AMD CPU speculation... and expert conjecture

[gamerk316]

Once again. Look at the FPS for the OC FX-4300 and the FX-8350, both are 30 FPS. There is no bottleneck, because OC the 8350 gives an improvement on the FPS that coincides with the OC.

First off, the FX-4300 gets 22 FPS, and 29 when OCd to 4.7GHz. The FX-8350 gets 30 FPS at stock, 33 when OC'd to 4.7GHz.

The FX-4300 is clearly bottlenecked, and only OCing to around 4.7GHz alleviates this. Even then, performance is lagging behind the OC'd FX-6300 and FX-8350, so I'm guessing at least one core is still overloaded, even with a 900MHz OC.

The FX-6300 also shows signs it is near a CPU bottleneck, though that could be due to its slower clock of 3.5GHz. When OC'd, it performs identical to the OC'd FX-8350 at the same clock (as one would expect).

Based on the FX results, I have to conclude the FX-4300 is CPU bottlenecked, and only a heavy OC will overcome it for this benchmark. The FX-6300 starts off bottlenecked, likely due to its slower base clock, but the bottleneck goes away with an OC. The FX-8350 is the only AMD CPU not bottlenecked for this particular Crysis 3 benchmark.

Makes one wonder how the i3 would fare with the same 4.7GHz OC...Faster then the FX-8350 prehaps?

The same about the i5 and i7. Both obtain the same FPS and OC both obtain more FPS.

Because neither is bottlenecked to start; performance increases will be dominated by per-core performance (clock). Nevermind the minimal gains of HTT; you can generally assume the i5 and i7 should match eachtother within a few FPS at the same clock (which they do).

Throw in the minimal FPS gain for a 1GHz OC (3 FPS), and I have to assume there's a GPU bottleneck occurring at 45FPS, so IB performance is probably greater then what the results show.

This is a clear symptom that "Roof" is not threaded enough. That is why the i3 perform so well. Whereas "Welcome" is heavily threaded, and the i7 is much faster than the i5 when HT is activated and the FX-8350 is much faster than the FX-4300.

In "Welcome", the FX-4300 @ stock is loaded at 95%. It is the dual cores, which are loaded at 100%.

And the dual core i3 outperforms the FX-4300. The FX-6300 does far better, but is then outperformed by the quad core i5. Seeing the recurring trend here? Intels duo outperforms the AMD quad. Intels Quad outperforms the AMD Sexta.

The reason for that is simple: Both the FX-4300 and i3-3220 are CPU bottlenecked. Between them however, the superior IPC of the i3 is enough to outperform the FX-4300's extra 500MHz and two extra cores. The FX-6300 however, is NOT CPU bottlenecked, hence the massive jump (9 FPS) even at a lower clock then the FX-4300 (300MHz slower).

Compare the FX-6300 against a quad core i5, however, and the trend starts to repeat itself, where the extra IPC of the i5 outperforms the two extra cores of the FX-6300. (The FX-8350's spot is likely due more to its clock rather then its number of cores; would need to clock the FX-8350 to 3.5GHz to confirm though.)

But its worth noting: In two cases, subtracting two cores but adding to IPC improves performance, even though those extra cores are being used. Which is what I'd expect. The i3 is running 9 threads on two cores, but due to its higher IPC, it can chug through those threads at a faster rate. By contrast, the FX-4300 can do twice as much work at a time, but each core is slower, so each thread takes longer to finish. Same trend repeats for the FX-6300 and the i5 lineup. Thats AMD's IPC problem going forward.

My point being, since its unlikely AMD will be able to increase clocks much farther, unless it fixes its IPC problem [lets see what SR does on that front], its chips will continue to perform behind Intels in games. No amount of threading is going to fix that problem. So when everyone on the forum repeats the BD mantra of "Buy PD; its performance will pass Intel's early next year", I take issue.

 

[Cazalan]

My point being, since its unlikely AMD will be able to increase clocks much farther, unless it fixes its IPC problem [lets see what SR does on that front], its chips will continue to perform behind Intels in games. No amount of threading is going to fix that problem. So when everyone on the forum repeats the BD mantra of "Buy PD; its performance will pass Intel's early next year", I take issue.

AMD will likely never get the performance crown back, but they can still compete on price/performance. Intel likes to keep their 60% margins. AMD is happy with 20-30%. Anyone can slide in this market. Even Apple has taken a big hit due to their high margin products.

AMDs desktop plans do seem to be in flux. Their Jaguar chips are selling well into OEMs. They have to focus on profitability first, and leverage from there. They need more investors which should come back when they're profitable again. Lack of a new FX chip hurts with enthusiasts but not so much with the PC market as a whole.

Kaveri being late sucks but AMD has limited resources to work with. If they spent more to get Kaveri out now they would remain in the red and hurt themselves more in the long run.

 

[juanrga]

@juanrga

games have ALWAYS favored cpus with l3 cache. athlon > duron ; pentium > celeron ; phenom > athlon ; fx > apu.

granted 16mb is probably a bit much, intel's normal cpus only have ~6mb and only the extremes have over 10mb. no l3 cache will have lower performance unless by some miracle the cpu core is strong enough to compensate the lack of. even if kaveri is faster than the I5, fx SR will be that much faster.

@juanrga

games have ALWAYS favored cpus with l3 cache. athlon > duron ; pentium > celeron ; phenom > athlon ; fx > apu.

granted 16mb is probably a bit much, intel's normal cpus only have ~6mb and only the extremes have over 10mb. no l3 cache will have lower performance unless by some miracle the cpu core is strong enough to compensate the lack of. even if kaveri is faster than the I5, fx SR will be that much faster.

It's not the presence of L3 but total cache size respective to each thread and the speed of that cache. Intel's caching design is radically different then AMD's, this results in some interesting results. Intel use's an ultra fast L2 cache that is fairly small by today's standards (256KB) vs AMD's larger slower 1MB (2MB per core). This is why Intel use's a large L3 cache, their 256KB isn't large but they can always fall back to the L3 prior to hitting system memory. Cache L2 cache miss's aren't a big deal for the Intel design, their absolutely crippling to the AMD one. This is also why you must be EXTREMELY careful with synthetic benchmarks. They tend to use incredibly small working data sets that are easy to predict and fit into cache memory, this in turn can mask real world performance issues.

SR introduces improvements in the L1/L2 caches. The L1-->L2 interface increases. The L1 cache increases in size, improving efficiency by a 30%. SR also introduces a decoded micro-op queue, that some consider a kind of cache. All those improvements together would reduce the effect of the presence/absence of a L3 cache. A FX SR with a L3 would be still faster, but I don't think that "much" is the word to be used here.

 

[juanrga]

My Kaveri/SR predictions just so I may throw it out there well in advance before any definite leaks come out;

x86 - Tweaks to Cache, IMC and subsystems along with evolutionay tweaks to the architecture will see performance per watt and performance per clock improvements. Over the A10 6800K at same clocks should see around 20-40% improvements depending on the application, around 15% general performance increase along with power improvements.

iGPU - Spectre should be around 2x faster than the HD8670D on Richalnd

I wait a 20% over Piledriver and probably a clock reduction from Richland: 4.1Ghz --> 4 GHz

Based exclusively in GPU cores, I wait about a 30% over Richland/Trinity plus a clock boost up to 900 MHz. This could add to about a 40% increase. How do you got the 2x estimation?

It seems Kaveri is clocked the same as Richland from hearsay so it will offer performance/clock increases over the PD based arch.

I believe the Spectre iGPU is around and between a HD7730 and HD7750 performance which is roughly double where Trinity is at, The GCN cores are far better at tessellation and compute than the VLIW parts so its GPGPU and General purpose(gaming) will improve significantly. At the same time OBR stated that FX9000 would be the last ever AMD traditional DT CPU which is transpiring to be true the same rumours at that time were that Kaveri would be at entry level gaming ability to be at that point the APU's needed roughly twice the performance.

Thanks by this very useful information. I researched a bit and it seems that GCN cores are about a 50% faster than VLIW5 cores. Considering the higher GPU clock and large number of cores (512 vs 384) I am able to obtain now your 2x estimation. If finally true, this is a very important advance.

 

[juanrga]

Once again. Look at the FPS for the OC FX-4300 and the FX-8350, both are 30 FPS. There is no bottleneck, because OC the 8350 gives an improvement on the FPS that coincides with the OC.

First off, the FX-4300 gets 22 FPS, and 29 when OCd to 4.7GHz. The FX-8350 gets 30 FPS at stock, 33 when OC'd to 4.7GHz.

The FX-4300 is clearly bottlenecked, and only OCing to around 4.7GHz alleviates this. Even then, performance is lagging behind the OC'd FX-6300 and FX-8350, so I'm guessing at least one core is still overloaded, even with a 900MHz OC.

The FX-6300 also shows signs it is near a CPU bottleneck, though that could be due to its slower clock of 3.5GHz. When OC'd, it performs identical to the OC'd FX-8350 at the same clock (as one would expect).

Based on the FX results, I have to conclude the FX-4300 is CPU bottlenecked, and only a heavy OC will overcome it for this benchmark. The FX-6300 starts off bottlenecked, likely due to its slower base clock, but the bottleneck goes away with an OC. The FX-8350 is the only AMD CPU not bottlenecked for this particular Crysis 3 benchmark.

Makes one wonder how the i3 would fare with the same 4.7GHz OC...Faster then the FX-8350 prehaps?

The same about the i5 and i7. Both obtain the same FPS and OC both obtain more FPS.

Because neither is bottlenecked to start; performance increases will be dominated by per-core performance (clock). Nevermind the minimal gains of HTT; you can generally assume the i5 and i7 should match eachtother within a few FPS at the same clock (which they do).

Throw in the minimal FPS gain for a 1GHz OC (3 FPS), and I have to assume there's a GPU bottleneck occurring at 45FPS, so IB performance is probably greater then what the results show.

This is a clear symptom that "Roof" is not threaded enough. That is why the i3 perform so well. Whereas "Welcome" is heavily threaded, and the i7 is much faster than the i5 when HT is activated and the FX-8350 is much faster than the FX-4300.

In "Welcome", the FX-4300 @ stock is loaded at 95%. It is the dual cores, which are loaded at 100%.

And the dual core i3 outperforms the FX-4300. The FX-6300 does far better, but is then outperformed by the quad core i5. Seeing the recurring trend here? Intels duo outperforms the AMD quad. Intels Quad outperforms the AMD Sexta.

The reason for that is simple: Both the FX-4300 and i3-3220 are CPU bottlenecked. Between them however, the superior IPC of the i3 is enough to outperform the FX-4300's extra 500MHz and two extra cores. The FX-6300 however, is NOT CPU bottlenecked, hence the massive jump (9 FPS) even at a lower clock then the FX-4300 (300MHz slower).

Compare the FX-6300 against a quad core i5, however, and the trend starts to repeat itself, where the extra IPC of the i5 outperforms the two extra cores of the FX-6300. (The FX-8350's spot is likely due more to its clock rather then its number of cores; would need to clock the FX-8350 to 3.5GHz to confirm though.)

But its worth noting: In two cases, subtracting two cores but adding to IPC improves performance, even though those extra cores are being used. Which is what I'd expect. The i3 is running 9 threads on two cores, but due to its higher IPC, it can chug through those threads at a faster rate. By contrast, the FX-4300 can do twice as much work at a time, but each core is slower, so each thread takes longer to finish. Same trend repeats for the FX-6300 and the i5 lineup. Thats AMD's IPC problem going forward.

My point being, since its unlikely AMD will be able to increase clocks much farther, unless it fixes its IPC problem [lets see what SR does on that front], its chips will continue to perform behind Intels in games. No amount of threading is going to fix that problem. So when everyone on the forum repeats the BD mantra of "Buy PD; its performance will pass Intel's early next year", I take issue.

By "bottleneck" I did mean GPU bottleneck.

You don't need to guess. Core loads figures were given before. No single core of the FX-4300 is overloaded by the engine.

I don't consider that a 33% improvement in crysis 3 due to HT is a "minimal gain". As stated above the i7 is "much faster than the i5 when HT is activated". That increase in performance due to using more threads is far behind IPC gains in last two Intel generations together: Sandy Bridge --> Haswell.

The FX-8350 is a 59% faster than the FX-4300. This means that current poorly threaded games are missing that 59% of extra performance from the chip.

One try more and the latter one. AMD is behind Intel in most current games because most current games are poorly threaded (this is due to current consoles archs.) and ignore the extra cores. Next consoles are 8-core archs. with weak cores, which means game developers cannot rely on a single strong core for doing most of the work and ignore the rest of cores in the chip.

Intel will be behind AMD in next gen games. This has been stated by main game developers. The last one is PlanetSide 2 dev, who said a pair of days ago (bold emphasis from mine):

We have the exact same kind of Achilles’ heel on the PC too. People who have AMD chips have a disadvantage, because a single core on an AMD chip doesn’t really have as much horsepower and they really require you to kind of spread the load out across multiple cores to be able to take full advantage of the AMD processors.

Our engine sucks at that right now. We are multi-threaded, but the primary gameplay thread is very expensive. The biggest piece of engineering work that they’re doing right now, and it’s an enormous effort, is to go back through the engine and re-optimize it to be really, truly multi-threaded and break the gameplay thread up. That’s a very challenging thing to do because we’re doing a lot of stuff – tracking all these different players, all of their movements, all the projectiles, all the physics they’re doing.

It’s very challenging to split those really closely connected pieces of functionality across in multiple threads. So it’s a big engineering task for them to do, but thankfully once they do it, AMD players who’ve been having sub-par performance on the PC will suddenly get a massive boost – just because of being able to take the engine and re-implement it as multi-threaded.

I’m very excited about that because I have a lot of friends, lots of people who are more budget minded, going for AMD processors because nine times out of ten they give a lot of bang for the buck. Where it really breaks down is on games with one really big thread. PlanetSide’s probably a prime example of that.

All triple-A game developers already chose the FX-8350 as best gaming cpu for future games. They chose the FX chip over the i5-3570k. The i3 was not even in the menu...

 

[blackkstar]

All triple-A game developers already chose the FX-8350 as best gaming cpu for future games. They chose the FX chip over the i5-3570k. The i3 was not even in the menu...

In before it loses to 4770k which costs nearly twice as much as FX 8350.

For $340 you can choose between 4770k and 7870 or FX 8320 and 7970.

At that point your CPU benchmarks mean nothing, 7970 is significantly ahead of 7870.

Also, it would be nice if you guys realized that a benchmark with a GTX 680 at 768p indicates that every single CPU is a bottleneck. The entire point of that sort of benchmark is to see which CPUs do best while they are the bottleneck. Extrapolating that FX 8350 is bottlenecking at 768p and that the Intel isn't is insane. If the 4.5ghz 3570k was not bottlenecking then the GTX 680 would be at 100% utilization in the benchmark. Can you get a GTX 680 to hit 100% utilization at 768p and become the bottleneck of performance as opposed to the CPU?

If the GPU is entirely the bottleneck the choice of CPU means nothing. Meaning that these graphs are useless for building a gaming rig unelss you are going to have enough GPU power that your CPU will become the bottleneck.

Some of you need to step back and be a little more critical of what you're posting. Instead of trying to find evidence to prove your point, perhaps you would be best off finding evidence and drawing conclusions from them.

The only thing the "roof" benchmark proves is that the "roof" benchmark is not heavily threaded and that Intel (surprise) does better when you don't have a heavily threaded workload.

Also you folks seem to think that if a game is threaded to run on 8 cores that it's always going to run on 8 cores. Any FX owner that's spent any time playing threaded games will know that if a game is capable of using 8 cores, it does not mean it will always use 8 cores.

I have played FC3 on my FX 8350 and seen horrible frame rates in certain areas while glancing over to my second monitor with a CPU usage graph going and seeing it's maxing out one core and the others are sitting idle.

"Our game engine uses 8 cores" does not mean the game uses 8 cores all the time. It means that it is capable of using 8 cores. If I wrote a game engine and it only used 8 cores to load a map and then went to using a single thread for everything, I could tell you my game engine uses 8 cores and not be lying to you, because it does use 8 cores, just not all the time.

 

[GOM3RPLY3R]

All triple-A game developers already chose the FX-8350 as best gaming cpu for future games. They chose the FX chip over the i5-3570k. The i3 was not even in the menu...

In before it loses to 4770k which costs nearly twice as much as FX 8350.

For $340 you can choose between 4770k and 7870 or FX 8320 and 7970.

At that point your CPU benchmarks mean nothing, 7970 is significantly ahead of 7870.

Also, it would be nice if you guys realized that a benchmark with a GTX 680 at 768p indicates that every single CPU is a bottleneck. The entire point of that sort of benchmark is to see which CPUs do best while they are the bottleneck. Extrapolating that FX 8350 is bottlenecking at 768p and that the Intel isn't is insane. If the 4.5ghz 3570k was not bottlenecking then the GTX 680 would be at 100% utilization in the benchmark. Can you get a GTX 680 to hit 100% utilization at 768p and become the bottleneck of performance as opposed to the CPU?

If the GPU is entirely the bottleneck the choice of CPU means nothing. Meaning that these graphs are useless for building a gaming rig unelss you are going to have enough GPU power that your CPU will become the bottleneck.

Some of you need to step back and be a little more critical of what you're posting. Instead of trying to find evidence to prove your point, perhaps you would be best off finding evidence and drawing conclusions from them.

The only thing the "roof" benchmark proves is that the "roof" benchmark is not heavily threaded and that Intel (surprise) does better when you don't have a heavily threaded workload.

Also you folks seem to think that if a game is threaded to run on 8 cores that it's always going to run on 8 cores. Any FX owner that's spent any time playing threaded games will know that if a game is capable of using 8 cores, it does not mean it will always use 8 cores.

I have played FC3 on my FX 8350 and seen horrible frame rates in certain areas while glancing over to my second monitor with a CPU usage graph going and seeing it's maxing out one core and the others are sitting idle.

"Our game engine uses 8 cores" does not mean the game uses 8 cores all the time. It means that it is capable of using 8 cores. If I wrote a game engine and it only used 8 cores to load a map and then went to using a single thread for everything, I could tell you my game engine uses 8 cores and not be lying to you, because it does use 8 cores, just not all the time.

+1

 

[griptwister]

Well, I guess FC3 is an example of a poorly threaded game? LOL! xD

Also, Interesting results... it's a 4770k Review, but look at who leads the benchmark...

http://www.overclockersclub.com/reviews/intel_core_i7_4770k/9.htm

 

[Cazalan]

All triple-A game developers already chose the FX-8350 as best gaming cpu for future games. They chose the FX chip over the i5-3570k. The i3 was not even in the menu...

With the desktop roadmap stall it makes me wonder if AMD will just scale up Jaguar clock speeds and go to 8 cores like PS4. We already know the Puma+ core is on queue for next year.

 

[mlscrow]

All triple-A game developers already chose the FX-8350 as best gaming cpu for future games. They chose the FX chip over the i5-3570k. The i3 was not even in the menu...

With the desktop roadmap stall it makes me wonder if AMD will just scale up Jaguar clock speeds and go to 8 cores like PS4. We already know the Puma+ core is on queue for next year.

That is exactly what AMD is doing. That is what the FX9590 is. A clocked up piledriver based chip, maintaining the 8 core count rather than moving to a Steamroller based chip. Since XBOX1 and PS4 are both going to be using 8-core AMD chips (Jaguar) and all video game developers will be coding with that hardware and architecture in mind, then they will want that same architecture for PC gamers. It makes sense, it's just not what us enthusiasts wanted. We all want an 8-core Steamroller CPU/APU, and even more so for an APU if Hybrid Xfire becomes standard and works perfectly.

 

[noob2222]

^^ actually ps4 and xb1 are jaguar cores not pd. Jaguar cores were designed for bulk silicon and are based off the bobcat APU, Kabini on steroids.

Now that I think about it, I wonder if PD/SR simply won't work on bulk silicon as part of the design itself is to take advantage of SOI.

 

[juanrga]

All triple-A game developers already chose the FX-8350 as best gaming cpu for future games. They chose the FX chip over the i5-3570k. The i3 was not even in the menu...

In before it loses to 4770k which costs nearly twice as much as FX 8350.

For $340 you can choose between 4770k and 7870 or FX 8320 and 7970.

At that point your CPU benchmarks mean nothing, 7970 is significantly ahead of 7870.

Also, it would be nice if you guys realized that a benchmark with a GTX 680 at 768p indicates that every single CPU is a bottleneck. The entire point of that sort of benchmark is to see which CPUs do best while they are the bottleneck. Extrapolating that FX 8350 is bottlenecking at 768p and that the Intel isn't is insane. If the 4.5ghz 3570k was not bottlenecking then the GTX 680 would be at 100% utilization in the benchmark. Can you get a GTX 680 to hit 100% utilization at 768p and become the bottleneck of performance as opposed to the CPU?

If the GPU is entirely the bottleneck the choice of CPU means nothing. Meaning that these graphs are useless for building a gaming rig unelss you are going to have enough GPU power that your CPU will become the bottleneck.

Some of you need to step back and be a little more critical of what you're posting. Instead of trying to find evidence to prove your point, perhaps you would be best off finding evidence and drawing conclusions from them.

The only thing the "roof" benchmark proves is that the "roof" benchmark is not heavily threaded and that Intel (surprise) does better when you don't have a heavily threaded workload.

Also you folks seem to think that if a game is threaded to run on 8 cores that it's always going to run on 8 cores. Any FX owner that's spent any time playing threaded games will know that if a game is capable of using 8 cores, it does not mean it will always use 8 cores.

I have played FC3 on my FX 8350 and seen horrible frame rates in certain areas while glancing over to my second monitor with a CPU usage graph going and seeing it's maxing out one core and the others are sitting idle.

"Our game engine uses 8 cores" does not mean the game uses 8 cores all the time. It means that it is capable of using 8 cores. If I wrote a game engine and it only used 8 cores to load a map and then went to using a single thread for everything, I could tell you my game engine uses 8 cores and not be lying to you, because it does use 8 cores, just not all the time.

We agree on many things.

I doubt that the FX 8350 will lose in many multithreaded benchmarks when compared to the 4770k.

We were comparing the CPUs, using the same GPU. Of course, you can use the difference in price between CPUs to obtain a better GPU, but then we would be comparing CPU+GPU pairs.

Benchmarks at 768p minimize the possibility of a GPU bottleneck and maximize the differences between CPUs, which is the point when one compares CPUs. In any case, I have provided benchmarks at 1080p, which is a realistic gaming resolution for many users. Of course, at higher resolutions the GPU can be the bottleneck and then the differences between CPUs minimize, but in that case you can use a faster GPU (or a dual, triple... setup) and increase again the differences.

That is right, but we are not discussing gaming rigs...

Effectively, the "roof" benchmark is not heavily threaded. I have been saying this for days.

In my previous post, I provided a quote from a game developer explaining how his multithreaded engine was not really using all the cores. This is typical for several current engines that are advertised as multithreaded, but are far from the ideal. I don't want to give names here.

Let me be clear about this. The next consoles use 8 weak cores. Engines cannot rely on all the cores for a part of the game and then use only 2 cores for another part of the game. Developers will need their engines to go wide for all the game, unless they want provide bad performance/gameplay. The same developer that I quoted admits that they are working in what he called a "true" multithreaded engine precisely for the new consoles.

As a consequence, games ported to PC will be well threaded and people with multicore AMD chips will see improvement in all games.

 

[juanrga]

All triple-A game developers already chose the FX-8350 as best gaming cpu for future games. They chose the FX chip over the i5-3570k. The i3 was not even in the menu...

With the desktop roadmap stall it makes me wonder if AMD will just scale up Jaguar clock speeds and go to 8 cores like PS4. We already know the Puma+ core is on queue for next year.

Look at Intel. 10x the resources of AMD and still failed to design a core (Haswell) that scales well from ultra-portable to high-end desktop/server. Intel focused on mobile with Haswell and has problems both at one extreme (tablets) and at the other extreme (top desktop).

The i7-4770k is about a 5% faster than the i7-3770k, has some performance regressions (i.e. is slower than the chip that replaces), run hotter, and OC poor.

Now take a look at tablets. Haswell has a too high power consumption and Intel has tried to lie about it. In case you did not notice Intel has been caught again, a pair of weeks ago, with fake power consumption figures for its Haswell tablet chips.

AMD Jaguar chip is beyond anything Intel can offer, but if AMD was to scale jaguar above 2GHz we would see similar problems to those experienced by Intel Haswell. I think that AMD dual arch. approach, Jaguar for low freq. and Steamroller for high freq., is fine.

AMD claims that Steamroller has double the performance of Jaguar

http://www.amd.com/us/press-releases/Pages/amd-unveils-2013june18.aspx

AMD means that Jaguar scales up to 2Ghz, whereas Steamroller scales up to 4 GHz, whereas maintaining the IPC.

AMD will maintain this dual arch. approach. Puma+ will replace Jaguar, whereas Excavator will replace Steamroller.

We did hear rumours that AMD will launch a version of the PS4 APU for the PC, but in IHMO it will be six-core and based in Steamroller.

Reasoning behind my guess:

* The 8-core jaguar PS4 performs like an i7 IB (in a PC). I am considering Windows overhead and ignoring HSA.

* The 4-core Steamroller Kaveri performs like an i5 SB. A 6-core Steamroller would perform like an i7 IB.

 

[gamerk316]

In my previous post, I provided a quote from a game developer explaining how his multithreaded engine was not really using all the cores. This is typical for several current engines that are advertised as multithreaded, but are far from the ideal. I don't want to give names here.

I'm assuming you are referring to Sony's remakes about Planetside?

http://www.eurogamer.net/articles/2013-08-05-sony-explains-lack-of-planetside-2-pc-and-ps4-cross-platform-play-teases-character-transfers

"It really is a challenge to optimize high-end PC games to be able to work on the pantheon of hardware that's available to players nowadays, it's just insane.

"The PS4 is a much more consistent, stable platform for us to be able to develop for. The big challenge with the PS4 is its AMD chip, and it really, heavily relies on multi-threading. We have the exact same kind of Achilles heel on the PC too. People who have AMD chips have a disadvantage, because a single core on an AMD chip doesn't really have as much horsepower and they really require you to kind of spread the load out across multiple cores to be able to take full advantage of the AMD processors.

"Our engine sucks at that right now. We are multi-threaded, but the primary gameplay thread is very expensive. The biggest piece of engineering work that they're doing right now, and it's an enormous effort, is to go back through the engine and re-optimise it to be really, truly multi-threaded and break the gameplay thread up. That's a very challenging thing to do because we're doing a lot of stuff - tracking all these different players, all of their movements, all the projectiles, all the physics they're doing.

"It's very challenging to split those really closely connected pieces of functionality across in multiple threads. So it's a big engineering task for them to do, but thankfully once they do it, AMD players who've been having sub-par performance on the PC will suddenly get a massive boost - just because of being able to take the engine and re-implement it as multi-threaded.

So you're right back to the root of the problem: The main gameplay thread is hard to break up in such a way that you don't end up reducing performance due to threads locking eachother out. Its doable on the PS4/XB1, due to being able to code to the metal, and being able to guarantee when threads run, less so on the PC.

Look, I've been writing software 20+ years. I've never delivered a program that has any significant problems. I've worked on highly threaded parallel processors, and I've worked on aged 16-bit minicomputers. The assumption by many here, that you can just arbitrarily thread things out (EG: Give a separate thread to each different part of the engine) is wrong. Trying to thread in such a way reduces performance due to various stalls (thread communication, deadline management, kernel lockout, etc). It takes a LOT more time and effort, and from my experience, you'll eventually find some setup your solution has significant performance issues.

So you'll continue to see what you are seeing today: The majority of gaming benchmarks favoring Intel due to its higher IPC, with a handful going AMD's way due to AMD's high clock speed. That isn't going to be changing anytime soon. If you have two threads that result in 80% of the process load, that favors the processor with at least two fast cores. Now say we manage to spread that workload over four threads instead. That STILL favors Intel, due to 80% of the work being on four threads, thus a faster quad would still outperform a slower 8-core CPU. AMD wouldn't see a performance lead, at the same clock as Intel, until you manage to split the load evenly between 8+ threads, which is not possible for games. Period.

3 years ago, the mantra in the BD forums was "Buy BD; games will catch up in 3 years". 3 years later, the mantra is "Buy ST; games will catch up in 3 years".

 

[BeastLeeX]

lol the 8 core jaguar in the ps4 performs similar to an i3 if not slower.

I hate to agree with this guy, but that sounds about right to me. Not to mention the PS4 will use 2 of it's cores for the OS at all times, and game developers can only use 4.5gb of ram, the gpu will also not use all of its cores in a game (at least this is most of the news I have heard). Right now it sounds like both console makers are making there consoles sound sooo good, but half of the hardware in the console is already cut down to support other task, so you won't get a full experience (or full potential of the console) when gaming. To me the console sounds like it will be even to a FX-6300, 7850, and 6gb of ram (only if you OC the 7850, which is a very simple thing to do with a card with such low heat output).

 

[The Q6660 Inside]

The PS4 will likely be on par with a FX6300, 7870 and 8GB of RAM similar to what Beast said. The FX6300 isnt exactly an i3, but it isn't exactly an i7 either...

