Slashdot Mirror
CPUs Do Affect Gaming Performance, After All
crookedvulture writes "For years, PC hardware sites have maintained that CPUs have little impact on gaming performance; all you need is a decent graphics card. That position is largely supported by FPS averages, but the FPS metric doesn't tell the whole story. Examining individual frame latencies better exposes the brief moments of stuttering that can disrupt otherwise smooth gameplay. Those methods have now been used to quantify the gaming performance of 18 CPUs spanning three generations. The results illustrate a clear advantage for Intel, whose CPUs enjoy lower frame latencies than comparable offerings from AMD. While the newer Intel processors perform better than their predecessors, the opposite tends to be true for the latest AMD chips. Turns out AMD's Phenom II X4 980, which is over a year old, offers lower frame latencies than the most recent FX processors."
Try cranking up the difficulty of an RTS on a not-so-good computer and you'll immediately notice how things slow down
Which idiot made that claim? Pretty much every hardware review site has CPU and GPU dependent games in their reviews when they review GPUs, CPUs, and OTB rigs.
Who thought that CPU's didn't bottleneck gaming performance? Who ever thought that? Only the smallest of tech demos only used GPU resources - every modern computer/console game I'm aware of uses, well, some regular programming language that needs a CPU to interpret instructions and is inherently limited by the standards of clock cycle and interrupt tied to those CPUs.
GPUs only tend to allow you to offload the strait-shot parallelized stuff - graphic blits, audio, textures & lighting - but the core of the game logic is still tied to the CPU. Even if you aren't straining the limits of the CPU in the final implementation, programmers are still limited by the capacity of them.
Otherwise, all our games would just be done with simple ray-traced logic, using pure geometry and physics, there would be no limits on the number or kind of interactions allowed in a game world, game logic would be built on unlimited tables of generated content, and we'd quickly build games of infinite recursion simulating all known aspects of the universe far beyond the shallow cut-out worlds we develop today.
But we can't properly design for that - we design for the CPUs we work with, and the other helper processors have never changed that.
Ryan Fenton
I don't recall ever reading on any PC hardware site anyone claiming that the CPU doesn't matter and all you need is a good graphics card. How on earth did anyone ever successfully submit that story?
The research into frame-rate latencies is really interesting, but the whole idea that *anyone* knowledgeable about PC gaming would have *ever* denied that the CPU was an important factor in performance is ridiculous. I am a consultant at a boutique PC builder (http://www.pugetsystems.com/) and I have always told gamers they want to get a good balance of CPU and GPU performance, and enough RAM to avoid excessive paging during gameplay. Anything outside of that is less important... but to ignore the CPU? Preposterous!
Then again, it is a Slashdot headline... I probably should expect nothing less (or more)!
William George
For years, stupid PC hardware sites have maintained that CPUs have little impact on gaming performance; all you need is a decent graphics card. That position is largely supported by FPS averages, as most GPU tests are run using the most powerful CPU to prevent the CPU from being the limiting factor, but the FPS metric doesn't tell the whole story. Examining individual frame latencies better exposes the brief moments of stuttering that can disrupt otherwise smooth gameplay. Those methods have now been used to quantify the gaming performance of 18 CPUs spanning three generations by some site that really has nothing better to do than to restate the obvious for morons. [ed: removed fanboy-baiting statements from summary]
This should be obvious to anyone who has done any realtime/interactive graphics programing. As the frame rate gets higher the amount of time the CPU has to process the next frame gets smaller. It also becomes more diffcult to properly utilise the CPU fully unless you are willing to add a couple of frames of latency to generate frames in the future which I'd speculate is not ideal for a game type application.
so... i should finally give in and buy the coprocessor for my 386!!
Mine is a 2600 also! But mine is made by Atari because I wanted a system made for gaming. I'm sure they're pretty much the same thing though.
Hm. First there is:
"...The FX-4170 supplants a lineup of chips known for their strong value, the Athlon II X4 series. Our legacy representative from that series actually bears the Phenom name, but under the covers, the Phenom II X4 850 employs the same silicon with slightly higher clocks."
and then:
"Only the FX-4170 outperforms the CPU it replaces, the Phenom II X4 850, whose lack of L3 cache and modest 3.3GHz clock frequency aren't doing it any favors."
How can I trust them if they are unaware of basic stuff any chip enthusiast should know? (The Phenom is the Athlon with level 3 cache. The Athlon has none.) They could have also touched on what the 2 AMD specific hotfixes were for.
I'm not shocked at the results, but I am skeptical of the degree of disparity.
They said frame rate *latencies* increased with the FX..... not that the frame rates went down.
My AC stalker: " I personally agree with your posts most of the time, but that won't keep me from modding you troll"
as this article points out it's not the number of frames per second that really matters:
it's the longest gap between subsequent frames which the eye picks up on.
you could cram 200 frames into the last 10th of a second, but if the other 0.9 seconds only has 1 frame, it'll feel like 1Hz.
i typically chart another metric next to traditional FPS which is 1 / (max inter-frame period in one second).
In a game, look at the sky. If your framerate shoots up, the video card was your bottleneck. If it doesn't, your CPU is.
The Internet King? I wonder if he could provide faster nudity.
I will take a mediocre cpu with a kick ass GPU than the other way around. Sure I have an under clocked phenom II at just 2.6ghz but with my ATI 7870 I plan to get it will blow away an icore7 extreme with the HD 4000 graphics by several hundred percent!
GPU is where it is at with games. Just like with Windows an SSD makes a bigger difference than a faster CPU booting up.;
http://saveie6.com/
Your eyes don't blur anything, blur you refer to (probably read about it somewhere) is why film looks smooth at 24 FPS, well it's because frames on FILM are blurred, your eyes have nothing to do with it. 30 FPS may look smooth but I definitely notice big difference between 30 and 60, even 60 and 90 looks different.
GPUs only tend to allow you to offload the strait-shot parallelized stuff - graphic blits, audio, textures & lighting - but the core of the game logic is still tied to the CPU. Even if you aren't straining the limits of the CPU in the final implementation, programmers are still limited by the capacity of them.
Your theory is basically valid, but the practical reality and the empirical evidence of the last, I dunno, 20 years or so, is that the graphics processing takes a significant amount of computing power. There's a reason that virtually every computer and every game console has a dedicated GPU. For that matter, a dedicated sound processing chip. It's all offloaded and the APIs have improved to the point that it doesn't seem like much work, but those specialized chips are burning an awful lot of power.
For a wide variety of games, the game logic just isn't that complicated, or rather, it doesn't require as much computing horsepower as the rendering. Sports games and FPS are the most obvious but I'm sure there's others. The most CPU intensive game I can think of is Civilization 4. I'm sure it's been surpassed, and yeah the AI still sucks, but late in games you can really tell that the CPU is chugging away.
The truth, of course, is that something will ALWAYS be a bottleneck. The argument seems to be: is it the CPU or GPU?
Yes... the reason why you notice a difference is because those additional frames *DO* have additional information in them. Your eyes, however, don't react quickly enough to discern the difference, and will blur images that occur too close to each other together. A theoretically slower frame, containing a proper blending of faster frames, would be indistinguishable from the faster framerate, as long as it is simply faster the threshold of vision persistence (approximately 1/20 to 1/30 of a second).
File under 'M' for 'Manic ranting'
No, not really.
I assume you are referring to the fact that when you look at the sky the game engine culls (skips rendering) most of the objects in the scene, therefore the GPU has less to do and if you are not CPU bound the frame rate shoots up. However when you are not looking at the sky BOTH the CPU and GPU load increases and your test does not reveal which has now become the bottleneck.
Your test only confirms the obvious: that it takes less resources (CPU and GPU) to render the sky than a full scene.
This also shows what many of us have been saying which is Bulldozer is AMD's Netburst.
Yes but not for the reason you think. Netburst introduced two things:
- An extremely deep pipeline, which was a stupid idea and ultimately netburst's demise and core's reboot from the ashes of pentium3. That's the thing most people are referring to when comparing both chips.
- HyperThreading. the ability to run 2 threads on the same pipeline (in order to keep the extremely long pipeline full). That's what's similar to buldozer's problems.
When HT was introduced, its impact on running windows software was catastrophic. That is simply due to the fact that Windows was optimized for SMP (Symmetric Multi Processors) where all CPU are more or less equal. Hyperthreadinng is far from symetric: it introduces 2 virtual processor which must share resource with the real one. You have to properly schedule threads so that no real cpu is idle while a virtual core is strugling. And you have to intelligently schedule threads to minimize cache misses. Windows simply wasn't designed for such architecture and definitely sucked at correctly juggling with the threads and the virtual processors. Proper Windows support came much later (and nowadays enabling hyperthreading under windows doesn't come as much a performance hit).
The "half core" of bulldozer are in the same situation. It's also a weird architecture (although less is shared between half-cores). It requires correctly assigning thread to processors, etc. Again current Windows ( 7 ) sucks at this, you'll have to wait for Windows 8 to see an OS properly optimized with this situation. Until then, the half-core design will come with a huge performance cost.
But that's only in Microsoft world.
On Linux the situation is different. Beside the Linux kernel being much more efficient for thread and process scheduling, Linux has another advantage: opensource code coupled with shorter release cycle. And thus the latest kernels available already support the special core model of bulldozer.
The end result is that bulldozers run much more efficiently under Linux than under Windows (as can be assert from the Linux benchmarks on Phoronix).
And they have decent performance per dollar.
Lets just hope that recent hire of the former Apple chip designer to AMD can right the ship, because otherwise when I can't score X4s and X6s anymore i'll have no choice but to go Intel.
What you'll benefit the most is waiting for a version of windows which does support the bulldozer model.
Although the bulldozer have some short-comings, which are in the process of being ironed out.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Speak for yourself, 30fps may be enough for you. It's not for many of us. We can notice the difference between 30fps, 60 fps, 85 fps and 120Hz.
We also have eyes that track and focus and thus reduce the blurriness of moving objects that we are looking at. For us, when we look at a fast moving object, the object is sharp and in focus, the background might be blurred. But if we look at the background the background is sharp and the object is blurred.
Whereas if you blur stuff before displaying it, there's no way your eyes can make it sharp even if you look at it.
The solution to more realistic graphics is to go 120Hz or more and let the limitations of human eyes do the blurring.
The other thing that many people miss is latency (which is related to the topic). A high FPS does not mean low latency. In theory you could have a crappy video system which produces a high average FPS but with a 1 second delay between what happens and what appears on the screen. That would be acceptable for movies with some audio delay corrections, but not acceptable for most games and interactive stuff. In fact there was a time where many displays had very bad latency (not one second but still too high).
*there* is the parallel.
There is parallel in the way people perceive them.
There is a big difference under the hood in practice.
I mean people see both and say "Bulldozer is the new Netburst" just as "Windows 8 is the Windoes Vista is the new WindowsME".
But the reasons behind are fundamentally different.
Netburst sucked and was hopeless. Bulldozer is suboptimal but there's room for improvement.
Intel went out chasing high numbers, what they got, was a chip that clocked moderately highly, but performed like ass anyway, and sucked power.
They got it, because they choose a design path which has many drawbacks, they sacrificed a lot just for the sake of higher GHz, Netburst doesn't bring much interesting thing to the table. I could maybe somewhat work a little bit today using the latest shrinking technologie, advanced cooling, and finally hit the 10GHz where the architecture should be competitive. While still sucking a lot of power.
But back in the Pentium IV days, there were no hope that anything could actually efficiently use it.
It "performed like ass" almost by design. Because all the things they neglected end up biting them in the long run, and become hard limits.
The only way to do something better was scrap the whole thing, move to something simpler, and stop favouring GHz at all cost, preferring it above anything else including power consumption.
Which they did. The Core family was done by improving over the older Pentium IIIs.
And they did it again in a way with the Atom family, which is not completely unlike the even simpler and older Pentium, giving an even lower power end result (though its difficult to compete with ARM in this range...)
The only solution to get Intel out of their solution was a garbage bin.
The only useful stuff which came out of the Netburst architecture was HyperThreading. Which was useless back in the Pentium IV era for lack of proper OS support. But worked better when it was reintroduced later in the Core era, just because Windows had some time to mature.
AMD went out chasing core count, what they got, was a chip that can't hold its own against chips with half as many "cores", and sucks power.
On the other hand bulldozers are limited by things which are improvable in the near future.
Some might be design flaws on the silicon, but these are stuff which can be fixed. And that means in the near future, not counting on some advanced technology 10 years from now to dramatically shrink the process. Part of the "sucks power" problem is fixable in hardware.
(And part of it is fixable by litteral "building architecture". AMD is a little bit late using older processes, simply for lacking manufacturing plants with the latest technology like intel).
But most problem aren't even hardware, but software.
- The OS and Kernel scheduler need to support its peculiar concept of half cores. There's a dramatic difference *already today* in using Bulldozer between Windows and Linux. Because current generation of kernel inside Windows 7 predates Bulldozer's release. Whereas Linux is not only fucking much more efficient, but support for half core was added long ago.
- The software needs to be written to take advantage of Bulldozer, specially using more cores. But *that is* the current general tendency anyway:toward multiprocessing, and multithreading. so that will happen naturally over time. Just look at the Google's Chrome: Each tab is (for security and sandboxing reasons) a separate (isolated) process. It's the most visible and known example, but other software follow the same trend. Being of Unix heritage, Linux uses multiprocessing much more heavily and thus has much more use cases where Bulldozer is useful (server tasks is one example).
(Also in the opensource world Bulldozer's other advantages are usually only a compiler switch- or a tool library upgrade- away. Software can take advantage of that rather quickly)
So yeah, a
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]