Will AMD cpus become better for gaming than intel with direct x12

Page 5 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
What I'm wondering is this. They switch up to dx12, mantle what have you where the api is no longer thread bound as it was before. This would account for much more thread usage and an 8c/16t cpu would appear maxed out in full communication with the gpu. However the api is only one aspect of the game, the rest of the game engine is still running on the cpu and may or may not depend on many threads same as before. It seems many people are getting confused, that if the api allows heavy multithreading the game itself is now magically multithreaded which may not be the case at all. Another reason why sites discussing dx12 said that while in theory it could bring about all this extra performance, don't suddenly expect ridiculous gains of 80% or more in games all of the sudden. That draw calls are only a portion of the entirety of what's going on. The ashes testing so far seems to confirm that. Under dx12 there are higher fps as expected, better usage of the gpu which is what dx12 set out to do. It would stand to reason that the faster/stronger cpu's with higher ipc still run the actual game code faster than those with weaker ipc which is why we're not seeing amd's suddenly matching/surpassing intel's cpu's.
 
If the developer gets results that differ from sites that were provided with an identical copy of the game, then he needs to explain why he gets different results using the same hardware. Until then we should dismiss his statement. In other words, if he states that using his hardware configuration with Windows 10 provides excellent results, then anyone else, including you and I, should be able to reproduce the same results within the expected margin of error. http://www.oxidegames.com/2015/08/16/the-birth-of-a-new-api/ Nowhere does he refer to AMD vs Intel processors.
 
Well. We have more evidence of it needing work for FX-8 CPUs. Test from someone on the Overclock.net forums;

FX-43xx@4.6Ghz
==Sub Mark Heavy Batch ==================================
Total Time: 57.973419
Avg Framerate : 37.767700 ms (26.477652 FPS)
Weighted Framerate : 38.071899 ms (26.266090 FPS)
CPU frame rate (estimated framerate if not GPU bound): 37.571934 ms (26.615612 FPS)
Percent GPU Bound: 6.452394%
Driver throughput (Batches per ms): 3170.016602
Average Batches per frame: 34691.609375


FX-83xx@4.6Ghz
==Sub Mark Heavy Batch ==================================
Total Time: 57.954021
Avg Framerate : 30.121634 ms (33.198730 FPS)
Weighted Framerate : 30.450636 ms (32.840034 FPS)
CPU frame rate (estimated framerate if not GPU bound): 26.878809 ms (37.204029 FPS)
Percent GPU Bound: 83.472687%
Driver throughput (Batches per ms): 4916.684082
Average Batches per frame: 36009.164063

Only a 6 fps increase going from FX-4 to FX-8. Should be closer to a least 60% even with inefficient scaling, rather than ~25%.
Word is that Ashes is using AVX2, which AMD CPUs don't support (they only support avx). I'll see what else I can find out.
 
The IPC of the 8 core CPU isn't significantly better than the 4 core; therefore the results are to be expected. You're trying to convince yourself, but you'll have a hard time convincing me that the FX-83xx is a very good gaming CPU and that probably wasn't AMD's goal when they designed it. The goal was to provide a good multi-tasking CPU at a competitive price and that goal has been achieved.
 
The IPC of the FX-8 and the FX-4 is the same.

I'm not trying to convince myself of anything. I'm just trying to figure out what's going on. Now it seems that the Phenom II X6 outperforms the FX-8 and is actually on par with an i5 4670k... Heavily overclocked, but still. And the developer stands on its point that the FX-8 should be doing better... So... Yeah. Things are confusing as of right now...

ab038afe_5EjgnZg.jpeg
 
What exact configuration did he use and what were the results? The "it should be doing better" statement implies that the posted CPU results are accurate. The game developer mentioned that Intel, AMD and NVidia had the DX12 code for up to one year. AMD obviously did something right since their GPUs perform above expectations in that DX12 game; actually they do so well that NVidia question the validity of that benchmark. Have you found a statement from AMD questioning the validity of that benchmark when using their CPU?

We all know the Phenom II X6 has a better IPC than the FX, but it isn't faster than the FX-8 at multi-tasking.
 


Again, this is EXPECTED BEHAVIOR.

Adding CPU cores only adds more performance in CPU Bound tasks. Games are dominated by GPU performance, not CPU performance, and as long as the CPU can send data to the GPU faster then the GPU can process, then adding more cores gives you no extra performance. ITS THAT SIMPLE PEOPLE.

Bottleneck is the GPU, as it has been for 20 years. Intel is faster then AMD simply because of it's superior single-threaded performance.
 


Could you be a doll and let Bohemia Interactive (ARMA 2, Arma 2 OA, Arma 3, VBS, and VBS 2), Bohemia Simulations (VBS 3) Eagle Dynamics (DCS World), Bethesda Softworks (Morrowind, Fallout 3, Oblibion, Skyrim), Lockheed Martin (MSF Rebuild) , and Frontier know that all of those imposing CPU limitations haven't existed for 20 years? Hell, those games should be running like gold on a P1!

You might want to tell the bench markers and players, while you're at it.

Thanks.

 
All the ARMA games were GPU bottlenecked day 1. Bethesda is one of the few exceptions to the rule, given how graphically light and computationally heavy their titles are, but even their stuff ends up GPU bottlenecked within a few months of release. And even then, GPU power mattered more then CPU power.
 
You can find several posts where members were recommended an AMD FX build based on the assumption that DX12 games will take advantage of the additional cores, making it a better gaming CPU than an i5. AMD's Mantle didn't make that happen; therefore DX12 can't.
 


If by "take advantage of" you mean use, then yes, DX12 will take advantage of more cores more easily then multithreaded DX11 will. If you mean "give a performance advantage", then no, it won't, for reasons I've explained over and over again.

Now, if AMD released a new CPU arch with more IPC then the i5, and if games suddenly become much more CPU bound because of lower CPU overhead allowing them to do more, and GPU development stalls, then maybe the situation changes. But for existing CPUs, FX competes with the i5 lineup, and DX12 won't change that.
 
Until I understand what's going on, I won't draw a conclusion. As for your 'explanations' gamerk316, I can tell you have no idea what you're talking about just by the language you're using.

And I never said the FX CPUs will be better than an i5. I said it will equal them. I might have said that it's a better choice due to the price/performance that it will have under DX12. But things are not making sense because both a Phenom II X6 and an Athlon 860K is beating the FX-8370, so something is going on.

It might indeed be that the FX CPUs will not have any improvement in performance, but as of now, I don't see anything that still excludes it, especially considering the Phenom II X6 performance increase. Remember that this CPU is slightly slower than the FX-8 in DX11 gaming, but now the FX-8 CPUs suddenly lose to it. Plus, the developer reporting the FX-8 not being CPU bound in his benchmark (a new build has already been released) might show something. We'll just have to wait and see.
 
I'm not sure why the difference in performance with a phenom ii x6 but the athlon 860k performing a bit better than the 8350 doesn't surprise me so much. It has better core performance than the fx 8350 even though it has fewer cores. Similar to the intel chips which have higher single core performance and outperform 8 core fx chips with only 4 cores. The api may be using all 8 cores for dx12 but it doesn't mean the game code itself is running efficiently on all 8 cores/threads. If it's reliant on fewer stronger cores, both the athlon 860k and intel cpu's have the advantage there.
 
And I never said the FX CPUs will be better than an i5. I said it will equal them. I might have said that it's a better choice due to the price/performance that it will have under DX12. But things are not making sense because both a Phenom II X6 and an Athlon 860K is beating the FX-8370, so something is going on.
Click to expand...

Both the PII X6 and Athlon have higher IPC then the FX-8370, so if the bench is largely insensitive to clock speed, that could explain what's going on. Farther hints that there really isn't any real performance benefit beyond four cores, even if the CPU load scales.

Would be interesting to see how a i3 holds up.
 
There is a fascinating discussion on AotS benchmark over on ocn, if anyone's interested I suggest they take a look there's some great posts and even an AotS dev now posting. Its just theory but the current opinion is the old 990 chipset pcie2.0 or the memory bandwidth could be why the fx8*** is not much better compared to fx4*** and 860k and behind intel with the cpu having more to feed the gpu with. What do you guys think? Gamerk?
http://www.overclock.net/t/1569897/various-ashes-of-the-singularity-dx12-benchmarks
 
Everything is simple. If you live somewhere, where electric price is cheap, go for AMD way, prices are really nice, but tdp is horrible. Amd products are cheaper, and performs about the same like i5, i7 and nvidia products. So I said, the main thing why I not go for AMD is high TDP.
 


It's not memory bandwidth; that's NA for discrete GPUs. Likewise, PCI-E 2.0 x16 isn't really stressed for single GPU configurations.

Look, it's simple people: If the CPU is giving the GPU frames faster then the GPU can output them, you aren't going to see any real performance benefit by speeding up the CPU. At that point, single-core performance is the only thing that will benefit the CPU, which is why the FX-8xxx is slightly faster then the FX-4xxx, due to the small clock speed advantage the FX-8xxx has. PII X6 is faster then FX due to its IPC edge. Intel is faster then AMD for the same reason. Done.

This goes back to what I've been saying for years: GPUs are gaming's bottleneck, so there's almost no benefit to continued threading. The only way to squeeze out additional gains is single-core IPC, which is where Intel has it's edge.

It's that simple.
 
Games are only GPU limited when assuming they are not limited by anything else.


Ashes of singularity clearly does not take advantage of all cores. It may try, and fail, trashing the threads so that when one thread is still doing stuff, the other threads have to wait. For many game developers, making an engine run on many cores, is not an easy task.
 


Insert faster GPU. Does game run faster? Yes? It's GPU limited.

RAM is not a major bottleneck due to the insane amounts of CPU cache these days. CPU effects are limited to single-core performance. PCI-E delays add latency, but don't really limit bandwidth, even on PCI-E 2.0.

For games, the GPU is almost always the primary bottleneck.

Ashes of singularity clearly does not take advantage of all cores. It may try, and fail, trashing the threads so that when one thread is still doing stuff, the other threads have to wait. For many game developers, making an engine run on many cores, is not an easy task.
Click to expand...

It uses them, and loads them all up. Beyond that, I'd need a tool like GPUView to see what's going on under the hood.

But as noted many times by myself [who happens to be a software engineer], since the CPU isn't the bottleneck, the ability to make the CPU finish its work faster doesn't impact FPS, since the CPU wasn't the bottleneck in the first place.

Now yes, making the CPU finish it's work faster allows you to do MORE within the same budget, so purely CPU processes such as AI can be expanded in DX12 without affecting FPS, independent of the GPU.

In short, you can do more work on the CPU without reducing FPS, but making the CPU faster won't improve FPS either.
 
^^ Results are what I'd expect. Intel gets GPU bottlenecked with a 4770k at 3.5GHz. By contrast, the FX never pushes the GPU to its maximum potential, even at 4.5GHz. You can see GTA loves single core performance due to the linear scaling you get as clock speed increases on the FX. One of the few examples out there of a genuine CPU bottleneck.

If AMD wants to perform the same as Intel, they need to stop worrying about adding cores, and fix their single core performance deficit.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom