AMD CPU speculation... and expert conjecture

[hcl123]

i have some questions about amd and arm:
1. what kind of license did amd buy? on arm's website, it says amd has 'processor license' for cortex a57 and a53 cores. is it the same as architecture license or different?
if it's the same, that means amd can customize the cores to better integrate to their own ip blocks.
if it's not, then arch license holders like [strike]calxeda[/strike], nvidia, apple, qualcomm might have advantage.
2. what will the licensees do when arm raises their royalties? 😀 the way arm has operated so far makes them look almost like a charity (amd has nothing on them).
3. is amd among arm's 'chosen' 3 partners for arm v8 testing?

since we were discussing roadmaps, i want to see if this following bit contradicts others:
in the earlier 'opteron' roadmap, amd showed that steamroller is on track for 2013, right? then in a later roadmap, amd showed that 'berlin' apu/cpu will have steamroller core in them. then amd is technically 'by the roadmap', they did debut steamroller in 2013. or did you guys expect something else instead of 'warsaw'?

edit: not sure about calxeda.

AMD has licensed the core, not the ISA. This means AMD cannot change the cores. This doesn't mean that AMD ARM SoC will be as everyone else SoC. AMD will incorporate custom stuff such as Freedom Fabric in its ARM SoC, offering advantages over competitors.

The reason why AMD didn't license the ISA is because developing a custom ARM core will take time and money. Nvidia did license ISA but its custom ARM cores are announced for 2015 (if there is no delays).

AMD will license ISA in a second phase and will develop its own ARM cores, but not today.

There is nothing else instead Warsaw. I don't understand why people insist on ignoring official roadmaps.

Steamroller is still late because it was suppose to be read Q3 this year by the original roadmap. Warsaw will probably be piledriver, AMD would have advertised it in the roadmap otherwise.

AMD has confirmed that Warsaw is Piledriver. AMD explained why and I also did.

Oh! you are wrong... either AMD as well Nvidia are "architectural licensees". They can change the not only the cores if they wish (edt), but also the SOCs.

I bet you, that either AMD and Nvidia have different cache arbitration than ARM reference designs, and AMD for sure could have different "pre-fetch" schemes than ARM references.

All is based on "core" definitions" of ARM of course, but none SoC of ARM has 10Gb Ethernet links on SoC, neither Freedom Fabric on SoC... and Nviida 4+1 CPU scheme of Tegra is not big.LITTLE either, and the GPU is quite different of Mali...

Different SoCs, different interconnects, different caches hierarchies, front-ends and back-ends of "cores" slightly different, enough to accommodate all those differences.

 

[hcl123]

No, he is doing what he always does...presenting skewed results that favor Intel architecture...his unofficial title is "Master of Misinformation" for a reason you know...

I think he is a professional... all it takes to deal with him is the ***TRUTH***

Windows Benchmarks are "optimized" for Intel, but are *NOT REPRESENTATIVE* of the software ppl use.period

Its a "curiosity" to say a mice thing... and he could fill another 140 pages of biased benchmarks that in the end, in REALITY, it means nothing.... but funny how he manages to derail the thread every time with his campaigns.

Answer to him only takes the 2 lines on bold everytime... instead ppl behave like suckers, suckered into polemics, politics and pointless bickering, where he has the advantage of "fake" graphics" (meaningless) spread allover.

In the end who said biased advertising and propaganda can't come to IT, even if there isn't almost any on TV or radio or newspapers ??... (edt)

... nowadays ppl complain about the traditional news outlets, specially dealing with "politics" news and their obvious lies... the presstitutes the pressluts... but i think and quite right, the Internet is much much worst, you have to be extra careful where you go, what you pick, what you believe (quite more easy an cheap to develop an elaborate scam by internet, than by TV, radio or newspapers)

 

[8350rocks]

I finally have word back from someone inside AMD, all I got in response to my query was this..."A relevant announcement regarding your question will be coming in the month of October. We will have news to share in September on a more unrelated note as well."

So...I think some of this speculation may be "jumping the gun" so to speak.

 

[gamerk316]

http://www.techpowerup.com/190493/amd-details-embedded-product-roadmap.html

http://www.techpowerup.com/img/13-09-09/slide-10-1024.jpg
oh look ARM is not replacing jaguar

Not until there is no more need for x86... until there is a *market for x86... and this is mostly dictated by "software ecosystem" than uarch.

Long time (if ever of course) before x86 takes a definite back seat(edt)... but i'm convinced that in the end it will (take a back seat), though in the "path" its not based on technical issues its based on market.

Too much legacy SW, too hard to properly emulate. I don't see X86 going away any time soon, even though its one of the worst designed architectures out there.

 

[juanrga]

The techreport link shows an i7-3770k ~ i5-3570k and i7-2600k ~ i5-2500k. If a 8 threads chip perform as a 4 threads chip, then that is not a true multitasking test. With true multitasking benchmarks one obtains FX-8350 > i7 > i5.

Translation: Any benchmark the 8350 doesn't win isn't valid.

Your translation is incorrect because you didn't pay enough attention to the part that you snipped:

The bit-tech link is another joke. Besides showing a Haswell i5 being slower than a two generation old Sandy Bride i5 (LOL), the imagined 'multitasking' benchmark puts an i5-3570k (4 threads) as being faster than an i7-4770k (8 threads) and this last one faster than a i7-3960X (12 threads).

Really? a $1000 12 threads chip being much slower than a $200 4 threads chip in a 'multitasking' test? LOL The multitasking is only in the headline.

With a real multitasking workload, the eXtreme 12 threads chip is the faster and then the FX-8350 > i7 > i5.

If a $1000 Intel i7-X chip (12 threads) performs poor than a $200 i5 chip (4 threads) then the problem is on the software used. Change the software and the 12 threads Intel chip will outperform the 4 threads Intel chip.

It is not AMD vs Intel, but optimized/modern software vs unoptimized/old.

to launch or not to launch, that is the question
http://www.fudzilla.com/home/item/32449-desktop-kaveri-to-show-up-in-q4?
apparently amd can't read their own roadmap. 😆

AMD (and members of industry confirmed) claims they are following the roadmap. Desktop Kaveri ships to customers (OEMs) at end of 2013 and will be available to public at early 2014. Mobile Kaveri comes some months latter.

 

[blackkstar]

No, he is doing what he always does...presenting skewed results that favor Intel architecture...his unofficial title is "Master of Misinformation" for a reason you know...

I think he is a professional... all it takes to deal with him is the ***TRUTH***

Windows Benchmarks are "optimized" for Intel, but are *NOT REPRESENTATIVE* of the software ppl use.period

Its a "curiosity" to say a mice thing... and he could fill another 140 pages of biased benchmarks that in the end, in REALITY, it means nothing.... but funny how he manages to derail the thread every time with his campaigns.

Answer to him only takes the 2 lines on bold everytime... instead ppl behave like suckers, suckered into polemics, politics and pointless bickering, where he has the advantage of "fake" graphics" (meaningless) spread allover.

In the end who said biased advertising and propaganda can't come to IT, even if there isn't almost any on TV or radio or newspapers ??... (edt)

... nowadays ppl complain about the traditional news outlets, specially dealing with "politics" news and their obvious lies... the presstitutes the pressluts... but i think and quite right, the Internet is much much worst, you have to be extra careful where you go, what you pick, what you believe (quite more easy an cheap to develop an elaborate scam by internet, than by TV, radio or newspapers)

Why are you the only other guy that gets that just because FX loses in Windows it isn't a slower chip everywhere.

A friendly reminder to some of you that my FX 8350 can beat 3930k CLOCK FOR CLOCK in some benchmarks when FX is getting optimized code and 3930k is running on Windows.

If AMD really wanted to destroy the competition, they'd be releasing Jaguar chips that run some sort of *nix or Android compiled to take advantage of all the Jaguar instructions. It would be a bloodbath, specially considering Atom doesn't have nearly as many instructions as Jaguar (and I'm talking about the new Atom, not the old one).

Also Hajifur and gamerk, why are you spending so much time fighting "MOAR COARS!" when people still find a reason to buy 3930k and greater? If MOAR COARS didn't work at all, Intel hex wouldn't be on top of the charts so much, specially considering that it's basically Sandy Bridge competing with Haswell at similar clockspeeds. I do feel that AMD's MOAR COARS problem is much bigger than the fact that software isn't using those cores, and that it's along the lines of Piledriver still struggling.

Intel hex winning composite scores like this shows me that the majority of benchmarks used in those compilations means that there's enough software that scales to 4+ cores in the categories where Intel hex wins to make up for Sandy Bridge's IPC deficit against Haswell.

Things like gaming has the 4770k in the lead, and it's because gaming isn't threaded as well or that perhaps the benchmarks used were older games which don't thread so well. But regardless it points me to believe that multi-threading in games is behind the rest of the curve in software heading towards multi-thread.

 

[juanrga]

I am using the following definitions (from the above links):

Uniform Memory Access (UMA) is a shared memory architecture used in parallel computers. All the processors in the UMA model share the physical memory uniformly. In a UMA architecture, access time to a memory location is independent of which processor makes the request or which memory chip contains the transferred data. Uniform Memory Access computer architectures are often contrasted with Non-Uniform Memory Access (NUMA) architectures.

What if more than one processor in a MPU (example 4 cores) makes a memory request at the same time ?

Sure happens all the time... and the answer is; "its arbitrated" ... the IMC (Integrated Memory Controller) has queues for requests

You continue misunderstanding the definition. Imagine that processor A makes an early request to a given address and the access time is T ns, and processor B access time is T+D ns, with D being the delay due to the shared resource being busy with the other request. The inverse case is also possible: processor B making an early request to a given address and the access time being T ns, whereas it is T+D ns for processor A.

The point is that both A and B have equivalent access to memory in equivalent situations, because there is no topological distinction in the memory model. The possible delay caused to A or B by busy resources doesn't split the physical memory into local plus remote.

MORE THAN OBVIOUS is you who don't lol

All that is very dandy... only you ***NEVER*** know what "T" is, until you probe. The access can go to a "remote" DRAM location, can go to disk, can go to network. "T" is a *wild* variable its not a constant.. and its so even for a UMA scheme.

You are only making your own interpretation of NUMA, based on wrong "assumptions"... the definition is not based on the "amount of time", the different "amount of time" is guarantied exactly by the hierarchical memory view of the arbitration of NUMA, while in UMA the access time is kind of flat, ~equal for every processing element of a MPU, if there isn't contention or disk going. And that is exactly what is in wikipedia.(they skip complicating with contention or disk/network going explanations, because wikipedia is supposed to be for everybody... the simpler the better...)(edt)

Funny way to reply without actually reply. LOL

In the first place, it is not Wikipedia thing for "everybody", but standard definition of UMA and NUMA

http://en.wikipedia.org/wiki/Non-Uniform_Memory_Access
http://technet.microsoft.com/en-us/library/ms178144%28v=sql.105%29.aspx
http://lse.sourceforge.net/numa/faq/

In the second place, T is not for accessing disk/network...

In the third place nobody said you that T is a constant, because it is not.

In the fourth place, do you still pretend that anyone can do a CPU+dGPU with UMA/hUMA?

 

[juanrga]

i have some questions about amd and arm:
1. what kind of license did amd buy? on arm's website, it says amd has 'processor license' for cortex a57 and a53 cores. is it the same as architecture license or different?
if it's the same, that means amd can customize the cores to better integrate to their own ip blocks.
if it's not, then arch license holders like [strike]calxeda[/strike], nvidia, apple, qualcomm might have advantage.
2. what will the licensees do when arm raises their royalties? 😀 the way arm has operated so far makes them look almost like a charity (amd has nothing on them).
3. is amd among arm's 'chosen' 3 partners for arm v8 testing?

since we were discussing roadmaps, i want to see if this following bit contradicts others:
in the earlier 'opteron' roadmap, amd showed that steamroller is on track for 2013, right? then in a later roadmap, amd showed that 'berlin' apu/cpu will have steamroller core in them. then amd is technically 'by the roadmap', they did debut steamroller in 2013. or did you guys expect something else instead of 'warsaw'?

edit: not sure about calxeda.

AMD has licensed the core, not the ISA. This means AMD cannot change the cores. This doesn't mean that AMD ARM SoC will be as everyone else SoC. AMD will incorporate custom stuff such as Freedom Fabric in its ARM SoC, offering advantages over competitors.

The reason why AMD didn't license the ISA is because developing a custom ARM core will take time and money. Nvidia did license ISA but its custom ARM cores are announced for 2015 (if there is no delays).

AMD will license ISA in a second phase and will develop its own ARM cores, but not today.

There is nothing else instead Warsaw. I don't understand why people insist on ignoring official roadmaps.

Steamroller is still late because it was suppose to be read Q3 this year by the original roadmap. Warsaw will probably be piledriver, AMD would have advertised it in the roadmap otherwise.

AMD has confirmed that Warsaw is Piledriver. AMD explained why and I also did.

Oh! you are wrong... either AMD as well Nvidia are "architectural licensees". They can change the not only the cores if they wish (edt), but also the SOCs.

I bet you, that either AMD and Nvidia have different cache arbitration than ARM reference designs, and AMD for sure could have different "pre-fetch" schemes than ARM references.

All is based on "core" definitions" of ARM of course, but none SoC of ARM has 10Gb Ethernet links on SoC, neither Freedom Fabric on SoC... and Nviida 4+1 CPU scheme of Tegra is not big.LITTLE either, and the GPU is quite different of Mali...

Different SoCs, different interconnects, different caches hierarchies, front-ends and back-ends of "cores" slightly different, enough to accommodate all those differences.

The information that I have is that AMD got a core license, but not an ISA license. All ARM products announced by AMD use standard ARM cores. If you have other information, please put here the link.

Also I have said before that Nvidia has an ISA license and can modify the cores. I don't understand why you go to say the same to me.

 

[juanrga]

I think he is a professional... all it takes to deal with him is the ***TRUTH***

Windows Benchmarks are "optimized" for Intel, but are *NOT REPRESENTATIVE* of the software ppl use.period

Its a "curiosity" to say a mice thing... and he could fill another 140 pages of biased benchmarks that in the end, in REALITY, it means nothing.... but funny how he manages to derail the thread every time with his campaigns.

Answer to him only takes the 2 lines on bold everytime... instead ppl behave like suckers, suckered into polemics, politics and pointless bickering, where he has the advantage of "fake" graphics" (meaningless) spread allover.

In the end who said biased advertising and propaganda can't come to IT, even if there isn't almost any on TV or radio or newspapers ??... (edt)

... nowadays ppl complain about the traditional news outlets, specially dealing with "politics" news and their obvious lies... the presstitutes the pressluts... but i think and quite right, the Internet is much much worst, you have to be extra careful where you go, what you pick, what you believe (quite more easy an cheap to develop an elaborate scam by internet, than by TV, radio or newspapers)

Why are you the only other guy that gets that just because FX loses in Windows it isn't a slower chip everywhere.

A friendly reminder to some of you that my FX 8350 can beat 3930k CLOCK FOR CLOCK in some benchmarks when FX is getting optimized code and 3930k is running on Windows.

If AMD really wanted to destroy the competition, they'd be releasing Jaguar chips that run some sort of *nix or Android compiled to take advantage of all the Jaguar instructions. It would be a bloodbath, specially considering Atom doesn't have nearly as many instructions as Jaguar (and I'm talking about the new Atom, not the old one).

Also Hajifur and gamerk, why are you spending so much time fighting "MOAR COARS!" when people still find a reason to buy 3930k and greater? If MOAR COARS didn't work at all, Intel hex wouldn't be on top of the charts so much, specially considering that it's basically Sandy Bridge competing with Haswell at similar clockspeeds. I do feel that AMD's MOAR COARS problem is much bigger than the fact that software isn't using those cores, and that it's along the lines of Piledriver still struggling.

Intel hex winning composite scores like this shows me that the majority of benchmarks used in those compilations means that there's enough software that scales to 4+ cores in the categories where Intel hex wins to make up for Sandy Bridge's IPC deficit against Haswell.

Things like gaming has the 4770k in the lead, and it's because gaming isn't threaded as well or that perhaps the benchmarks used were older games which don't thread so well. But regardless it points me to believe that multi-threading in games is behind the rest of the curve in software heading towards multi-thread.

It seems evident that some people don't want accept facts. As AMD representatives have stated recently Microsoft ecosystem is against innovation. AMD chips run fine under linux, because software is well multi-threaded.

What I find really funny is that some Intel fans fight the "moar cores", when Intel has announced an octo core Haswell. I find it also interesting that some fans believe that one sane person will buy a $1000 six-core Intel chip to run a single thread application or something as superpi. LOL

Also everyone in the industry knows that next gen games will be well multithreaded. Only AMD haters seem to deny this.

 

[juanrga]

blackkstar you do realise the i7 3930k SB is about 80% faster clock for clock then a piledriver cpu don't you. The fx8350 won't beat the 3930k at anything relevant especially 100% cpu it will get destroyed. fx8350 has less performance then an i7 2700k or 3770k or 4770k at stock in most multithreaded tasks. What you have to understand is intel i7's are barely getting used in games but the better single threaded performance and memory bandwidth gets the higher fps in games.

No. In multithreaded tasks (8+ threads) the FX-8350 is faster than i7-2700k, i7-3770k, and i7-4770k. And must be behind the i7-3930k. The FX-9590 would be faster than the i7-3930k.

 

[gamerk316]

No, he is doing what he always does...presenting skewed results that favor Intel architecture...his unofficial title is "Master of Misinformation" for a reason you know...

I think he is a professional... all it takes to deal with him is the ***TRUTH***

Windows Benchmarks are "optimized" for Intel, but are *NOT REPRESENTATIVE* of the software ppl use.period

Its a "curiosity" to say a mice thing... and he could fill another 140 pages of biased benchmarks that in the end, in REALITY, it means nothing.... but funny how he manages to derail the thread every time with his campaigns.

Answer to him only takes the 2 lines on bold everytime... instead ppl behave like suckers, suckered into polemics, politics and pointless bickering, where he has the advantage of "fake" graphics" (meaningless) spread allover.

In the end who said biased advertising and propaganda can't come to IT, even if there isn't almost any on TV or radio or newspapers ??... (edt)

... nowadays ppl complain about the traditional news outlets, specially dealing with "politics" news and their obvious lies... the presstitutes the pressluts... but i think and quite right, the Internet is much much worst, you have to be extra careful where you go, what you pick, what you believe (quite more easy an cheap to develop an elaborate scam by internet, than by TV, radio or newspapers)

Why are you the only other guy that gets that just because FX loses in Windows it isn't a slower chip everywhere.

A friendly reminder to some of you that my FX 8350 can beat 3930k CLOCK FOR CLOCK in some benchmarks when FX is getting optimized code and 3930k is running on Windows.

If AMD really wanted to destroy the competition, they'd be releasing Jaguar chips that run some sort of *nix or Android compiled to take advantage of all the Jaguar instructions. It would be a bloodbath, specially considering Atom doesn't have nearly as many instructions as Jaguar (and I'm talking about the new Atom, not the old one).

Also Hajifur and gamerk, why are you spending so much time fighting "MOAR COARS!" when people still find a reason to buy 3930k and greater? If MOAR COARS didn't work at all, Intel hex wouldn't be on top of the charts so much, specially considering that it's basically Sandy Bridge competing with Haswell at similar clockspeeds. I do feel that AMD's MOAR COARS problem is much bigger than the fact that software isn't using those cores, and that it's along the lines of Piledriver still struggling.

Intel hex winning composite scores like this shows me that the majority of benchmarks used in those compilations means that there's enough software that scales to 4+ cores in the categories where Intel hex wins to make up for Sandy Bridge's IPC deficit against Haswell.

Things like gaming has the 4770k in the lead, and it's because gaming isn't threaded as well or that perhaps the benchmarks used were older games which don't thread so well. But regardless it points me to believe that multi-threading in games is behind the rest of the curve in software heading towards multi-thread.

blackkstar you do realise the i7 3930k SB is about 80% faster clock for clock then a piledriver cpu don't you. The fx8350 won't beat the 3930k at anything relevant especially 100% cpu it will get destroyed. fx8350 has less performance then an i7 2700k or 3770k or 4770k at stock in most multithreaded tasks. What you have to understand is intel i7's are barely getting used in games but the better single threaded performance and memory bandwidth gets the higher fps in games.

No. In multithreaded tasks (8+ threads) the FX-8350 is faster than i7-2700k, i7-3770k, and i7-4770k. And must be behind the i7-3930k. The FX-9590 would be faster than the i7-3930k.

FX never beats IB clock-for-clock; that would indicate FX wins in single-threaded benchmarks as well. Clock both at the same speed, use the same number of cores, and IB always wins.

Secondly, again, as I've noted for four years now: You can not just "make" software into a parallel multithreaded design. That is the reason you still typically see two main threads (main app thread + main render thread), and maybe 1-2 extra threads to offload some processing. This isn't going to change any time soon, for reasons I've explained over and over and over again. Games do not naturally scale to multiple cores. The parts of the engine that do (rendering + physics) are already offloaded to the GPU, leaving the serial processing to the CPU. That favors Intel's approach.

blackkstar you do realise the i7 3930k SB is about 80% faster clock for clock then a piledriver cpu don't you. The fx8350 won't beat the 3930k at anything relevant especially 100% cpu it will get destroyed. fx8350 has less performance then an i7 2700k or 3770k or 4770k at stock in most multithreaded tasks. What you have to understand is intel i7's are barely getting used in games but the better single threaded performance and memory bandwidth gets the higher fps in games.

No. In multithreaded tasks (8+ threads) the FX-8350 is faster than i7-2700k, i7-3770k, and i7-4770k. And must be behind the i7-3930k. The FX-9590 would be faster than the i7-3930k.

As I've said since before BD released: In the few cases where all 8 cores get used, it will win against Intel. In all other cases, AMD looses.

The clock-for-clock advantage in favor of Intel looks to be in the 30-40% range currently, though this is offset somewhat by AMD's higher baseline clock speed.

The bit-tech link is another joke. Besides showing a Haswell i5 being slower than a two generation old Sandy Bride i5 (LOL), the imagined 'multitasking' benchmark puts an i5-3570k (4 threads) as being faster than an i7-4770k (8 threads) and this last one faster than a i7-3960X (12 threads).

Really? a $1000 12 threads chip being much slower than a $200 4 threads chip in a 'multitasking' test? LOL The multitasking is only in the headline.

With a real multitasking workload, the eXtreme 12 threads chip is the faster and then the FX-8350 > i7 > i5.

Serial software favors the higher base clock of the 3570k. If you load to 4 cores, i would expect the faster chip to win, even if its a generation older. The 3570k is clocked at 3.4, and 3960x at 3.3. The 4770k loosing to the 3570k is interesting; possibly negative result of HTT, slightly different cache layout (cache sensitive app?), poor optimization, or other factors. Can't say without looking at what the code does. but the 3570k winning over the slower clocked 3960x does make sense if you don't scale beyond four cores.

Which, BTW, is AMD's problem with FX: They made the assumption that devs could simply scale up games to use more cores. And as devs are starting to find out, as the rest of us developers did back in the 70's and 80's, that isn't normally the case.

I'd go so far as to say the current generation of console will hold back PC gaming for the next decade, due to being designed around weak, individual cores, and developers finding they can't get games to scale well. So we'll get a decade of GPU bound titles filled with more and more graphical fluff, with very little innovation in gameplay, design, physics, audio, or anything else. I'm predicting a decade of stagnation in gaming.

 

[noob2222]

I have shown before that companies that rely on speed don't give a sh@@ about low power junk.

I fully understand that a 8-16 core arm cpu is going to be 2-4x as powerful as a 2-4 core x86 cpu. why are you so adamant that the number of cores means nothing?

aside from that, I love your hints that you have secret information that your sharing openly. as I stated, if what you claim is true, AMD is going to suck bad unless HSA is adopted within the next year. 64bit isn't eve fully utilized 10 years later, how long till HSA is mainstream? that how long AMD will suck hind teat if what you claim is true.

as for ARM, I have seen nothing that gets me excited about the architecture. sure, works somewhar ok in my phone aside from all the app crashes, os lockups, and general slow responce. somehow this is going to "dominate the world" without the support of discrete gpus and being as fast as the slowest x86 cpu available per core, aka via nano.

yes everyone wants to see intel's stranglehold broken, but I don't see any evidence that ARM will work on a desktop environment.

Seattle is one of fastest chips made by AMD. I estimate the 16 thread version will be at the same level of performance than a 16 threads Xeon E5-2650 (which is one of the fastest chips made by Intel and used in some of the fastest and efficient supercomputers in the world).

Calling that low power junk is funny.

Who said you that the number of cores means nothing?

The rest of your posts was replied before. I already explained why adoption of HSA will not be so tortuous and slow as was the adoption of AMD64.

http://www.techpowerup.com/190493/amd-details-embedded-product-roadmap.html

oh look ARM is not replacing jaguar

The claim was that AMD will replace jaguar servers by ARM servers the next year.

In the embedded market jaguar will be replaced by its successor "enhanced jaguar", because the whole market will be not ready for switching from x86 to ARM in only one year. That is also the reason why AMD will release Warsaw for servers. Warsaw is aimed to big corporations and other clients that will be slow on moving from x86 servers to ARM servers.

In the long run ARM will replace x86 in supercomputers, desktops... both AMD and Nvidia have claimed so and I agree.

AMD places the ARM HieroFalcon chip in the High-Performance tie, whereas jaguar and its successor are not. I find it funny that some AMD haters continue to say that AMD ARM chips are low power junk. LOL

Where is your proof that ARM is competetive? I have shown you already that 1.5ghz Kabini is 2x as fast as samsung's 2.0ghz A15. Also funny in the embedded roadmap, E-Jaguar uses 5-25W while the ARM is 15-30W. I showed Anand and toms tests of atom vs A9, with the A9 losing by a mile in power consumption.

Where is your proof that ARM is the best, fastest, most stable thing ever created by man? Your words are not proof, half of the stuff you say is barely believeable. \/ \/

The information that I have is that AMD got a core license, but not an ISA license. All ARM products announced by AMD use standard ARM cores. If you have other information, please put here the link.

Also I have said before that Nvidia has an ISA license and can modify the cores. I don't understand why you go to say the same to me.

So basically what your trying to say here is that you work for AMD and are disclosing NDA information without any proof, but anyone that argues with you has to offer proof?

As if we are supposed to just take your word for it.

 

[christoffe9311]

why is everyone going on about the stuff no ones cares about, non of the parts that are being mentioned are performance parts, i don't even know why amd are putting the steam roller cores in a apu first they work back to front, who even uses a apu seriously. They should sort out main steam part should be top priority.

 

[griptwister]

Lol, looks like Nvidia is attempting to compete with AMD's TessFX

http://www.overclockarena.com/the-witcher-3%E2%80%B2s-fur-tech-will-most-probably-bring-your-gpus-to-their-knees/

 

[8350rocks]

No, he is doing what he always does...presenting skewed results that favor Intel architecture...his unofficial title is "Master of Misinformation" for a reason you know...

I think he is a professional... all it takes to deal with him is the ***TRUTH***

Windows Benchmarks are "optimized" for Intel, but are *NOT REPRESENTATIVE* of the software ppl use.period

Its a "curiosity" to say a mice thing... and he could fill another 140 pages of biased benchmarks that in the end, in REALITY, it means nothing.... but funny how he manages to derail the thread every time with his campaigns.

Answer to him only takes the 2 lines on bold everytime... instead ppl behave like suckers, suckered into polemics, politics and pointless bickering, where he has the advantage of "fake" graphics" (meaningless) spread allover.

In the end who said biased advertising and propaganda can't come to IT, even if there isn't almost any on TV or radio or newspapers ??... (edt)

... nowadays ppl complain about the traditional news outlets, specially dealing with "politics" news and their obvious lies... the presstitutes the pressluts... but i think and quite right, the Internet is much much worst, you have to be extra careful where you go, what you pick, what you believe (quite more easy an cheap to develop an elaborate scam by internet, than by TV, radio or newspapers)

Why are you the only other guy that gets that just because FX loses in Windows it isn't a slower chip everywhere.

A friendly reminder to some of you that my FX 8350 can beat 3930k CLOCK FOR CLOCK in some benchmarks when FX is getting optimized code and 3930k is running on Windows.

If AMD really wanted to destroy the competition, they'd be releasing Jaguar chips that run some sort of *nix or Android compiled to take advantage of all the Jaguar instructions. It would be a bloodbath, specially considering Atom doesn't have nearly as many instructions as Jaguar (and I'm talking about the new Atom, not the old one).

Also Hajifur and gamerk, why are you spending so much time fighting "MOAR COARS!" when people still find a reason to buy 3930k and greater? If MOAR COARS didn't work at all, Intel hex wouldn't be on top of the charts so much, specially considering that it's basically Sandy Bridge competing with Haswell at similar clockspeeds. I do feel that AMD's MOAR COARS problem is much bigger than the fact that software isn't using those cores, and that it's along the lines of Piledriver still struggling.

Intel hex winning composite scores like this shows me that the majority of benchmarks used in those compilations means that there's enough software that scales to 4+ cores in the categories where Intel hex wins to make up for Sandy Bridge's IPC deficit against Haswell.

Things like gaming has the 4770k in the lead, and it's because gaming isn't threaded as well or that perhaps the benchmarks used were older games which don't thread so well. But regardless it points me to believe that multi-threading in games is behind the rest of the curve in software heading towards multi-thread.

You're preaching to the converted here...I gave up on trying to get the Intel nuts to look at Linux benchmarks...though there are plenty where the FX series really shines against anything Intel.

That discussion was hashed out somewhere around pages 80-90 IIRC.

Their argument was..."what does the world run on?" When I pointed out that was android...their next argument was "what do the AAA title games run on?"

 

[juanrga]

No. In multithreaded tasks (8+ threads) the FX-8350 is faster than i7-2700k, i7-3770k, and i7-4770k. And must be behind the i7-3930k. The FX-9590 would be faster than the i7-3930k.

FX never beats IB clock-for-clock; that would indicate FX wins in single-threaded benchmarks as well. Clock both at the same speed, use the same number of cores, and IB always wins.

The first part "FX never beats IB clock-for-clock" is incorrect. The last part "Clock both at the same speed, use the same number of cores, and IB always wins" is correct. However this is also true for Intel. Take an i7 eXtreme chip disable four cores and downclock and it will run like an i3.

The question here is that no owner of an $1000 i7 eXtreme chip will be disabling cores and downclocking. No owner of a FX-8350 will be disabling four cores and downclocking at 3.5GHz. The owners of those chips will be using the native clocks or even overclocking and using software that utilizes all the cores.

Which, BTW, is AMD's problem with FX: They made the assumption that devs could simply scale up games to use more cores. And as devs are starting to find out, as the rest of us developers did back in the 70's and 80's, that isn't normally the case.

I'd go so far as to say the current generation of console will hold back PC gaming for the next decade, due to being designed around weak, individual cores, and developers finding they can't get games to scale well. So we'll get a decade of GPU bound titles filled with more and more graphical fluff, with very little innovation in gameplay, design, physics, audio, or anything else. I'm predicting a decade of stagnation in gaming.

Today the main game engines scale well. I already provided you a copy of a cpu profiling of a future PS4 game. The game uses six-cores for running six game threads. And the OS and the dev tools (including the profiler) run in the remaining two cores. A total of 8 cores working fine.

I also provided you quotes from game developers that prefer "moar cores":

Digital Foundry: Let's talk about next-gen console. What's your take on the general design in terms of CPU and graphics processing power?

Oles Shishkovstov: We are talking PS4, right? I am very excited about both CPU and GPU. Jaguar is a pretty well-balanced out-of-order core and there are eight of them inside. I always wanted a lot of relatively-low-power cores instead of single super-high-performance one, because it's easier to simply parallelise something instead of changing core-algorithms or chasing every cycle inside critical code segment (not that we don't do that, but very often we can avoid it).

I reject your pessimistic vision. The entire game industry has applauded the next gen consoles (specially the PS4) and how they will push games to a superior level. Even Nvidia agrees that new consoles will benefit PCs

http://www.techspot.com/news/52881-nvidias-tony-tamasi-says-new-consoles-are-good-news-for-pc-gamers.html

I have shown before that companies that rely on speed don't give a sh@@ about low power junk.

I fully understand that a 8-16 core arm cpu is going to be 2-4x as powerful as a 2-4 core x86 cpu. why are you so adamant that the number of cores means nothing?

aside from that, I love your hints that you have secret information that your sharing openly. as I stated, if what you claim is true, AMD is going to suck bad unless HSA is adopted within the next year. 64bit isn't eve fully utilized 10 years later, how long till HSA is mainstream? that how long AMD will suck hind teat if what you claim is true.

as for ARM, I have seen nothing that gets me excited about the architecture. sure, works somewhar ok in my phone aside from all the app crashes, os lockups, and general slow responce. somehow this is going to "dominate the world" without the support of discrete gpus and being as fast as the slowest x86 cpu available per core, aka via nano.

yes everyone wants to see intel's stranglehold broken, but I don't see any evidence that ARM will work on a desktop environment.

Seattle is one of fastest chips made by AMD. I estimate the 16 thread version will be at the same level of performance than a 16 threads Xeon E5-2650 (which is one of the fastest chips made by Intel and used in some of the fastest and efficient supercomputers in the world).

Calling that low power junk is funny.

Who said you that the number of cores means nothing?

The rest of your posts was replied before. I already explained why adoption of HSA will not be so tortuous and slow as was the adoption of AMD64.

http://www.techpowerup.com/190493/amd-details-embedded-product-roadmap.html

oh look ARM is not replacing jaguar

The claim was that AMD will replace jaguar servers by ARM servers the next year.

In the embedded market jaguar will be replaced by its successor "enhanced jaguar", because the whole market will be not ready for switching from x86 to ARM in only one year. That is also the reason why AMD will release Warsaw for servers. Warsaw is aimed to big corporations and other clients that will be slow on moving from x86 servers to ARM servers.

In the long run ARM will replace x86 in supercomputers, desktops... both AMD and Nvidia have claimed so and I agree.

AMD places the ARM HieroFalcon chip in the High-Performance tie, whereas jaguar and its successor are not. I find it funny that some AMD haters continue to say that AMD ARM chips are low power junk. LOL

Where is your proof that ARM is competetive? I have shown you already that 1.5ghz Kabini is 2x as fast as samsung's 2.0ghz A15. Also funny in the embedded roadmap, E-Jaguar uses 5-25W while the ARM is 15-30W. I showed Anand and toms tests of atom vs A9, with the A9 losing by a mile in power consumption.

Where is your proof that ARM is the best, fastest, most stable thing ever created by man? Your words are not proof, half of the stuff you say is barely believeable. \/ \/

It is based in AMD decision to substitute Opteron X for Seattle. Opteron X is faster than Kabini and AMD claims Seattle will be 2x-4x faster than Opteron X, but with improved efficiency. Also you avoid what reviewers, including Anand, are saying about AMD ARM chip.

Therefore you want to compare TDPs of Steppe Eagle and HieroFalcon? What one was a "high-performance" chip, whereas the other is an "low-power" chip doesn't mean anything for you? Compare like to like. E.g. Opteron X to Seattle. AMD did. Read what they say.

No. In multithreaded tasks (8+ threads) the FX-8350 is faster than i7-2700k, i7-3770k, and i7-4770k. And must be behind the i7-3930k. The FX-9590 would be faster than the i7-3930k.

No its not, 2700k and fx8350 is close but the 3770k and 4770k both comprehensively as in most multithreaded tasks beat the fx8350. In fact it will be hard to find any real world task the fx8350 beats either the 3770k or 4770k at.
The i7 3930k absolutely destroys the fx9590 its not even close stock cpu vs stock cpu so your credibility there goes right out the window of being an amd fanboy and the 3930k has massive overclock potential to.

edit: yes I know the fx8350 is good at video encoding especially on 2nd pass but thats all its good for and the i7 3930k at 4.5ghz will blow away the fx9590 at less power consumption.
http://www.amazon.co.uk/Intel-BX80619I73930K-i7-3930K-6-Core-Processor/dp/B00681D9ZI

No. The FX-8350 is clearly faster than i7-2700k, i7-3770k, and i7-4770k in multithreaded tasks (8+ threads). In fact can be much faster. My only doubt is if the FX-9590 would be faster than the i7-3930k, because I don't have benchmarks of the latter.

Ah! And power consumption means really nothing in the desktop for people who has a high-end GPU and a 1200W PSU.

 

[palladin9479]

Answering some people's questions / posts all at once.

AMD is focusing on APU's first because that's the one area they absolutely dominate in. The value of an APU in a low cost SFF system can not be dismissed. And while we debate and discuss "high end, blah blah" most of the sales are in the budget / low power range to OEM's doing much the same. A low power APU in a sub $500 system is going to yield more overall revenue then a high powered 6~8 core DT chip. AMD won both console design's on the merit of cost vs performance on those APU's. It's also for this reason that AMD is going to be waiting to release any performance DT parts, I suspect we'll see something before July 2014.

About ARM vs x86, it's no contest here. x86 is the more powerful uArch, ARM is the more power efficient one. ARM doesn't sale vertically very well, it's very difficulty to make a "powerful" ARM computer. Instead it scales horizontally, you can make "many" low powered ARM computers on the cheap. This lends them to being used in radically different computing then x86, different tools for different problems. Everything in computing is on a "problem vs solution" mindset. You pick out the right tool for the job and engineer your solution around that tool.

Finally the performance difference between SB/IB/HW and PD/BD is very easy to understand once people take off the war paint. SB/IB/HW and even K10 have 3 ALU's per individual core, PD/BD has 2 ALU's per individual core. This yields Intel an immediate 20~30% performance advantage before taking into account their better cache system and AMD's resource arbitration issues. AMD went this route to slim down their modules so as to put more of them in one location, they were going for a horizontal solution. This has benefits and drawbacks that people can argue all day. This is why attempting to debate "single core" performance for a chip that was deliberately handicapped in that way is kinda pointless. They've done some interesting things to attempt to mitigate that weakness but no matter what they do, 2 ALU's will always be slower then 3 ALU's.

 

[The Q6660 Inside]

Hafijur, we have had enough of you BULL$HIT. GTFO off this thread, we dont want lies every 2 seconds. Ever heard of something called GPU dependancy/bottleneck? Heck @1440p+ a Q9650@3.8 will not likely bottleneck dual 780s or TITANS in SLI. A 8350 will do fine with any GPU out there, when you come across a pesky single threaded game, you realize nobody gives jack $hit about 30 more FPS when you are already at 80+.

 

[The Q6660 Inside]

Answering some people's questions / posts all at once.

AMD is focusing on APU's first because that's the one area they absolutely dominate in. The value of an APU in a low cost SFF system can not be dismissed. And while we debate and discuss "high end, blah blah" most of the sales are in the budget / low power range to OEM's doing
much the same. A low power APU in a sub $500 system is going to yield more overall revenue then a high powered 6~8 core DT chip. AMD won both console design's on the merit of cost vs performance on those APU's. It's also for this reason that AMD is going to be waiting to release any performance DT parts, I suspect
we'll see something before July 2014.

About ARM vs x86, it's no contest here. x86 is the more powerful uArch, ARM is the more power efficient one. ARM doesn't sale vertically very well, it's very difficulty to make a "powerful" ARM computer. Instead it scales horizontally, you can make "many" low powered ARM
computers on the cheap. This lends them to being used in radically different computing then x86, different tools for different problems. Everything in computing is on a "problem vs solution" mindset. You pick out the right tool for the job and engineer your solution around that tool.

Finally the performance difference between SB/IB/HW and PD/BD is very easy to understand once people take off the war paint. SB/IB/HW and even K10 have 3 ALU's per individual core, PD/BD has 2 ALU's per individual core. This yields Intel an immediate 20~30% performance advantage before taking into account their better cache system and AMD's resource arbitration issues. AMD went this route to slim down their modules so as to put more of them in one location, they were going for a horizontal solution. This has benefits and drawbacks that people can argue all day. This is why attempting to debate
"single core" performance for a chip that was deliberately handicapped in that way is kinda pointless. They've done some interesting things to attempt to mitigate that weakness but no matter what they do, 2 ALU's will always
be slower then 3 ALU's.

+1

 

[The Q6660 Inside]

Gah, evil forum double posts!

 

[noob2222]

@juanrga

you pretty much proved my point. you have no proof whatsoever of your "facts" other than marketing tactics deployed to ensure AMD can sell some of the server ARM cpus they are designing.

 

[griptwister]

Hafijur, we have had enough of you BULL$HIT. GTFO off this thread, we dont want lies every 2 seconds. Ever heard of something called GPU dependancy/bottleneck? Heck @1440p+ a Q9650@3.8 will not likely bottleneck dual 780s or TITANS in SLI. A 8350 will do fine with any GPU out there, when you come across a pesky single threaded game, you realize nobody gives jack $hit about 30 more FPS when you are already at 80+.

I applaud you sir! LOL!

 

[juanrga]

Okay show us then, unless its the ssd benchmarks.
Suprised that no other companies come close to intel for sales.
http://linuxgizmos.com/arm-rising-amd-falling-says-report/

AMD winner:

http://www.techpowerup.com/188980/amd-winner-in-q2-intel-up-nvidia-down-according-to-jon-peddie-research.html

The fx8350 only has 8 threads anyway and no the 3770k amd 4770k even beat the fx9590 most of the time. intel cpus have more room to be much faster with overclocking while amd cpus with these test all 8 cores used already lose to intel.

Interesting to see a benchmark using haswell double gflops potential over ivy bridge on here floaing point performance for vp8:
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-3.html

3770k 4770k beating the fx9590 at multithreaded tasks:
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-5.html
mediacoder 4770k beats fx8350 but 2 seconds the 3770k loses to fx8350 but considering how much the 4770k wins by compared to 3770k you can expect the 3770k to go faster.
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-7.html

Yes, the FX-8350 has 8 threads. This was said before. The FX-8350 is clearly faster than i7-2700k, i7-3770k, and i7-4770k in multithreaded tasks (8+ threads).

Funny how you change your version. First, it was a 2700k beats a 8350, after it was changed to "2700k and fx8350 is close", now you start mentioned overclocked i7...

The first benchmarks page that you give us show a 8350 being a 35% faster than i7-3770k. You couldn't achieve 8350 level of performance even with a 3770k @ 4.5Ghz... and the 8350 can OC as well.

Yes, it is interesting that Haswell got a theoretical twice faster FPU than IB and that benchmarks show a 60-75% increase, but there is a problem: people who is serious about FP computations, doesn't use something as slow as a Haswell CPU, but fast GPU/accelerators. AMD offers a FirePro GPU with more GFLOPs than 13x i7-4770k working together. The new Apple Mac Pro includes FirePro.

AMD FirePro > i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k + i7-4770k

Do you get it?

I can see the i7-3770k being a 3% faster than the FX-9590 on the poorly optimized 7-zip. However, I can see the FX-9590 being a 12% faster than the i7-3770k on MediaCoder.

As I said the FX-8350 is faster than i7-2700k, i7-3770k, and i7-4770k in really well multithreaded tasks (8+ threads).

Also if people have a high end gpu setup no one in there right mind would buy an amd cpu to go with it as its like wasting valuable fps of what the card could do and many £'s worth of performance. In that sense intel cpus even more value for money not only do they have there igpu cost more to get to 22nm but they also make graphics cards perform great while amd cpus can sabotage 30-50fps off games performance noticeable in 120hz monitors.

This is plain wrong. No one purchase a high-end GPU to play at 768p and at high resolutions and high details the difference between CPUs is small, which means that AMD CPUs are good enough for gaming.

 

[juanrga]

@juanrga

you pretty much proved my point. you have no proof whatsoever of your "facts" other than marketing tactics deployed to ensure AMD can sell some of the server ARM cpus they are designing.

And again you ignored stuff what I wrote and the third-party tech analysis of Seattle, including Anandtech, which praise the chip as follows:

The chip that has the potential to give Intel some real headaches is “Seattle”. It is a pretty revolutionary design for being an AMD CPU.

[...]

Single threaded performance will be similar to the Opteron X1150, but throughput should up to 4 times higher. There is little doubt in our minds that this might well be one of the best micro server CPU of 2014 (based upon the paper specs). It looks like the Intel Avoton will have a very potent challenger in Q1 2014.

[...]

the combined AMD, ARM and Seamicro technology inside AMD’s new Seattle CPU look extremely promising: these are probably the best specs of a micro server CPU we have seen so far. And since all the right components are now in place, it looks like the micro server is ready for prime time. There is little doubt that Seamicro servers will continue to thrive in their niche market while HP's Moonshot and Dell's Viking will make the market much more popular. So there is good chance that AMD will make a big comeback in 2014 in the server market.

Funny that the single thread performance of the first ARM 64 bit chip is already on the same level than x86 and still some people believes ARM is something about phone level performance.

 

[noob2222]

funnny considering its not even out yet. all the hype about arm catching x86 is based on 20/16nm process. arm a-57 is based on arm a-15, you ignored that tested irl (not according to paper) 1.5ghz x86 > 2ghz a-15.

let's see what ARM speculates market shares by 2017.

http://www.anandtech.com/show/7112/the-arm-diaries-part-1-how-arms-business-model-works/4

enterprise dt/server gains of "flat".