and the lowest in true scientific terms. So when you submit to Orb you are saying my machine performed this well or this badly at this point in time. The whole issue of CPU liability goes out the window. Stick to facts. Is the cpu test in 3Dmark pointless because no game switches off graphic acceleration, so why should it even be there. True or false? if its a Gamers benchmark then like my previous answer encorporate the CPU test into the Graphics test in order to reach true graphics benchmarks.

 

[smids]

I don't quite see what chaos theory has to do with this unless you are talking about the random stray electron theory and I doubt something like this would produce such an anomaly as to warrant the idea that it has been affected in any way.

The point is, 3DMark uses and relies on the CPU even when running the graphics tests - proof, take a 2.0ghz A64 and run it, then try with a 2.6 ghz processor - yes, the scores are different even though the exact same test and setup was run except the CPU. What are the chances of 2 people running exactly the same system etc? Very low is the answer. Systems might be similar but there are so many more variables like RAM timings on both the card and the system RAM, FSB, RAM frequency etc. The whole point of the CPU score is to put all that into perspective and produce a score which is representative across the board irrespective of the graphics card used so long as they all render using the same method (which 3DMark ensures). Thus people can conclusively say whether their system, when contrasted against another can be down to the system or the card, this being the point of the benchmark.

The whole point is for comparison, that is what benchmarks aim to achieve and this is the reason it is included. It is not like it took them time to devise the test either, so what if it is an extra test, you can always disable it if you don't want it to run.

 

[Jack Bauer]

To me a bench mark is there for a reference point and if you are a true scientist you must accept the Chaos Theory. What may be constant one minute is not the next.

I am a scientist and I can honestly say that I have never had to consider Chaos theory in any studies I have done..
At the end of the day its a benchmark like smids said, I dont see how any more explanation is nessacerry. It compares systems without the graphics card.

 

[smids]

I am a scientist and I can honestly say that I have never had to consider Chaos theory in any studies I have done..
At the end of the day its a benchmark like smids said, I dont see how any more explanation is nessacerry. It compares systems without the graphics card.

Yeah I did 2 science A-levels (though am not a scientist) including Physics and can honestly say, I never once even thought or suggested that my results might be affected by chaos theory - if that was the case, then all of science is a fallacy as it is all warped by random things causing random results. The fact that a class full of students can pull off roughly the same results is a testament to that. And the fact that people can prime computers for hours on end whilst crunching billions of numbers and still not produce errors shows how uncommon a stray electron affects results.

 

[james.miller]

yet you've both chosen to ignore one point i made

Current cpu tests in futuremark programs consistes of turning all gpu acceleration off. nobody does that.

 

[smids]

yet you've both chosen to ignore one point i made

True, your point is correct, but the only way to compare a system with another (i.e. the CPU score) is to take the graphics card out of the equation. By disabling all acceleration, the CPU is doing all the rendering, as I'm sure you know, meaning systems can be benched no matter what card is present. This allows you to compare your system with another system and see if the system plays a big part in the GPU scoring the score it does, or if it is just the graphics card.

 

[james.miller]

yeah but it has no real world use. what good is telling us how fast the cpu can push pologons around when nobody wnat to do that with a cpu? a better test would be something physics bound - at least then we would have a usefull benchmark.

What they have atm is as useful as a sandra cpu bench for telling us how fast are pc's really are.

 

[smids]

yeah but it has no real world use. what good is telling us how fast the cpu can push pologons around when nobody wnat to do that with a cpu? a better test would be something physics bound - at least then we would have a usefull benchmark.

What they have atm is as useful as a sandra cpu bench for telling us how fast are pc's really are.

LOL - but you have to accept that whilst it is totally useless as a standalone benchmark, it is required for comparison purposes.

Say I have a 7900GTX. My friend has an X1900XTX. We have similar but different systems. I have an FX60 and he has an FX57. Everything else in our systems are the same.

We both run 3DM'06 to find out which card is better. He gets 5500 and I get 6000 (made up). Now, does this mean that his card is slower than my card? No, because they have been run on quite different systems. We therefore want to know which is actually faster.

We would have to look to the ORB for people with roughly the same CPU scores to be able to rule out the system as causing the difference in the cards' performance. The only way to do this is to find someone with a similar CPU score. An FX60 will score about 1900, whilst the FX57 about 1000. When you find someone similar, that means their card has had roughly the same raw power behind it as yours and you can make a comparison of which card is better. As it stands between me and my imaginary friend , there can be no comparison drawn. That is the point of the CPU test - to enable a good comparison, something which all benchmarks are geared up for.

I totally agree, the test alone is about as useful as Sandra 2005, but coupled with the graphics score, can help people make cross comparisons.

 

[Steves_pc]

Granted we wouldnt bother to turn off the gfx cards features, but is that just done as a representation as to how well the cpu/bus etc can perform?, albeit done in a graphical manner as opposed to running prime etc.

I agree with the OP, an explanation of what the tests are for or supposed to show would be more helpful instead of just being eye candy (the 1st time anyway)
Personally it seems more of a competition to get the highest score instead of a usefull benchmark to highlight system inadequacies.

 

[jeffa123]

Admittedly there are a hell of a lot of users whom tend to put a half decent PC together only to overclock it so much (so that it only lasts 5 minutes or so after the test) so they can post the Highest score. A processor clocked over the rate it was sold at invalidates its warranty and the same goes for graphics GPU's and Ram. This is a personal thing and nothing against those whom do, but I prefer to buy a piece of equipment and doing with it what the manufacturer designed it to do. If it doesn't measure up then it goes back. If I bought a CPU which was supposed to work at 3,8 GHZ and it worked at 2.8 it would go back. Gone are those days when we could fry an egg on both the Motherboard and AMD Athlon 1.2 GHZ frying underneath it. The heart of a processor is the Motherboard and if the motherboard's timings are out and your Southbridge has gone awry then it doesn't matter what GPU or CPU you have onboard...........Everything and nothing matches up so you end up with a badly constructed machine. As to my reference to the Chaos theory try to imagine all those machines being overclocked to get that one benchmark that puts them on top. How many millions use a benchmark to test a machines stabillity and the many other reasons a benchmark can be used. There is no one result and no one combination which can give an absolute under the circumstances of one BM. If you are looking for comparison fine but at any one given moment some CPU task might cause an interrupt and the BM changes. Sandra Asks you not to even touch your mouse when testing but that still does not prevent USB queue interrupt and link library overrides which for any encounter for a BM must make a test in true state be false.
To have a truer test you would have to boot directly into the benchmark with no O/S in the background in order to give an honest reading from CPU to GPU and RAM effectiveness. I must be boring you by now so I will finish on that note.