AMD CPU speculation... and expert conjecture

[juanrga]

Juan is just using the forum to promote his blog site.

I don't know what makes laugh me more your repetitive lies against me or you personal delusions.

I also notice the double standard in this forum. You are one of those here who complain that this thread is not about ARM, but you don't complain when others posters have been posting about ARM during last days...

The on-topic argument is only used when your mistakes are corrected. Hypocrisy at its best.

 

[tracker45]

Kaveri will be competitive with Intel's Core i5-4670K processor

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


Well, they could always be lying. AMD don't have a good reputation for reaching expectations.

 

[de5_Roy]

Kaveri will be competitive with Intel's Core i5-4670K processor
http://techreport.com/news/25707/all-signs-point-to-kaveri-being-an-evolutionary-upgrade
Well, they could always be lying. AMD don't have a good reputation for reaching expectations.

kaveri will be competitive with core i5 the same way richland is competitive with core i5. amd isn't lying, simply hiding the whole truth under pile of bench/marketing.
amd has been unable to match intel's consistency of execution (despite some of intel's massive failures), that's why they're trying to branch out to markets where they can compete better.

 

[-Fran-]

The dead of the FX platform was suspected since the beginning of this year and confirmed latter. It has been explained here during months that AMD is abandoning FX and Opteron CPUs by APUs. Don't understand people surprise.

That's where our interpretations of the slides differ, I guess.

I don't think they're abandoning it, but putting them on hold until they get a competitive product. In the particular case of the FX line, it's not dead, but it won't have a refresh in long long time. Similar, but not the same. Killing it would be indicated by removing them completely from their roadmaps and, for example, a symptom would be no more AM3+ MoBos in the market for the following year. I haven't read anywhere that Asus, MSI or any other OEM would stop making AM3+ MoBos in the coming years. In the server landscape, they have such little market share that "killing the line" can be interpreted as "stop trying to sell a crappy product there".

Like I said a while back, I think they'll wait for a lot of factors to come in together in order to pick up with the FX/Phenom/Opteron (X86) line. And I also agree they're trying to make the APUs a thing for servers, but it's still in is infancy against a well established market of Xeons, Sun Blades (haha) and IBM Power CPUs. It will take a lot of time to switch the boat into the APU direction. HSA is the key component in that regard.

Also, I think once the money given to them for the research bears fruition, we'll have a more clear look on what AMD will do for real.

Cheers!

 

[elemein]

]Either way; the basis rests, I dislike the FX module design (finally something OT.)

I don't like the module design either. It has trade offs made for server workloads at the time it was being developed.

They should scale up Jaguar for higher performance.

It has tradeoffs. Compare to Intel HTT: HTT is REALLY cheap to add to the chip (like 10% extra die space), for a decent performance gain in some subset of tasks. Its basically free for Intel to throw in and charge a $50 premium.

AMD's CMT is a lot more powerful, but a LOT more expensive, since you are basically duplicating everything but the CPU dispatchers and schedulers. Its ALMOST like adding a full core. The downside however, is AMD made each individual core very weak.

What AMD failed to recognize (and I called), is a four core chip can be faster then an 8 core chip, if the cores of the 4 core chip are faster and NO INDIVIDUAL CORE IS BOTTLENECKED AT ANY POINT IN TIME. That's where Intel has positioned itself going forward; it doesn't need 20% CPU gains per generation. As long as they keep the CPU fast enough where they can keep any individual core from being bottlenecked, they will hold the performance crown, regardless of how many cores AMD slaps on.

By contrast, AMD has the problem where because it has weaker cores, it is MUCH easier for any single one to bottleneck, and kill performance of the entire CPU. Hence why they are reliant on programs being heavily threaded, and why BD/PD stinks at any single-threaded benchmark. This also explains why the FX-4xxx series does so badly compared to i5's.

Intel's SMT implementation is actually even cheaper than 10% to add. Much lower (maybe 10% of a core, but not 10% of a whole die-- oh heck no!)

AMD has a very funny way of moving forward. Programs are naturally becoming more and more threaded, and, when it comes to pure Integer performance; all 4 FX modules of an 8320 + their clock advantage can definitely beat an i5-4670K in something purely integer-based with all 4 FX modules at full steam.

It's when you throw in a vector or floating point instruction that things get very "slow" for the FX, and of course, only have a few integer EUs to rely on per core makes single or double threaded tasks a pure fright for Piledriver. Meanwhile, Intel has a massive amount of throughput potential for ANY calc type-- it has four integer EUs per core, 4 decoders, a total of 768-bits of pure floating point throughput (256 bitx3) per core-- and honestly, it's very powerful in comparison to even a full FX module.

But lord, FX clocks easily and HIGH, and it has some ISA advantages...

It's a close battle, but in most cases, the i5 wins.

 

[szatkus]

Kaveri will be competitive with Intel's Core i5-4670K processor

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


Well, they could always be lying. AMD don't have a good reputation for reaching expectations.

Read again.

If one measures x86 performance on its own, Kozak said, "we'll lose."

Kaveri will be supieror on CPU+GPU power. Just like Richland.
Also, A10's competition is i3.

 

[tracker45]

What is x86 performance ? Does that mean just pure cpu performance ?

 

[etayorius]

Kaveri, 20% faster on CPU workloads only compared to Richland... that pretty much fits the i5 2500k performance, but the i5 4670k would be about 15-20% faster than Kaveri, i think the performance gains are quite good and would put Kaveri 40-45% faster than my PhenomII x4 980, if FM2+ will be all long up to late 2015 i think i will just build a Kaveri System with a R9 280X, HSA and MANTLE will be worth it and most probably even be faster than a i7 3770k IN Gaming.

If Excavator APU will be FM2+ then i have another reason to stick to FM2+

 

[juanrga]

The dead of the FX platform was suspected since the beginning of this year and confirmed latter. It has been explained here during months that AMD is abandoning FX and Opteron CPUs by APUs. Don't understand people surprise.

That's where our interpretations of the slides differ, I guess.

I don't think they're abandoning it, but putting them on hold until they get a competitive product. In the particular case of the FX line, it's not dead, but it won't have a refresh in long long time. Similar, but not the same. Killing it would be indicated by removing them completely from their roadmaps and, for example, a symptom would be no more AM3+ MoBos in the market for the following year. I haven't read anywhere that Asus, MSI or any other OEM would stop making AM3+ MoBos in the coming years. In the server landscape, they have such little market share that "killing the line" can be interpreted as "stop trying to sell a crappy product there".

Like I said a while back, I think they'll wait for a lot of factors to come in together in order to pick up with the FX/Phenom/Opteron (X86) line. And I also agree they're trying to make the APUs a thing for servers, but it's still in is infancy against a well established market of Xeons, Sun Blades (haha) and IBM Power CPUs. It will take a lot of time to switch the boat into the APU direction. HSA is the key component in that regard.

Also, I think once the money given to them for the research bears fruition, we'll have a more clear look on what AMD will do for real.

Cheers!

I agree with you on that "dead" is not the correct word. AMD is maintaining the FX line until they have ready an APU replacement: a six-core or octo-core APU.

The desktop roadmaps are simply confirming information was leaked many months ago. E.g. the docs saying that the 9590 was the last FX CPU.

 

[juanrga]

Kaveri, 20% faster on CPU workloads only compared to Richland... that pretty much fits the i5 2500k performance, but the i5 4670k would be about 15-20% faster than Kaveri, i think the performance gains are quite good and would put Kaveri 40-45% faster than my PhenomII x4 980, if FM2+ will be all long up to late 2015 i think i will just build a Kaveri System with a R9 280X, HSA and MANTLE will be worth it and most probably even be faster than a i7 3770k IN Gaming.

If Excavator APU will be FM2+ then i have another reason to stick to FM2+

Excellent resume!! Under ordinary CPU workloads:

Kaveri ~ 80% of i5-4670k

Using MANTLE:

[ Kaveri + R9-290X ] ~ [ i7-4770k + R9-290X ]

MANTLE with HSA:

[ Kaveri + R9-290X ] > [ i7-4770k + R9-290X ]

 

[blackkstar]

I am having a very hard time believing this.

AMD basically just spend a ton of money and made all this effort to land all this pull in how games are developed and now they're going to turn around and make HSA feature only available to the sub $150 market for CPU + GPU? And then create Mantle for hardcore gamers where they're running Intel CPUs on a platform without HSA? Meaning that this 2015 roadmap just completely doomed HSA platforms to low end budget systems and mobile.

I was feeling optimistic about HSA but if that roadmap is true and this is what AMD is going to do, HSA is DOA and Mantle won't do anything at all to compliment HSA.

I really don't understand how it would make sense to abandon a market where you have a significant software advantage which adds massive amounts of value to a product only to go compete in a very difficult market.

@gamerK Yes I know the memory subsystems are different, but in the past you had different memory subsystems as well as different chips in general. My point was that even though xbone and ps4 are somewhat different, they are significantly more similar than 360 and ps3. It should be significantly simpler to make a multi-platform game that how it was on the past.

However xbone looks to me like it's going to end up not getting as many games as PS4 and PC in the long run. Multi-plat between Windows, Linux, OSX, and PS4 will be significantly easier because you can use Mantle across the whole stack. xbone embedded memory buffer will be the system's undoing and creating the more difficult to develop for and more expensive system doesn't work.

It is not going to be completely trivial to port from xbone to ps4 to PC, but it will be better than going n64 to PS1 to PC, PS2 to Xbox to Gamecube to PC, SNES to Genesis to PC, etc. I would expect the next generation of consoles after xbone and ps4 to be even more PC-like to make things easier. I'd imagine if xbox is still a brand by that time that game developers would be begging MS to get rid of the ram buffer. It's going to be a bane in the existence of game developers for the next few years, because it's going to be the thing that stops the two from being easy ports.

 

[noob2222]

Juan is just using the forum to promote his blog site.

I don't know what makes laugh me more your repetitive lies against me or you personal delusions.

I also notice the double standard in this forum. You are one of those here who complain that this thread is not about ARM, but you don't complain when others posters have been posting about ARM during last days...

The on-topic argument is only used when your mistakes are corrected. Hypocrisy at its best.

In the entire 2 days you were gone this was the discussion about ARM.

jdwii said:
In 2017 or so Arm will have 10% market share in servers and Amd wants some of that. ...

ya ... thats it .. one half of jdwii's post. you are pretty twisted to say "ermago all I read was ARM this and ARM that."

one posting.

Then you showed back up.

 

[noob2222]

Kaveri, 20% faster on CPU workloads only compared to Richland... that pretty much fits the i5 2500k performance, but the i5 4670k would be about 15-20% faster than Kaveri, i think the performance gains are quite good and would put Kaveri 40-45% faster than my PhenomII x4 980, if FM2+ will be all long up to late 2015 i think i will just build a Kaveri System with a R9 280X, HSA and MANTLE will be worth it and most probably even be faster than a i7 3770k IN Gaming.

If Excavator APU will be FM2+ then i have another reason to stick to FM2+

HSA in gaming has to be specifically programmed in. I can't see Kaveri catching the 3770k in any situation other than specific HSA software, games aren't likely to be on that list for quite a while. HSA is going to be easy to implement on highly parallel workloads such as video encoding and file compression.

2 AMD modules without L3 cache vs 4+4 Intel cores ... CPU wise its not going to happen.

 

[szatkus]

What is x86 performance ? Does that mean just pure cpu performance ?

CPU in Kaveri is x86, so yes.

 

[szatkus]

Using MANTLE:

[ Kaveri + R9-290X ] ~ [ i7-4770k + R9-290X ]

For Mantle CPU's brand doesn't matter.

MANTLE with HSA:

[ Kaveri + R9-290X ] > [ i7-4770k + R9-290X ]

HSA is not on/off button. Game have to actually comprise some GPGPU computations.

 

[-Fran-]

Using MANTLE:

[ Kaveri + R9-290X ] ~ [ i7-4770k + R9-290X ]

For Mantle CPU's brand doesn't matter.

MANTLE with HSA:

[ Kaveri + R9-290X ] > [ i7-4770k + R9-290X ]

HSA is not on/off button. Game have to actually comprise some GPGPU computations.

And also, it will depend on the workloads the GPU can take its hands on. I'll side with gamerk on this point, since it is true that typical workloads in current software are not GPU/parallel friendly at all. The HSA promise needs a lot of re-work done to the core of the problems being solved with software. This means a new approach, plus the promise from AMD & friends to make it as seamless as possible.

Anyway, not so much gloom and doom either. I can speak from the Java paradigm being HSA friendly in the coming years. The major problem (again, from my coding world) will be to let older folks in the Ops teams to understand what HSA means and why switching from a Xeon that has been performing "fine" all these years to an AMD APU will make perfect sense. That is going to be HARD, please trust me on that one, haha. I just had a 2-week fight with our tech support (infrastructure) fellas trying to migrate from 32bits JVM to 64bits (1.6r45) because they were afraid of how the garbage collector would work, hahaha. Man, those long ass meetings trying to make them understand how the GC needed to be tuned for the workloads for the app. Old fixed people with their old school paradigms are the most dangerous thing in tech IMO 😛

Cheers!

 

[szatkus]

And also, it will depend on the workloads the GPU can take its hands on. I'll side with gamerk on this point, since it is true that typical workloads in current software are not GPU/parallel friendly at all. The HSA promise needs a lot of re-work done to the core of the problems being solved with software. This means a new approach, plus the promise from AMD & friends to make it as seamless as possible.

Juan doesn't accept arguments like those, so I wanted to begin with something easier

 

[anxiousinfusion]

Old fixed people with their old school paradigms are the most dangerous thing in tech IMO 😛

This is true beyond your wildest dreams. Toms seriously needs an upvote button.

 

[de5_Roy]

I am having a very hard time believing this.

AMD basically just spend a ton of money and made all this effort to land all this pull in how games are developed and now they're going to turn around and make HSA feature only available to the sub $150 market for CPU + GPU? And then create Mantle for hardcore gamers where they're running Intel CPUs on a platform without HSA? Meaning that this 2015 roadmap just completely doomed HSA platforms to low end budget systems and mobile.

hsa is not for gaming alone.
hsa optimization for entry level lineup is a safe way to ensure fast market penetration in x86. besides, it's not like amd can't upgrade the northbridge in am3+ platform and tune it for hsa standards.
mantle is proprietary despite amd's assurance. imo, the best way for mantle to survive would be to assimilate it's functions into directx or some kind of co-license.

i said it before, ARM (and gang) might turn out to be the biggest gainer out of hsa instead of amd.

 

[gamerk316]

However xbone looks to me like it's going to end up not getting as many games as PS4 and PC in the long run. Multi-plat between Windows, Linux, OSX, and PS4 will be significantly easier because you can use Mantle across the whole stack. xbone embedded memory buffer will be the system's undoing and creating the more difficult to develop for and more expensive system doesn't work.

I note the assumption that Intel and NVIDIA will support Mantle, which is about as likely as AMD supporting PhysX. Intel and NVIDIA control about 80% of the total GPU market, so yeah, not happening.

What I view as more likely is MSFT adding a Mantle like layer to DirectX, similar to how it combined Pixel Model 1.1 (NVIDIA) and Pixel Model 1.4 (AMD) into Shader Model 2.0 in DX9. Course, this would leave consoles + Linux out of date...

Right now, it looks like AMD is following the money, and thats low-power APU's. Which is what I've been saying for years now that AMD SHOULD be doing: Undercut Intel at the mid-low range, and make a comparatively powerful all in one chip, and get in with the OEM's. Build from the bottom up, and eat away at Intel from the bottom.

 

[juanrga]

Juan is just using the forum to promote his blog site.

I don't know what makes laugh me more your repetitive lies against me or you personal delusions.

I also notice the double standard in this forum. You are one of those here who complain that this thread is not about ARM, but you don't complain when others posters have been posting about ARM during last days...

The on-topic argument is only used when your mistakes are corrected. Hypocrisy at its best.

In the entire 2 days you were gone this was the discussion about ARM.

jdwii said:
In 2017 or so Arm will have 10% market share in servers and Amd wants some of that. ...

ya ... thats it .. one half of jdwii's post. you are pretty twisted to say "ermago all I read was ARM this and ARM that."

one posting.

Then you showed back up.

You continue lying. First you omit parts of jdwii's post:

jdwii said:
Arm is a side project i do know that is a fact the one and only person i know who works at Amd even says that. Its a SIDE PROJECT.

In 2017 or so Arm will have 10% market share in servers and Amd wants some of that. APU's with Arm+GPU is enough to get that as well. Compared to other brands that make Arm AMD actually has a relationship with Arm and their the only ones who make high-end GPU graphics capable of producing some power...

Second, you lie about what I said.

Third, you deliberately ignore the other two posters who wrote about ARM. Add one more if you also count posterior Cazalan's post after my return.

 

[juanrga]

Kaveri, 20% faster on CPU workloads only compared to Richland... that pretty much fits the i5 2500k performance, but the i5 4670k would be about 15-20% faster than Kaveri, i think the performance gains are quite good and would put Kaveri 40-45% faster than my PhenomII x4 980, if FM2+ will be all long up to late 2015 i think i will just build a Kaveri System with a R9 280X, HSA and MANTLE will be worth it and most probably even be faster than a i7 3770k IN Gaming.

If Excavator APU will be FM2+ then i have another reason to stick to FM2+

HSA in gaming has to be specifically programmed in. I can't see Kaveri catching the 3770k in any situation other than specific HSA software, games aren't likely to be on that list for quite a while. HSA is going to be easy to implement on highly parallel workloads such as video encoding and file compression.

2 AMD modules without L3 cache vs 4+4 Intel cores ... CPU wise its not going to happen.

MANTLE.

 

[szatkus]

Kaveri, 20% faster on CPU workloads only compared to Richland... that pretty much fits the i5 2500k performance, but the i5 4670k would be about 15-20% faster than Kaveri, i think the performance gains are quite good and would put Kaveri 40-45% faster than my PhenomII x4 980, if FM2+ will be all long up to late 2015 i think i will just build a Kaveri System with a R9 280X, HSA and MANTLE will be worth it and most probably even be faster than a i7 3770k IN Gaming.

If Excavator APU will be FM2+ then i have another reason to stick to FM2+

HSA in gaming has to be specifically programmed in. I can't see Kaveri catching the 3770k in any situation other than specific HSA software, games aren't likely to be on that list for quite a while. HSA is going to be easy to implement on highly parallel workloads such as video encoding and file compression.

2 AMD modules without L3 cache vs 4+4 Intel cores ... CPU wise its not going to happen.

MANTLE.

 

[juanrga]

Using MANTLE:

[ Kaveri + R9-290X ] ~ [ i7-4770k + R9-290X ]

For Mantle CPU's brand doesn't matter.

Nobody said that CPU brand matters. What MANTLE makes was explained before. I will explain it again.

MANTLE eliminates CPU bottlenecks. It was shown during APU13 that under MANTLE a FX-8350 downclocked to 2GHz offers the same performance than an FX-8350 @ 4GHz and that an i7-4770k. The two FX and the i7 were paired with a 290X.

MANTLE with HSA:

[ Kaveri + R9-290X ] > [ i7-4770k + R9-290X ]

HSA is not on/off button. Game have to actually comprise some GPGPU computations.

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.

 

[juanrga]

Using MANTLE:

[ Kaveri + R9-290X ] ~ [ i7-4770k + R9-290X ]

For Mantle CPU's brand doesn't matter.

MANTLE with HSA:

[ Kaveri + R9-290X ] > [ i7-4770k + R9-290X ]

HSA is not on/off button. Game have to actually comprise some GPGPU computations.

And also, it will depend on the workloads the GPU can take its hands on. I'll side with gamerk on this point, since it is true that typical workloads in current software are not GPU/parallel friendly at all. The HSA promise needs a lot of re-work done to the core of the problems being solved with software. This means a new approach, plus the promise from AMD & friends to make it as seamless as possible.

Anyway, not so much gloom and doom either. I can speak from the Java paradigm being HSA friendly in the coming years. The major problem (again, from my coding world) will be to let older folks in the Ops teams to understand what HSA means and why switching from a Xeon that has been performing "fine" all these years to an AMD APU will make perfect sense. That is going to be HARD, please trust me on that one, haha. I just had a 2-week fight with our tech support (infrastructure) fellas trying to migrate from 32bits JVM to 64bits (1.6r45) because they were afraid of how the garbage collector would work, hahaha. Man, those long ass meetings trying to make them understand how the GC needed to be tuned for the workloads for the app. Old fixed people with their old school paradigms are the most dangerous thing in tech IMO 😛

Cheers!

You are right about your ecosystem, but you are missing an important detail here. Your fixed seniors can choose to do the things as they always did and use a Xeon. Game developers cannot chose for new consoles. AMD master plan started with placing a weak 8-core CPU in both consoles. Game developers are being forced to wide their engines/games to use more than a pair of threads or they will lose performance.

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.

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.

Similar thoughts about HSA. Game developers are being forced to offload the CPU and use the GPU to run heavy computations such as AI, physics... Mark Cerny (PS4 lead architect) has given plenty of talks and interviews about this

http://gamingbolt.com/mark-cerny-explains-the-ps4s-gpu-can-perform-asynchronous-complex-processes

http://www.gamasutra.com/view/feature/191007/inside_the_playstation_4_with_mark_.php?page=2