AMD CPU speculation... and expert conjecture

[juanrga]

http://gamer.blorge.com/2013/06/16/controversy-over-xbox-one-games-running-off-windows-7-at-e3/

Of course XboxOne hardware optimizations and cloud will do those high-PC outdated in a pair of years.

(*) The hard part is to compute the performance gain introduced by the custom components found in the consoles and in the cloud system. The cloud offers a theoretical 4x boost for the Xbox One, but nobody know how many of that theoretical performance will be finally used by the games

Uhhh...no, since Etherent (not even the connection; the actual ethernet link) would become the primary system bottleneck. Anything relying on external access will slow down your processing.

But they plan to use the cloud for tasks not affected by latency and bottleneck. Of course the real improvement in performance will be not the theoretical 4x, but the cloud will not slow down processing. Microsoft already showed demos of games using cloud.

Intel isn't ignoring HSA. Their drivers have been upgraded for Haswell and perform quite well now for compute. It's one of the more shocking things about Haswell. Their OpenCL performance of the HD4600 is faster than the A10-6800K.

http://www.tomshardware.com/reviews/core-i7-4770k-haswell-review,3521-3.html

Intel is not member of the HSA foundation and has no plans for HSA support in any future chip, so far as I know.

OpenCL is not HSA. OpenCL is a mere interface to access the GPU in your computer for non-graphic tasks. HSA is about blurring the traditional distinction between CPU and GPU.

http://www.tomshardware.com/reviews/fusion-hsa-opencl-history,3262-8.html

Look at the PS4 APU. Most people think of it as a 8-core CPU plus a given Radeon GPU, but it is something new. There is no real distinction between CPU and GPU at the software level.

It is the programmer who splits the APU into a CPU and a GPU and this split is dynamical. You can use the CPU to assist the GPU or you can use the GPU to assist the CPU. This is very related to the claims that the PS4 has not bottlenecks.

And in other news: The HSA chip will drive software development so much, the XB1 at least was demoed on a Win7 PC with an Intel CPU and NVIDIA graphics.

http://www.fudzilla.com/home/item/31703-xbox-one-demoed-running-gtx-card

The irony might be as bad as the XB1's pre-sale numbers.

The irony is that some demos at E3 were running on real Xbox1 whereas others run in a high-end gaming PC with a GTX-780. This seems to confirm my prediction that the minimum PC for matching a next gen console is a i7 + GTX-780

 

[montosaurous]

Unless you get discounts or actually need it, I don't see the point of buying any Haswell-E chip.

 

[hcl123]

Eh I prefer to use AMD for their price/performance ratio. Haswell was a bit of a failure for a 'tick', I won't disagree with you there. And this thread isn't about consoles either. Lets face it: the days of the Athlon are over. Intel has a much larger R&D department and a superior architecture at the moment, it will take awhile for AMD to catch up. FX kinda does remind me of the Phenoms, and the i5s of Core 2 Duos. The C2D were better for gaming at first, but who's better for gaming now? That's right, the Phenoms. Then the Phenom IIs were released and brought AMD pretty close to Intel in gaming. Sandy Bridge came out 2 years later and since then it's been a long, uphill battle for AMD. My guess is Steamroller and Excavator will be like the Phenom IIs where the current chips are like the Phenoms. And then Skylake will come out and be like Sandy Bridge. Same thing over and over again. History may not always repeat itself, but it's the only guide we got.

Typical mistake... in your analyses you got to separate what is due to software and what is due to hardware. In Games it depends a lot on drivers, and even so, the higher the resolution the less the CPU importance is... above 1980p the results are pretty identical on the same GPU, even with CPUS that have a notable gap in performance.

About intel uarch advantage another big mistake. You are seeing it by the side of "brand" preferences, and soon will be like a football/soccer game, and what matters is the result of the game, even if the penalty(bench soft) the referee marked wasn't penalty at all (review). If there is anything "Hasfail tock (not tick) " points is that actual intel uarch ( "core") is end-of-line. They must come with something much better in the future, more multithreading prone.

And i believe AMD will not abandon Cluster Multi-Threading and quite possibly Vertical Multi-Threading, not even after Excavator. So i tend to believe AMD has a much better uarch with this BD "line", more scalable, more flexible and easier to change, and much better for multi-threading "THE FUTURE" (else MPUs with higher number of cores doesn't make sense)...

Winning or losing on benchmarks is totally orthogonal, since this is much more due to software and its variations than the hardware... make me wrong, only takes to reveal the compiler used, flags of it used, and md5 (checksum) the bench soft, that is, to measure hardware properly then software must not change.

 

[hcl123]

Lol, If this FX 9530 (or what ever it is called) is priced at $400, I may as well just pick up a i7 3930K from microcenter. Or just buy a FX 8350 and OC it. I don't even know what AMD is doing anymore lol.

You don't know because when OC, which every enthusiast do, the chip doesn't fail right away. You can even grab any piece and quite OC it, if it fails, get lower and lower until stable. Yes you can pick any CPU and OC only its not "CERTIFIED", you can even lose the guaranties if you claim you have done it on a RMA issue (and this either intel or AMD).

The Fx9370 and 9590 are "CERTIFIED" to run at those speeds, it can even be the same as an 8350 in reality, but its certified. To do this "certification" the fab process must be very good for "high clocks" (even if it means OC above the usual TDP marks), otherwise is a considerable risk, prone to a flush of RMA, to do it ( that is the advantage AMD is trying to point... i think...)

 

[montosaurous]

Maybe I over emphasize the importance of a CPU because I've seen how big of a difference an FX 4100 can make going from a Core 2 Duo E4400 (with same GPU). CPU can make a huge difference in games, it's not always the GPU.

 

[The Q6660 Inside]

Eh I prefer to use AMD for their price/performance ratio. Haswell was a bit of a failure for a 'tick', I won't disagree with you there. And this thread isn't about consoles either. Lets face it: the days of the Athlon are over. Intel has a much larger R&D department and a superior architecture at the moment, it will take awhile for AMD to catch up. FX kinda does remind me of the Phenoms, and the i5s of Core 2 Duos. The C2D were better for gaming at first, but who's better for gaming now? That's right, the Phenoms. Then the Phenom IIs were released and brought AMD pretty close to Intel in gaming. Sandy Bridge came out 2 years later and since then it's been a long, uphill battle for AMD. My guess is Steamroller and Excavator will be like the Phenom IIs where the current chips are like the Phenoms. And then Skylake will come out and be like Sandy Bridge. Same thing over and over again. History may not always repeat itself, but it's the only guide we got.

Typical mistake... in your analyses you got to separate what is due to software and what is due to hardware. In Games it depends a lot on drivers, and even so, the higher the resolution the less the CPU importance is... above 1980p the results are pretty identical on the same GPU, even with CPUS that have a notable gap in performance.

About intel uarch advantage another big mistake. You are seeing it by the side of "brand" preferences, and soon will be like a football/soccer game, and what matters is the result of the game, even if the penalty(bench soft) the referee marked wasn't penalty at all (review). If there is anything "Hasfail tock (not tick) " points is that actual intel uarch ( "core") is end-of-line. They must come with something much better in the future, more multithreading prone.

And i believe AMD will not abandon Cluster Multi-Threading and quite possibly Vertical Multi-Threading, not even after Excavator. So i tend to believe AMD has a much better uarch with this BD "line", more scalable, more flexible and easier to change, and much better for multi-threading "THE FUTURE" (else MPUs with higher number of cores doesn't make sense)...

Winning or losing on benchmarks is totally orthogonal, since this is much more due to software and its variations than the hardware... make me wrong, only takes to reveal the compiler used, flags of it used, and md5 (checksum) the bench soft, that is, to measure hardware properly then software must not change.

The P6 uarch has been around for almost 18 years and is still used. AMD only recently gave up on the K/Athlon uarch. P6 still is fairing strong since its upgrade to Core/Core 2 and the smaller upgrades such as Haswell and Sandy Bridge.
But it may eventually reach its limit and that is what drove AMD to abandon 10h. How Intel will handle the Core/P6 uarch reaching its limit is a question we will have to wait to be answered.

 

[The Q6660 Inside]

Eh I prefer to use AMD for their price/performance ratio. Haswell was a bit of a failure for a 'tick', I won't disagree with you there. And this thread isn't about consoles either. Lets face it: the days of the Athlon are over. Intel has a much larger R&D department and a superior architecture at the moment, it will take awhile for AMD to catch up. FX kinda does remind me of the Phenoms, and the i5s of Core 2 Duos. The C2D were better for gaming at first, but who's better for gaming now? That's right, the Phenoms. Then the Phenom IIs were released and brought AMD pretty close to Intel in gaming. Sandy Bridge came out 2 years later and since then it's been a long, uphill battle for AMD. My guess is Steamroller and Excavator will be like the Phenom IIs where the current chips are like the Phenoms. And then Skylake will come out and be like Sandy Bridge. Same thing over and over again. History may not always repeat itself, but it's the only guide we got.

Typical mistake... in your analyses you got to separate what is due to software and what is due to hardware. In Games it depends a lot on drivers, and even so, the higher the resolution the less the CPU importance is... above 1980p the results are pretty identical on the same GPU, even with CPUS that have a notable gap in performance.

About intel uarch advantage another big mistake. You are seeing it by the side of "brand" preferences, and soon will be like a football/soccer game, and what matters is the result of the game, even if the penalty(bench soft) the referee marked wasn't penalty at all (review). If there is anything "Hasfail tock (not tick) " points is that actual intel uarch ( "core") is end-of-line. They must come with something much better in the future, more multithreading prone.

And i believe AMD will not abandon Cluster Multi-Threading and quite possibly Vertical Multi-Threading, not even after Excavator. So i tend to believe AMD has a much better uarch with this BD "line", more scalable, more flexible and easier to change, and much better for multi-threading "THE FUTURE" (else MPUs with higher number of cores doesn't make sense)...

Winning or losing on benchmarks is totally orthogonal, since this is much more due to software and its variations than the hardware... make me wrong, only takes to reveal the compiler used, flags of it used, and md5 (checksum) the bench soft, that is, to measure hardware properly then software must not change.

The P6 uarch has been around for almost 18 years and is still used. AMD only recently gave up on the K/Athlon uarch. P6 still is fairing strong since its upgrade to Core/Core 2 and the smaller upgrades such as Haswell and Sandy Bridge.
But it [strike]may eventually reach[/strike] is very close to its limit and that is what drove AMD to abandon its aging K based 10h for a completely new module based uarch. How Intel will handle the Core/P6 uarch reaching its limit is a question we will have to wait to be answered.

 

[griptwister]

OH MY GOODNESS!!! $960... AMD has got to be kidding me!!!

http://www.rwlabs.com/newsitem.php?id=6663

 

[hcl123]

Seems like a pre-order BS, every article points to the same vendor with nothing in stock

A classical game of RIP-OFF, which is usually what they do with intel, above $1000 pieces (brrr... insane!), and the FX (high clock) should land very close to the performance of HPDT of intel, specially on Linux (if not pass).

As stated in 2 articles, AMD pointed this for OEM only... for now... but if i was a vendor, i would buy some from those OEMs, inflate the price... yet get some good business going lol

 

[noob2222]

OH MY GOODNESS!!! $960... AMD has got to be kidding me!!!

http://www.rwlabs.com/newsitem.php?id=6663

I said it before. the phenom twkrs were given out for free and sold on ebay for $1,500 to 10s of thousands. if amd wants to make money on them this time, I say go for it. the problem is all these sites are acting like its going to be some easy to get part. quite the opposite, very few will pass the testing, even fewer will hit the stores.

think about it this way. how much did it cost AMD to test for the PII TWKR, package it, and ship them out. how much profit did those that received a twkr make vs how much profit AMD made?

 

[hcl123]

While I agree that x86 was designed for serial operations from onset. The innovations of HSA and hUMA are going in a new direction. I think Intel will ultimately have to capitulate to the ever increasing number of proponents of HSA and hUMA...

An Unified Memory Addressing/Access scheme, hardly will be exclusively at reach of AMD or HSA... intel will come with their own, that it will be incompatible( for sure at the hardware level).

Intel will never capitulate, they gain the most by having others capitulate and then spin & propaganda their products to the extreme.

Classical trend of "forward flight" upon a fail, if for some reason a fail is perceived, make the SKUs more expensive, spin it to the stratosphere... expect Hasfail to be more expensive in the future, its incredible but its true, they could even have a celeron level performance at $1000, yet make it the most sold compared with any competition, its not their fault, users like the brand no matter what, they only take a natural advantage.

 

[hcl123]

EDIT;

I said it before. the phenom twkrs were given out for free and sold on ebay for $1,500 to 10s of thousands. if amd wants to make money on them this time, I say go for it. the problem is all these sites are acting like its going to be some easy to get part. quite the opposite, very few will pass the testing, even fewer will hit the stores.

That is not what TPU and techreport suggests, this will be OEM not RETAIL... this will not be a commemorative chip inside a wood case. Just can't sell OEM in very low quantity, and also neither will go inside wooden cases.

They should have plenty, otherwise the FX8350 also will be diminute and the price couldn't be below $200. From 4.2Ghz to 5Ghz max turbo is not much, its ~20%, but the indicated TDP diff is 76% ) ... the retail version could be a 4.1/4.4Ghz inside the 140W TDP... plenty of them...

 

[jdwii]

Why would Amd do something so foolish sense that's true I’ll assume we won’t see a CPU update on piledriver this year, I’m really hating this whole mobile crap, i can personally care less about my laptop and having 12 hours of battery when I plug it in whenever I can all i want is a simple fast laptop that can play decent games at the screens resolution at med-high settings when it comes to desktops i want the best i can get for the money not the same thing every single year with a minor tweak. Anyways why does a laptop even need 12 hours of battery life? Plane rides (because every single person is on a plane all the time) i guess but in high school generally teachers would let me plug my laptop into an outlet to let it charge if needed and college students definitely can, and in the car its 20$ at most for a power invertor. My old A8 Llano can do 6-7 hours of basic use and can play a game for around 3 hours, this is easily enough for me and I would assume anyone else in the world besides people on plane rides. Kids at my college actually want more performance per dollar on laptops not so much on the battery in fact most of them play games and do everything else plugged in. When it comes to smartphones I understand and tablets (which no one is actually buying a windows tablet at BestBuy and my friends report the same at Staples)

 

[hcl123]

Why foolish ? ... intel has chips above 150W (they de-flate (edit) to lower 130-140W marks) with water blocks on the market. If one is foolish so is the other.

Foolish would be to have this on 1866Mhz memory. Perhaps that is the reason for OEM only (along with better served with water blocks like intel), 2400Mhz is not yet available on AMP Radeon brand... OEMs will consider all, most users wont, but this last ones would bitch louder than anyone else.

 

[anxiousinfusion]

AMD's "Berlin" server processors will see Steamroller cores, 1H 2014.

http://semiaccurate.com/2013/06/17/amd-announces-seattle-berlin-and-warsaw-for-servers/

 

[esrever]

steamroller FX doesn't seem to be in the making as of yet, the APU will be out this year likely.

 

[sarinaide]

The direction you speak of is coming, in my mind, I cannot see anyway that HSA does not contribute to a far more parallel capability of hardware. At which point is does become a coding language issue, though software will evolve. Adobe is already working on implementations for HSA in it's programs. I think there will be other companies doing the same to push HSA along even further.

Intel isn't ignoring HSA. Their drivers have been upgraded for Haswell and perform quite well now for compute. It's one of the more shocking things about Haswell. Their OpenCL performance of the HD4600 is faster than the A10-6800K.

http://www.tomshardware.com/reviews/core-i7-4770k-haswell-review,3521-3.html

Faster CPU, since HSA is not "really" in effect right now, the HSA Foundation only stipulated full adapted HSA based applications by 2014, starting with the New Adobe CS, since the existing Compute benches are still CPU driven that is why.

A10 5800k speeds up by 130% ... intel speeds up by 40% ... also essentially your comparing a dual module stripped down AMD to a full quad+HT Intel. not really a fair assumption just looking at the raw numbers.

The OpenCL GPU benchmarks are only using the embedded GPU. The HD4600 is on i3, i5 and i7 CPUs, so it is a fair comparison. It has CPU and combined scores mixed in but I was ignoring those due to the massive price difference.

A10 5800k speeds up by 130% ... intel speeds up by 40% ... also essentially your comparing a dual module stripped down AMD to a full quad+HT Intel. not really a fair assumption just looking at the raw numbers.

The OpenCL GPU benchmarks are only using the embedded GPU. The HD4600 is on i3, i5 and i7 CPUs, so it is a fair comparison. It has CPU and combined scores mixed in but I was ignoring those due to the massive price difference.

most of them aren't gpu only, most of it is cpu+gpu or cpu. The one that is IGP only is luxmark, even then their numbers look pretty fishy.

Compared to :

CPU+gpu is the only test here but has hasbeen 477-k:

So why is it toms test is the only one out there that has the 5800k slower than the 3770k on IGP only when the rest are nearly twice as fast, and even competetive to combined against the 4770k. toms showed it only 1/3 the power instead of 2445 to 2354.

In fact on GPU only tests, toms is the only one that showed even a remotely close race on igp, with Intel actually winning with the 3770k ... thats a big wtf?

Its because TH has a bad reputation amongst legitimate reviewers, personal biases warp results and frankly you never trust online reviews. Intel has been butthurt about their IGPU deficiencies for a long time, so its a chance to salvage face. I will post results soon but TH had Skyrim benches showing the HD4600 higher than it really achieves, I did a 1268x800(16 4xMSAA 4xAF and High settings managed around 60FPS lock skyrim has that is the HD packs included, the game is playable at 1600x900 with the same settings contrary to what Toms found. I also said a long time ago, Skyrim needs a GTX8800 maximum to be able to play well, since the HD7660D is about as good as a 8800/9800 and better in that it is DX11 supported the numbers given on low settings are essentially to low, and I don't know how they bypass Skyrim's 60FPS limiter.

 

[de5_Roy]

AMD's "Berlin" server processors will see Steamroller cores, 1H 2014.

http://semiaccurate.com/2013/06/17/amd-announces-seattle-berlin-and-warsaw-for-servers/

i have been waiting. so long. for something from amd about steamroller.
the speculation on the absence of 'high end' desktop(sr-fx) is mildly disappointing but it seems to be in line with amd's current execution. i didn't like comparisons with pd opteron and jaguar opteron-x vs berlin though. that's poor marketing tactics. woulda liked to see how berlin compares against llano, trinity powered opteron apus instead or kyoto/16 core pd.
berlin itself seems very promising, so i'm guessing kaveri will be quite good. overall very nice to see moar steamroller-related info out of amd. now i can't help wanting moar info on kaveri... even after seeing berlin's configuration (4c), i don't want to rule out 6 core and gddr5 (at least on the dt mobo...)....

 

[montosaurous]

Would be awesome if they released Steamy in line with the original roadmap: Q4 this year. Sadly though, it will probably be sometime in 2014, hopefully the first half. But if they could get Steamroller here in Q4 and Excavator here in Q4 2014 they'd really be rolling... Steamrolling Intel that is.

 

[esrever]

AMD's "Berlin" server processors will see Steamroller cores, 1H 2014.

http://semiaccurate.com/2013/06/17/amd-announces-seattle-berlin-and-warsaw-for-servers/

i have been waiting. so long. for something from amd about steamroller.
the speculation on the absence of 'high end' desktop(sr-fx) is mildly disappointing but it seems to be in line with amd's current execution. i didn't like comparisons with pd opteron and jaguar opteron-x vs berlin though. that's poor marketing tactics. woulda liked to see how berlin compares against llano, trinity powered opteron apus instead or kyoto/16 core pd.
berlin itself seems very promising, so i'm guessing kaveri will be quite good. overall very nice to see moar steamroller-related info out of amd. now i can't help wanting moar info on kaveri... even after seeing berlin's configuration (4c), i don't want to rule out 6 core and gddr5 (at least on the dt mobo...)....

berlin and kyoto are both APUs with is why AMD is likely comparing them. From what I read, Seatle looks to be just a cpu soc with no or very little gpu resource which are unneeded for microserver data centers. My guess for high end server cpus will be waiting for DDR4.

 

[griptwister]

steamroller FX doesn't seem to be in the making as of yet, the APU will be out this year likely.

We will see steamroller FX drop Q4 no doubt. AMD keeps saying that.

 

[esrever]

Whats interesting is AMD thinks A57 cores are about as powerful per core as jaguar while being more energy efficient. I wonder if its going on 20nm at TSMC or on Glofo's 28nm. I really wouldn't have expected A57s to perform as well as jaguar per core.

 

[sarinaide]

AMD have put OBR and Coolaler.com out of business, not only where leaks completely wrong, nobody is sourcing them parts anymore as I believe AMD send samples but observation to, then take it back when done and ensure the the data too, basically they have nothing to base results on so its just a thumbsuck.

Steamroller on AM3+ will only be out in the first quarter of 2014 this was said a long time back from AMD themselves, Kaveri will be out in Q4 and this is how AMD normally release new micro processors, Release the FM socket first, either way the Kaveri cores will be pitted against llano, trinity/richland cores and will give an indication as to how efficient SR's arch is so its not really that much of a catastrophe and to be honest I don't care about AM3+ anyways.

 

[gamerk316]

But they plan to use the cloud for tasks not affected by latency and bottleneck. Of course the real improvement in performance will be not the theoretical 4x, but the cloud will not slow down processing. Microsoft already showed demos of games using cloud.

Sorry, but the cloud is a joke. Nevermind how only 30% of the planet is actually connected to the internet, you still have a very large percentage of the country stuck on ASDL 768/1.5mbit, which is abysmally slow. Even at 50mbit like I get, you would have the CPU waiting for extended periods of time (Miliseconds!) doing NOTHING but waiting for data to come across the ethernet port.

Nevermind what happens when the cloud decides to go down for its first upgrade...

Intel is not member of the HSA foundation and has no plans for HSA support in any future chip, so far as I know.

OpenCL is not HSA. OpenCL is a mere interface to access the GPU in your computer for non-graphic tasks. HSA is about blurring the traditional distinction between CPU and GPU.

http://www.tomshardware.com/reviews/fusion-hsa-opencl-history,3262-8.html

Look at the PS4 APU. Most people think of it as a 8-core CPU plus a given Radeon GPU, but it is something new. There is no real distinction between CPU and GPU at the software level.

It is the programmer who splits the APU into a CPU and a GPU and this split is dynamical. You can use the CPU to assist the GPU or you can use the GPU to assist the CPU. This is very related to the claims that the PS4 has not bottlenecks.

By people who don't know any better. Remember: CPU's suck at parallel workloads, and GPU's suck at serial ones. At the end of the day, there is the distinction, and due to how different workloads are, you WILL always have one waiting on the other for some form of data.

In that regards, is where HSA has its weakpoint: Due to having a unified address space, software locks become a MUCH bigger performance bottleneck. No one is paying attention here yet, but I'm willing to bet those bottlenecks will offset any advantages of HSA over the long haul.

The irony is that some demos at E3 were running on real Xbox1 whereas others run in a high-end gaming PC with a GTX-780. This seems to confirm my prediction that the minimum PC for matching a next gen console is a i7 + GTX-780

But heres my point: If the games were optimised for AMD/AMD, why demo on Intel/NVIDIA? And if optimized for consoles then ported over to the PC, then why demo the PC version?

The answer, as I've TRIED to explain for years now, is that the PC is still the lead development platform, and always has been. The "console bottleneck" is, and has always been, a myth.

 

[gamerk316]

And for the hundredth time: I HATE the new quoting system; I can never figure out which quotemsg in a chain belongs to the post I'm trying to copy over.

Seriously: Everyone hates it. Fix it.