Originally posted by: orangat
And the fact is: the PublicRelations department says that the FX is the ultimate gaming processer. In the link you gave us, a slower clocked X2 4800 beats a faster clocked FX 55 in Splinter Cell.I don't see any instances like that happening with a slower clocked single core beating a faster clocked dual core - ESPECIALLY an FX series.. This is why I use common sense when choosing my processer, not the PR department.
Orangat, I had a thought with respect to your comments regarding the sum of CPU utilization across both cores. I think you are not taking concurrency into account. Let's assume that no one application will be allowed to use more than a combined 100% of CPU time across both cores (i.e. core 0 50%, core 1 50%). I'm not aware of any such limitation, but as you pointed out, that is clearly happening in the three graphs where a game has been loaded and is running D3D (note that no such horizontal symmetry appeared in the Google Earth, Outlook, or DVD playback graphs). Even if this limitation is real, the two cores are still executing code in parallel that would otherwise have to be executed serially. Games are realtime simulations, and the amount of processing that can take place in a frame (17 ms, say) is a critical component of performance. Concurrency is the big win here.
Edit: I think it is also interesting to look at the top scores in the Crystalmark thread.
Edit: I think it is also interesting to look at the top scores in the Crystalmark thread.
I updated the original post with data gathered from 15 minutes of BF2 multiplayer play using the recently released nVidia 81.82 beta drivers with dual core optimizations. The difference is amazing. Check it out.
Would it be possible to reliably benchmark multiplayer gaming? I've heard X2 owners claim noticeable performance increases vs. single cores in crowded on-line environments.
Well, depends on what benchmarks you want to use. I only captured processor core utilization here, and played BF2 for 15 mins online to see what the results would be. Perfmon is capable of measuring a crapload of stuff.
I'm sorry to say, but one major flaw with all this analysis is the assumption that Perfmon.exe is acurately reporting CPU core utilization. It has been concluded in these forums (and elsewhere I expect) that it does not. If I recall correctly, unless a task has affinity set properly, perfmon will not properly report core utilization. If that is indeed the case, much of this testing is inconclusive.
Don't tasks with incorrect affinity properties perform poorly, unstably, or not at all? The OP didn't report any such conditions.
I guess i am a bit baffled by this thread only because SMP has always functioned as the origional thread starter has demonstrated.It is no scecrete 2 cores or 2 CPU load balance and that no one single app can use more then 50% of each core.100% of anything is all you can ever get from any app no matter if it is run dual core or single core . .
To answer the point about perfmon, I have been curious both about its accuracy in reporting CPU utilization, and its own effect on the numbers given that it is capturing a sample per second. I may try to quantify this by capturing perfmon's own cpu usage and subtracting.
With respect to the comment that no app can use more than 50% of a core, I don't see any such limitation when I run BF2 under the dual core-optimized nVidia drivers. It did appear that way in the previous runs, but I am not sure whether this is an artifact. As for it being obvious that dual cores load balance, I agree, but there remains a fair bit of debate here about whether dual core's have a positive effect on software like games. Thus this thread.
With respect to the comment that no app can use more than 50% of a core, I don't see any such limitation when I run BF2 under the dual core-optimized nVidia drivers. It did appear that way in the previous runs, but I am not sure whether this is an artifact. As for it being obvious that dual cores load balance, I agree, but there remains a fair bit of debate here about whether dual core's have a positive effect on software like games. Thus this thread.
Mark, I'm still unsure what you mean by the concurrency advantage. And from your previous statements (from other thread), do you mean to say that the X2 feels more fluid/playable even if it has similar fps scores in games compared to single cores?
And I don't see cpu utilizations being the end-objective of benchmarks. What you should be doing is benching the fps. If Nvidia drivers could offload some of the triangle setup to the 2nd core, it would be huge.
And I think the X2 benches slightly faster in games at the same clockspeed compared to single core is because of the extra L2 cache since both cores come with their own cache.
The 4200+ has 512+512 while the 4800 has 1M+1M which is leaps and bounds over single cores.
The 4400+ also has 1m + 1m, but you have to bear in mind that this is only a benefit for each core individually. So the "system" never sees the benefit of 2m cache. What it sees is the benefit of 2 processors, each with 1m cache.
What I mean by concurrency is simply the advantage of parallel processing. The two cores can execute instructions simultaneously. A single core can only execute them serially (not precisely true, since there is some parallelism built into modern CPUs, but effectively true). So if we assume that 100% of some "ideal" CPU is all any program can have access to, it is still better to have 50% of your work done one one core, and the other 50% done in parallel on the other core.