 

[etayorius]

Has anyone read this:

http://vr-zone.com/articles/hold-for-publication-why-its-a-big-deal-that-amd-is-delaying-kaveri/49389.html

It pretty much says the FX 9590 is the Last FX... so no more SR FX and most probably no more AM3+ CPUs, i believe only APUs for FM2+, it may just be true since we know AMD scrapped anything related to CPUs in their Roadmap...

 

[amdfangirl]

Very disappointing to say the least. I was hoping for a Steamroller upgrade.

 

[The Q6660 Inside]

inb4PhenomIII AMD has had its mouth shut about Steamroller FX, so perhaps they just give up on it and move on to APU.

 

[Ranth]

Has anyone read this:

http://vr-zone.com/articles/hold-for-publication-why-its-a-big-deal-that-amd-is-delaying-kaveri/49389.html

It pretty much says the FX 9590 is the Last FX... so no more SR FX and most probably no more AM3+ CPUs, i believe only APUs for FM2+, it may just be true since we know AMD scrapped anything related to CPUs in their Roadmap...

Are there other sites saying the same? Just seems a bit too much rumorish to me..
Also that part with the watermarks was quite weird.

 

[noob2222]

ATOM 8 CORE will embarass the jaguar cores. Intel may have 10x the resources but they have most of the cpu market for a reason as they are years ahead of where amd are.

First off jaguar isn't designed to compete with the I3 so might as well say its going up aganst the 4770k, what the heck lets compare a $50 cpu to Intel's $1000 extreme instead. Kinda hard to go wrong with this statement isn't it? At least you got one thing right, by comparing 2 entirely different market chips.

Atom 8-core ROFL. Kabini is twice as fast as clovertrail in single threaded apps and nearly 3x in multithreaded. And on top of that, thats jaguar at 1.5ghz and Atom at 1.8. I don't think ATOM's 25% increase is going to make it embarass anything other than your ego.

http://www.anandtech.com/show/6974/amd-kabini-review/3

 

[noob2222]

Has anyone read this:

http://vr-zone.com/articles/hold-for-publication-why-its-a-big-deal-that-amd-is-delaying-kaveri/49389.html

It pretty much says the FX 9590 is the Last FX... so no more SR FX and most probably no more AM3+ CPUs, i believe only APUs for FM2+, it may just be true since we know AMD scrapped anything related to CPUs in their Roadmap...

Are there other sites saying the same? Just seems a bit too much rumorish to me..
Also that part with the watermarks was quite weird.

they are speculating that the absence of a roadmap means it will never exist. If AMD does abandon entheusiasts, they might as well end up like VIA. They have a cpu, but who cares.

 

[Cazalan]

GF 28nm is apparently good enough for Qualcomm. They're shifting 20% to GF from TSMC to meet demand. Which is much higher than AMD volumes BTW. Makes you wonder what the issue is with Kaveri. Or did Qualcomm buy their way to the front of the line.

http://www.digitimes.com/news/a20130806PD215.html

 

[Cazalan]

That isn't going to be changing anytime soon. If you have two threads that result in 80% of the process load, that favors the processor with at least two fast cores. Now say we manage to spread that workload over four threads instead. That STILL favors Intel, due to 80% of the work being on four threads, thus a faster quad would still outperform a slower 8-core CPU. AMD wouldn't see a performance lead, at the same clock as Intel, until you manage to split the load evenly between 8+ threads, which is not possible for games. Period.

It will have to change or PS4/XBone games will just plain suck. A single Jaguar core isn't that strong. They really don't have a choice in the matter. It's do or die.

That's when you get the biggest breakthroughs in technology when you're forced to evolve.

 

[daerohn]

I can not understand why people say Intel is years ahead of AMD. AMD is trying to change the whole CPU achitecture. I believe the performance problem of AMD chips is because there is no software to use its full potential. As I have said in my previous post, Intel forces most of the OEM's to optimize their software solely for Intel chips. Saying AMD APU's that will be used in xbox and ps4 are weak is also nonsense. Do you really beleive that these companies will invest their money for a sure to fail system which they will be selling in the next 5-6 years? When you look in the progress of their previous hardware, the first games that were released could'nt use the full potential of the hardware but in time they are fully optimized. The point is that in the next few years games and softwares will be optimized for AMD. which in the end will show us who has the leading edge technology in CPU's. Now it is too early to say Intel is way ahead of AMD