Originally posted by: munky
Not when it comes to increasing framerates.
Rofl what? So if framerates are above refresh, a faster CPU can't increase frame rates? LMAO.
You're comparing multi-gpu at 1280 to 1920. I can't stress enough how ridiculous that is.
Yep its necessary to find the "inflection point", to clearly demonstrate CPU bottlenecking, which I've done time and time again. What's ridiculous is the fact you still don't understand this simple metric even after nearly a decade of software/hardware reviews which has used CPU/GPU scaling as the standard for performance benchmarking.
You may as well throw 10 gpu's at a game with the fastest imaginable cpu and not necessarily get any higher fps. What's your point?
Except that's provably not the case, as reviews that do use faster CPUs do show an improvement in FPS. You just choose to ignore the results and want to blame the amount of RAM, the chipset, and basically anything else.
Single card and dual-card are done on a 2.93GHz system. Tri-SLI is done on a 4.2GHz system:
- Guru3D 280 GTX in SLI and Tri-SLI
Quake Wars again has the same IQ settings, yet is a little difficult to explain. With the baseline test system I mentioned that the performance of 2-way SLI was slower than a single card due to the the game being CPU bound. Two cards need more CPU cycles than one.
With the overclocked system at 4200 MHZ, we rule out that fact and give it much more CPU cycles. It shows clearly with 3way SLI, though Quake Wars still does not scale extremely well. Then again at nearly 90 FPS at 2560x1600 .. you can't really complain either 
- GRAW2
Again, same image quality settings are applied, everything is set to high. Now this is ridiculous to see, yet in the previous SLI results I already told you that we hit a CPU bottleneck with the 2.9 GHz Core 2 Duo Extreme processor.
This is the result of the massive overclocked CPU in combo with the power of 3 GPUs, it's just ridiculous how fast that really is ... but it can get even sicker though, let's move towards the game FEAR.
- UT3
We again see with all these cards in play that we have a CPU limitation which stops us from moving past that 140ish FPS mark.
Some of the most interesting results show that even with a single GPU, a faster CPU may yield better, or the same results as multi-GPU due to CPU bottlenecking. Feel free to look over the 4870X2 results, you'll see similar although most of the are using slow 3GHz CPUs and therefore, they have to use GPU limited settings like "free AA" in order to show any separation between other parts.
Again, your whole premise is based on the ridiculous idea that if you throw more gpu's at a game and don't get an improvement then it must be cpu limited.
Except I've reinforced that premise with examples that show using a faster CPU does yield improvement in previously CPU bottlenecked situations. There can certainly be other limitations, like driver or game engine frame caps in some titles, but its quite obvious faster CPUs are needed for the current GPU solutions. Either that or Devs are going to have to code better for multi-CPU to take advantage of current CPUs or make more GPU intensive games.
Until a next gen high end gpu arrives, that may be the only way of reaching 50-60 fps at high rez, an you're claiming that what we really need are faster cpu's... :roll:
Why would we need a faster next gen high end GPU if we don't get faster CPUs and it'll still be bottlenecked anyways? You've already argued how adding a 2nd GPU "always adds more FPS", except when it doesn't (at 1280 or when CPU overhead drops it lower than single), so why would you dismiss multi-GPU or ignore the fact multi-GPU requires more CPU cycles than single in order to scale? I've already tried explaining this to you, if a CPU can produce a certain number of frames per second and modern GPUs are already reaching that limit, how would adding a faster GPU help?
I got news for you: nobody plying ME at 76fps is gonna whine about being bottlenecked by anything.
So is 76FPS fast enough or not? Once again you contradict yourself in an attempt to try and prove a point without even remembering what you posted in the previous reply. What happened to:
The "corrolation" you're seeing is the result of not having a fast enough single gpu to run modern games at high rez and high framerates, not some magical cpu-limitation theory you invented.
Despite satisfying your subjective 76FPS requirements, I'm quite sure anyone interested in SLI would like to know multi-GPU will only yield "Free AA" beyond a single card due to CPU bottlenecks.
Another news flash: Nobody with half a brain is paying 2-3x as much for SLI/CF when they're getting 60+ fps with a single gpu.
Of course they're not when they're only getting "Free AA" and the same low frame rates.
If you already had drops below 60 and you enabled vsync, would you still get those drops or not? Vsync has no relevance to the topic, and you had no reason of mentioning it other than debating more pointless trivia.
Vsync absolutely has relevance. You were claiming 60-80FPS averages "were enough" because of 60Hz refresh, at which point I said that is false as you would still see frame drops below refresh
UNLESS you had Vsync enabled. So again, how would you have frame drops below refresh with Vsync enabled with an average FPS of 60-80? You can't. So once again, unless Vsync is enabled and you are averaging 60FPS, your subjective 60FPS and 60Hz "is enough" argument is ridiculously flawed. I've already shown clearly a 62FPS Average doesn't mean crap, much less being "enough" with the COD4 FPS vs. Time graphs.
Those drops are not caused by cpu bottlenecking, so what's your point?
My point is that you don't know that for certainty, as the FPS vs. Time graphs showed there is meaning in the detail, to which you replied "Those details don't matter when the person playing the game won't notice a difference." I can certainly see the difference between 20-30FPS and 60FPS. If adding a faster GPU or more GPUs isn't increasing frame rates but adding a faster CPU does I think its pretty safe to say the CPU is indeed the bottleneck.
Are you the guy who claimed SW has nothing to do with HW? Because what you just said is no less ridiculous.
Nope, but how does your example have anything to do with what you claimed, how multi-GPU is inefficient by design? How does sharing data between frames make single-GPU more efficient than multi-GPU when multi-GPU simply makes a copy for both GPU in discrete frame buffers? Once again, multi-GPU isn't inefficient by design, its even been shown to scale beyond 100%. That alone proves your claim is impossibly false.
Which again is irrelevant for the reasons I already mentioned.
And your subjective reasons have no bearing on the objective evidence showing clear CPU bottlenecking. That's what it really boils down to, I show clear evidence of CPU bottlenecking, you claim it doesn't matter because its above whatever subjective threshold you set for that particular example.
Your 48fps theory was
already proven wrong, which makes the argument moot.
It actually furthers my point, just as your last example did (by showing higher FPS with a faster CPU), but thanks.
Nobody with half a brain spends 2-3x as much on multi-gpu unless they need it at high rez because it's the only way of getting playable framerates at the moment.
And nobody with half a brain would try to dismiss such clear evidence of CPU bottlenecking whether they think a single GPU is capable of producing playable framerates or not.
Exactly which game improved by 3-4 fps at 1920?
The same titles I mentioned earlier....AC, Witcher, Oblivion....
What happened to shifting the whole paradigm forward? Crysis is gpu-bound to 17fps, and you're rejoicing because your minimum increased from 5 to 12? How about showing something at settings people would actually use?
I said it
can shift the entire result set, as it clearly did for WiC. Its even more obvious in cases where a single GPU with a faster CPU is outperforming multi-GPU with a slow CPU. Without seeing duration of the test and an FPS vs. Time Graph its impossible to say for sure but one obvious possibility is that Crysis is more GPU intensive than CPU intensive at that setting of Very High, which shouldn't be a real surprise....
And yes, if you can increase your minimum frame rates by 75% that would be reason to be excited, especially if minimums were single digits.....
Facebook
Twitter