In practice it appears that there is no "100% limitation." My most recent tests of BF2 with the new nVidia beta drivers show the app often getting %140 - 160% of a single core by heavily utilizing both. Unless there is something wrong with the test methodology this should put the argument to bed. The previous "reflection graph" as you called it appears to have been the result of the way that the unoptimized nVidia release drivers worked.
What I mean by concurrency is simply the advantage of parallel processing. The two cores can execute instructions simultaneously. A single core can only execute them serially (not precisely true, since there is some parallelism built into modern CPUs, but effectively true). So if we assume that 100% of some "ideal" CPU is all any program can have access to, it is still better to have 50% of your work done one one core, and the other 50% done in parallel on the other core.
In practice it appears that there is no "100% limitation." My most recent tests of BF2 with the new nVidia beta drivers show the app often getting %140 - 160% of a single core by heavily utilizing both. Unless there is something wrong with the test methodology this should put the argument to bed. The previous "reflection graph" as you called it appears to have been the result of the way that the unoptimized nVidia release drivers worked.
I suppose you're right. I'm just trying to make sense of the benches in games.
But if it doesn't actually increase performance because cpu util is always capped at 100%, is it even relevant? I'm trying to understand why you said fps didn't matter as much.
Great, so games are already running faster under X2. Some actual fps numbers for comparison would be awesome.
Well, I didn't say that FPS didn't matter as much. What I said was that it isn't the only thing that matters. Frames per second measures how fast the system can construct and render a frame of video. In a single threaded loop architecture (most current games) anything that slows down that loop will slow down the frames per second. But it's not quite that simple. Windows uses multiple threads in the 3d graphics and sound subsystems, and in the I/O layer, as examples. So you can be chugging along rendering 75 fps and have network updates lag, causing sprites to jump around on screen. You've still got high performance graphics, but you don't have an overall high performance game at that point. There are lots of other potential examples. These conditions will become more common as more games are designed for concurrent processing.
On the cpu question, think about it like this: at 50% utilization a given core is executing instructions just about as fast as it possibly can, because there are no bottlenecks; its utilization is way below max. Now imagine a unit of work. You can either hand the whole unit to a single processor, or divide it in half and hand each half to a seperate processor. If you do the latter then that unit of work will complete in roughly half the time. I think where you are getting confused is in thinking that 50% core utilization equates to 50% speed. It doesn't. The cpu is still chewing through instructions as fast as it can. It's just that it could chew through twice as many in the same amount of time, if the rest of the system could keep up.
On the cpu question, think about it like this: at 50% utilization a given core is executing instructions just about as fast as it possibly can, because there are no bottlenecks; its utilization is way below max. Now imagine a unit of work. You can either hand the whole unit to a single processor, or divide it in half and hand each half to a seperate processor. If you do the latter then that unit of work will complete in roughly half the time. I think where you are getting confused is in thinking that 50% core utilization equates to 50% speed. It doesn't. The cpu is still chewing through instructions as fast as it can. It's just that it could chew through twice as many in the same amount of time, if the rest of the system could keep up.
If the cpu is occupied by some other spurious process other than the game, you wouldn't get an fps of 75 in the first place.
And if network updates happen or the virusscan starts churning away, an X2 will help but it won't help disk contention at all and benchmarks will still plunge.
I think you'd better get some fps benchmarks asap if you think your bf2/cod are running twice as fast compared to a single core cpu. I think you'll be surprised.
I think there is a utility to disable one core.
I think you missed my first point. I didn't say that the cpu was occupied by "some spurious process" (what does one of those look like), but rather that the CPU and graphics engine can be rendering frames at full speed while an i/o thread is idling waiting for a late network update. The point is that once you decouple that central loop you get many more different kinds of performance scenarios.
On the second point, it is easy to disable one core for a given process by setting its affinity, but again my point got lost in translation here. Even if the dual core can perform the CPU intensive parts of the game code twice as fast (and I am willing to bet it can come close to that in the right situation) that doesn't mean you will get twice the fps. The same factors you mentioned in your first para, i.e. disk contention, graphics bottlenecks, etc., apply.
On the second point, it is easy to disable one core for a given process by setting its affinity, but again my point got lost in translation here. Even if the dual core can perform the CPU intensive parts of the game code twice as fast (and I am willing to bet it can come close to that in the right situation) that doesn't mean you will get twice the fps. The same factors you mentioned in your first para, i.e. disk contention, graphics bottlenecks, etc., apply.
You made a bold claim that performance must have improved drastically because the work completed at half the time, why don't you bench the fps and see? How did the same factors like disk contention and other bottlenecks suddenly not affect single core measurement the same way?
The best objective benchmark is still fps, not cpu util. I'm surprised you aren't benchmarking fps after the big change after the Nvidia driver update.
The best objective benchmark is still fps, not cpu util. I'm surprised you aren't benchmarking fps after the big change after the Nvidia driver update.
[You made a bold claim that performance must have improved drastically because the work completed at half the time]
Haha, I said nothing of the kind. I was speaking only of CPU throughput. If that sums up your definition of performance then we're probably wasting time here.
Haha, I said nothing of the kind. I was speaking only of CPU throughput. If that sums up your definition of performance then we're probably wasting time here.
Online benchmarking isn't reliable.. you can't really compare one session to the next with accuracy. I have FRAPS and have benchmarked BF2 online, and it's different everytime. That being said though, games tested with the new drivers have shown FPS improvements, at least in that link from the thread a couple weeks ago about the new drivers. I also have seen a few people comment about fps going up in games for them, but I have no idea if they ran official benchmarks, or were basing that on subjective analysis.
It seems you keep searching for some way to convince those of us with dual core that the advantages are all in our head or something.
I don't know if you can benchmark "smoothness", but anyone with a dual core can tell you that is a major benefit of them. You can argue till your blue in the face, but you don't have a dual core and you don't really know until you try it. Looking at benchmarks on review sites isn't gonna show you how smooth dual cores are in REAL gaming- like online with various apps running in the background. Benchmarks are in perfect conditions on clean installs.. and have nothing to do with real-world computing. You know- like having anti-virus, anti-spyware, AIM, TeamSpeak, VOIP, IE, TCP/IP comunications with the 64 person server, Firewall, blah blah blah.. I don't see any benchmarks showing how single cores do when the CPU has to divide it's time to take care of that stuff, also. Basically, dual core simply enables any applications to have access to CPU cycles without having to wait for another task.
You keep trying to convince everyone (and yourself) that dual cores don't have any worthwile advantages- but they really do!
ElTorrente, I never said X2 don't have any worthwhile advantages. You making a strawman argument again.
Apart from games X2's are great cpus and at speeds above 4200+ X2s are start to become faster than single cores.
Apart from games X2's are great cpus and at speeds above 4200+ X2s are start to become faster than single cores.
What I am wondering now if cpu cores are now being utilized at a much higher cpu usage yet the games fps did not increase much...Is the video card the bottleneck??? I dont see why utilizatioon would be that high with no improvemen. i think I would rather have lower usage if performance will not increase...
How about someone with an SLI 7800 system run it and see if the differences are bigger and noticeable...
How about someone with an SLI 7800 system run it and see if the differences are bigger and noticeable...
[I don't know if you can benchmark "smoothness"]
I don't know if you can benchmark it, but you can show how the cpu's are being used, and anyone who understands computers can then infer it.
[What I am wondering now if cpu cores are now being utilized at a much higher cpu usage yet the games fps did not increase much...Is the video card the bottleneck??? I dont see why utilizatioon would be that high with no improvemen. i think I would rather have lower usage if performance will not increase... ]
Well in my case Duvie I run with vsynch enabled, because at 1600 x 1200 I get significant tearing if I don't. So BF2 won't run faster than 60 frames per second no matter how fast my machine is. Assuming that, unbridled, it would run at 75-80 fps, then that is a lot of processing power left over that can go into sound, physics, network i/o, etc.
I don't know if you can benchmark it, but you can show how the cpu's are being used, and anyone who understands computers can then infer it
. [What I am wondering now if cpu cores are now being utilized at a much higher cpu usage yet the games fps did not increase much...Is the video card the bottleneck??? I dont see why utilizatioon would be that high with no improvemen. i think I would rather have lower usage if performance will not increase... ]
Well in my case Duvie I run with vsynch enabled, because at 1600 x 1200 I get significant tearing if I don't. So BF2 won't run faster than 60 frames per second no matter how fast my machine is. Assuming that, unbridled, it would run at 75-80 fps, then that is a lot of processing power left over that can go into sound, physics, network i/o, etc.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter