AMD CPU speculation... and expert conjecture

Page 384 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.


According to last Steam survey Intel and NVIDIA have 67% of the GPU PC market. But according to last prospects AMD will gain significant market share thanks to new Radeon cards. Also don't forget that AMD has about 100% of the new consoles market.
 


Because with Mantle GPU was bottleneck. They could replace i7 with i5 or maybe even with i3 and result would be the same.

Without Mantle we have similar situation. Most modern games (except BF3, C3 and few other titles used in hardware reviews) runs well on Pentium/Athlon. Still there are some games which are CPU-bound (Assasin's Creed, GTA, basically most sandbox and RTS games).
 


MANTLE removes CPU bottleneck and future games will look as in the famous slide #13 of AMD October talk, where an 4C APU gave practically the same framerate than FX-6350/8350. Where is the surprise folks? I have said this plenty of times.

But, at the same time, MANTLE improves multi-core support: a 4C Kaveri will perform better than i3.

It is a case per case problem. In some games 4C Kaveri will perform as an SB i5 or poor, in other games 4C Kaveri will perform as an i7-4770k, and in some games it will perform better... it depends. Etayorius wrote an excellent and succinct summary.
 
It's hard to trust benchmarks, so I wrote my own 🙂

I've written a Java app to compare CPUs and GPUs, and I have some results for a stock i7-2600K and an 8350 overclocked to 4.4GHz. When using Java, it is a pretty convincing win for the 8350. I conclude that the Java just-in-time compiler was not written by Intel!

http://www.headline-benchmark.com/results/b5b4278b-300e-47be-9698-bddda9a21d0f/79361afa-c8b8-47ac-af0c-2abe5090e7b0

http://www.headline-benchmark.com/cpu/5646722353070080/5761569778565120

My results would be improved by more data points, I do admit. You can always go to the download page on the site and contribute some more data if you like 🙂 Be aware that it is a work-in-progress and has bugs (e.g. crashes on GTX 770). Also, you have to have Java installed.
 


Mantle doesn't ALWAYS remove CPU bottleneck, just decreases rendering overhead. Game's logic could bottleneck CPU also.

BTW, Richland without Mantle was sometimes better in gaming than i3 :>


And in some game Pentium perform as i7-4770K (if GPU is bottleneck), so what's the point? Only cases where CPU is bottleneck are good to measure pure gaming performance of CPU.
 


Yeah, Intel has a lot of GPU market share and so does Nvidia, but you need to look at it in gaming rigs. The total computer market isn't interested in playing the latest AAA titles. When it comes to gaming AMD has about 33% of the market according to Steam Hardware Survey. That's already significantly better penetration than you're suggesting with 20% for AMD overall.

I think you miss the point of Mantle. Mantle is a part of the drivers, not an alternate framework. There is no reason why MS would add a Mantle like layer to DX and then have it make it impossible to use on Linux. For that to happen AMD would have to drop Linux Mantle support.

But we have been over this already. Market share does no matter. Game developers want out of Windows ecosystem and they want it so bad they're pushing for Linux gaming, which has a <2% desktop market share depending on where you look.

Mantle is a tool to help game developers leave Windows ecosystem. Having an easy port to Mantle on Linux and then encouraging people to switch by making Linux bonus packs or something would work wonders.

It would be a lot better to try and get people with GCN cards to switch to Linux as opposed to just trying to tap the Linux market. I do feel like that's what companies are aiming for. Note how Valve keeps talking up Linux performance improvements and then they launch SteamOS. They want people to switch because they want to not depend on Windows anymore after Windows 8 failure and seeing how MS has no qualms with throwing people under the bus and making significant changes to the interface.

And, Mantle is not comparable to PhysX in any way. PhysX was additional work for eye candy that solved no problems. You can see that it's just meaningless eye candy by the way that Nvidia had to pay companies to make PhysX games and there was only one good one every year or two while people are jumping on Mantle left and right.

You're also forgetting that Mantle opens up a new market of mobile devices and low end rigs where owners of said hardware wouldn't be playing games at all because they can't afford the hardware and now they can. Mantle is literally bringing gaming to people who couldn't game before because they didn't have a strong enough device.

You can sit here and say that Intel and Nvidia have 80% of the total market but practically every Intel GPU is unable to play games. Nvidia mostly has dGPUs in gaming rigs. AMD can swoop in and take mobile gaming on 35w and lower SoCs and make it tolerable. The thing you are missing is that Mantle would make it possible to play a decent game on Temash and Kabini, something that's out of reach right now.
 


Mantle as you would believe with marketing will make it so an Intel Atom can run a game at maximum resolution. What will this accomplish as far as AMD sales go? MANTLE has nothing to do with cpu processing power at all other than making it seem that even a low performing cpu can accomplish the same task as a HPC. Guess what happens when you don't run Mantle or run Nvidia hardware? its just left with the Low Performance CPU. How many "GPU bound games" are deemed to be the best metric for measuring CPU performance?

Mantle will really only accomplish 1 thing for AMD. It will boost sales for radeon cards. It will boost sales for the i3 and lower cpus because Intel will leverage it to their advantage. Mantle will not magically make Kaveri faster than Hasbeen as it has nothing to do with the cpu.

Nobody said that HSA is on/off. That is why I separated the case of MANTLE alone from the more advanced case of MANTLE+HSA.

The first MANTLE games will not use HSA. With time, games will use MANTLE and HSA. HSA is included in next consoles. We already saw demos of GPGPU computations in consoles: e.g. physics running in PS4 GPU.

There is also a chance that with time Mantle and HSA will be ignored by programmers. What does AMD have left if that happens? There is no guarantee that Mantle or HSA will be AMD's saving grace instead of another disaster.
 


http://www.headline-benchmark.com/results/083b462f-9ef2-46f6-b1bb-a3c83c1f9cce/b5b4278b-300e-47be-9698-bddda9a21d0f

There's mine. I'll try and grab the code to see if there's anything worthy to be improved. And my 2700K at 4.6Ghz looks quite strong in almost everything.

Cheers!
 


Optimizations on/off? No locking issues eating performance? HTT/Cache impacts? I'd LOVE to have a look at the source code, so I can pick away at it.
 
http://www.headline-benchmark.com/results/41a7eed7-7dbf-45d3-8d03-d0ef407b00f6

Two quick questions, though maybe my mind isn't working right now (130 AM right now):

1: How come the pixel shader test was the only one done in a fullscreen window? I ask, because my AV went to game mode during that time, so the other results were probably suppressed a little...(Will attempt to pass Yuka in the GPU tests tomorrow 😛)

2: Three cores were doing a decent amount of work during the memory tests, and was a "spiky" pattern; not sure what was going on there...(Programmer in me trying to figure out what's going on under the hood; i do that)

3: I used the Win32 version; the x64 one just "goes away" after I click OK. Will debug a bit more tomorrow (Java issue?)
 
gamerk:

1. the first two gpu tests try to stress textures and geometry, so they use a lower pixel count to exclude any limiting related to rasterization. The shader test uses full screen because it scales by pixel - using full screen lets the tests finish a few seconds sooner. (otherwise I would have to do more iterations of the test)

2. The memory test is single threaded, but possibly the work was being evicted from core to core. if I could set core affinity in Java I would do so.

3. My guess: you have Java 32-bit runtime - do "java -version" in a CMD prompt and if it does not mention "64-bit" then you have the 32-bit version.

Thanks guys for running the tests!
 


I didn't write "ALWAYS". You did. It has been explained here for days that MANTLE removes DX/OGL API overhead. Draw-calls were explicitly mentioned and discussed.

I know quad-core Richland can outperform an i3 when four cores are used. I based my BSN* article Kaveri scores in Trinity/Richland ones.

I know that a Pentium can perform as an i7-4770k (if GPU is bottleneck). Some GPU-bound benchmarks were given in this thread and reposted often.

If you are going to play games at 720p and low settings, go for "CPU benchmarks". If I am playing at 1080p or higher and medium or high settings, then those "CPU benchmarks" are useless...
 
Interesting thread. It seems that bottom line on whether there will be a Steamroller based FX-series CPU isn't at all confirmed by AMD, right? The only thing we do have, is a roadmap that indicates that there won't be a Steamroller FX in 2014, but that is not the same as a statement from AMD on this matter. These roadmaps are as they see the world right now and as they also write in the first slides of the roadmap, it is subject to immediate change if their view of the world changes.
 
1. the first two gpu tests try to stress textures and geometry, so they use a lower pixel count to exclude any limiting related to rasterization. The shader test uses full screen because it scales by pixel - using full screen lets the tests finish a few seconds sooner. (otherwise I would have to do more iterations of the test)

Its 830 now and that makes a lot more sense. Might be worth noting AV might come into play for the non-full screen tests.

2. The memory test is single threaded, but possibly the work was being evicted from core to core. if I could set core affinity in Java I would do so.

You can probably invoke the Windows API manually and do it that way, but you'd preempt the Java threading API's and probably have a bit of re-writing to do. And you'd break linux compatibility in the process. Welcome to the world of fine-tuning ports. 😀

In any case, the thread was jumping a LOT between three cores, one of them a HTT core. Most of other tests were more stable in this regard.

Probably not a huge factor in the final performance, given it IS a memory test, but worth mentioning non the less.

3. My guess: you have Java 32-bit runtime - do "java -version" in a CMD prompt and if it does not mention "64-bit" then you have the 32-bit version.

I'm 99.9% sure that's it too, just didn't have time to check. I'll make sure tomorrow, but that's probably it. Still, might want to throw SOME error message if this happens. Nothing frustrates a user (and tech support) more then "Hi, I was running your application, and it went away by itself. Can you fix it?" support queries.

(I actually get yelled at a LOT at work for throwing too many error messages, but I'm a firm believer that every exception should be handled, at the very least, with the name of the function that crashed, what the error actually was, and the relevant data for the exception, so I can debug the thing. If I can provide a stack trace, even better.)
 
I do catch most exceptions and popup a 'submit error report' window (which sends me the stack trace), but I don't catch this one clearly.

AV should not affect the results - I test the GPU tests at different CPU clock speeds (eg 40% underclock) and the results have to not vary. So there is headroom for a bit of background AV.
 


Mantle is primarily a marketing thing. They get game devs to talk about AMD for free or very low cost. It gets some buzz in the industry. I doubt Intel would ever mention Mantle.

Kaveri really just needs to beat an i3, or whatever product Intel has in the $150 price range. The clock speeds aren't bad considering the process change. (3.7Ghz Base/4Ghz Turbo).
 


The latest roadmap leak shows no Steamroller FX likely ever, as it goes through 2015 with Piledriver. In 2016 they would likely jump to Excavator FX, but probably only if they had a large buyer lined up (like Verizon).

Intel only ships ~2% of their parts as 8+ core so considering AMDs strained resources it makes sense for them to stick with more mainstream APUs. Perhaps the mysterious 6 core APU will appear at the 20nm node.
 
http://www.techspot.com/news/54821-ea-sports-cancels-fifa-manager-series-after-13-years.html

FIFA Manager heavily leverages off the FIFA engine. So the fact the series is canceled because it is "too expensive to bring into the next generation" kinda leans more evidence toward my argument that ports aren't going to be fundamentally easier. Then again, this shouldn't be shocking, given how games like FIFA 14 and NBA 2k14 for the PC are running on current-gen, and not the theoretically easier to port next-gen versions.
 
Not sure if this was posted or not:

http://techreport.com/news/25707/all-signs-point-to-kaveri-being-an-evolutionary-upgrade

To begin with, Adam Kozak, AMD's marketing chief for client processors, told folks during a press briefing that Kaveri will be competitive with Intel's Core i5-4670K processor. (That's a $225 offering and the cheapest quad-core Haswell CPU with an unlocked upper multiplier.) When pressed for details after the briefing, however, Kozak clarified that Kaveri should only be equivalent in terms of combined CPU and GPU compute power. If one measures x86 performance on its own, Kozak said, "we'll lose." However, Kozak expects Kaveri's integrated graphics, bolstered with Mantle support, to be better than the latest version of Intel's HD Graphics.

Emphasis mine.

So, about what we've all been saying; without the GPU and HSA enabled software, i5s win. Even AMD's marketing department said so!
 
^^ that's old news for a thread that moves this fast. it's so old that it has been used to make false claims and misinforming already!
anywho, the leaked promo slide in the last para of article is for mobile (laptop) version of kaveri, so power, performance, clockrate scaling are involved.
meanwhile, haswell i5 4670k is so overpriced that undercutting it is a piece of cake for an entry level apu like kaveri. as long as amd prices kaveri sanely like they did with previous ones, it'll be good.
 


Juan sais otherwise and Juan is never wrong, just ask him.
 



Or after 14 years people are just tired of playing these types of games. I know I haven't played one in over a decade.
 
I'm sticking to a 15% improvement on average for CPU performance compared to the A10 6800K. 4.2Ghz downclocked to 3.7Ghz is a 13% reduction in clock speed and at most Amd only improved per clock performance by 25-30%. Its GPU however will be massively bottlenecked by its memory controller and DDR3 memory more so this time around. However you can't avoid the fact that the APU was playing BF4 at 1080P at 40FPS. Now this is playing at Low-Medium on the A10 6800K http://www.youtube.com/watch?v=21odBrgz-q8

I will say the GPU is probably around 25% more powerful which is great since it sticks to the same power envelop and probably within the same price.

APU Kaveri
CPU 15% Improvement on average
GPU 25% Improvement on average
Again this is without Mantle or HSA enabled apps.
 


Not even close.



There is no Excavator FX, just as there is no Steamroller FX. If you refer to Carrizo neither it is a 6-core APU nor it is made at 20nm.



That link was posted at least twice before you. It was also discussed. He mentions an i5-4670k and, sorry to disappoint you, but as has been said here for months, the Kaveri CPU is at the i5-2500k level (in integer performance). The i5-4670k is about 20% faster than i5-2500k.

Your "i5s win" is an exaggeration.
 
Status
Not open for further replies.