AMD CPU speculation... and expert conjecture

Page 377 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.


I think this may help answer some questions you may have.

http://uncyclopedia.wikia.com/wiki/Proof_that_everything_on_the_internet_is_true
 


And as a software dev, I am going to argue this point to death.

Its not like anyone uses X86 assembly, or manually drops in X86 opcodes; its all handled by the compiler. So you are going to see easier to accomplish ports, but you aren't going to see any extra performance come out of the porting process. Any low-level optimizations you make on consoles have to be ripped out during the PC port, because they simply WILL NOT WORK.

So this whole "AMD is going to gain a performance edge due to consoles" argument is nonsense.
 


Technically, under US Law, a monopoly of X86 processors wouldn't be recognized as a monopoly, since the CPU market is quite vibrant right now. You have plenty of Windows alternatives, and nothing is stopping you from Windows Apps via WINE on Linux using an ARM CPU.

So please stop it; X86, compared to PPC, ARM, and MIPS, is a minority player in the CPU market.
 


It did help to understand that you are trolling when you are not too busy bashing AMD.
 


Except the fact that WINE requires an x86 CPU.
 


Fixed it for you
 


I'm not even talking graphically; the graphics subsystem needs to be ripped apart during the porting processes. I'm just talking CPU instructions here (threading models, memory subsystem, etc).

Anytime you change platforms, every low-level optimization you made is instantly made invalid.
 


How can that be? You've had the answer for a day or 2 now...

See my post that you have quoted on multiple occasions now as to the differences.

EDIT: Here's some reading from a dev forum to further explain what I am talking about.

http://devgurus.amd.com/thread/166601
 


That's understood, but it also doesn't really matter. ARM also has a monopoly on the ARM ISA, as they're the sole provider of ARM licenses. They 100% control the pricing, distribution and royalty collection mechanism. At any time they can exert that monopoly to increase royalties and suck more profits out of the market which they already seem to be doing. ARM licensees are so far on the hook now. They either pay up or jump to MIPS.



Indeed I was only mentioning the Goliaths in the ARM market that they will be fighting. Do you want to also add all the medium and small ARM competitors which there are dozens? Almost all of which are bigger than AMD. MediaTek is already plowing over the smaller ARM players and are now bigger than AMD/NVidia combined and doubled.

Yes there is an ARM vs x86 fight but it doesn't improve AMDs position as far as competition goes. It makes it far worse in my opinion.

I'm seeing another DOT.COM or networking bubble on the horizon for the ARM solution providers and that will lead to a lot of messy consolidations and hostile take overs.



I say the ISA is largely irrelevant. AMDs issues have almost entirely been with management and execution of their own roadmaps. Just like Intel they didn't focus on the low power segment soon enough.

Heck they sold their low power graphic IP to Qualcomm who is now a 120B Goliath.
 


You have not still provided an answer to my question. The first time you ignored it. The secnid time you gave a generic marketing talk, the third and succesive times you are been dodging the question. Now you provide a link that doesn't answer it.

The question was:


 
http://www.rebellion.co.uk/blog/2013/11/21/rebellion-throws-weight-behind-amd-mantle

"We are proud to announce that we will be working closely with AMD to add Mantle support to our in-house game engine, Asura."

"We believe that supporting Mantle will enable us to stay on the bleeding edge of PC gaming and ensure that we don’t leave any performance on the table when it comes to offering gamers amazing experiences."
 


Im not bashing AMD, im bashing people who think an APU is going to be an awesome replacement for HEDT and only listens to marketing hype and badly written articles claiming AMD is making 100% profit on a product they are selling.

I am bashing people who post and repost articles that aren't even remotely true, hence the link I provided for any people who might need it.
 


WINE on ARM only runs Windows RT programs.

If you want to do x86 WINE you will have to use QEMU for everything.

Take a guess how that will run?

As reference WINE alone usually takes a decent performance hit (up to 40% slower usually).

As anecdotal evidence, the interface on Photoshop CS6 on my 5ghz FX 8350 with 7970 lags a little in WINE in Gentoo. Meanwhile it runs perfectly fine in Windows and I don't have that issue with other programs.

WINE Is far from perfect. I can't even play League of Legends on it without the game running at 50fps instead of 300 and the sound going in slow motion.

All WINE does it allow you to use Windows PE executables and Dynamic Link Libraries from Windows. It's not an emulator, it's far more closer to Valve's DirectX to OpenGL wrapper.

However, WINE will have the following limitations:

1. Wine for Windows x86 can not run Windows RT software without ARM emulation

2. Wine for Windows RT can not run Windows x86 applications without x86 emulation

3. It is not a replacement for Windows at all. Photoshop runs faster in a VM for me than it does in WINE.

As for why there is no ARM port of Photoshop, it's because there's no suitable desktop operating system for ARM. Which is why AMD going ARM only puts them in an awkward situation as they would have to basically bow out of the desktop and laptop market completely. The closest they could come to a retail laptop would be a Chromebook. Yes, Photoshop would run fine on an ARM ISA (assumingly) but on what OS? ChromeOS? Android?

And AMD bowing out of desktop and laptop puts them in a completely horrible market segment. Now they are like Nvidia who must pitch Android devices as great at playing games as their main selling point. Must we review Tegra and Tegra Zone to show that this is far from an optimal solution?

It would also completely invalidate their dGPU lineup. Firepro would be DOA after AMD abandons x86. Intel would snip PCIe in the buttocks and tell everyone to buy an LGA implementation of Xeon Phi.

Intel would more than likely slowly start phasing out PCIe in HEDT. They would have no reason to keep it around, those slots would only be used to sell competitor's parts. In workstation it would be the difference between customers buying Xeon Phi and customers buying Firepro or Quadro/Tesla.

When you're speaking of AMD going ARM only in the future, I don't think you're understanding the massive implications that would cause and that you're thinking that Read and everyone at AMD wouldn't see this coming.

If AMD goes ARM only, they're going to need to use ARM to offset their GPU business.
 


good point.

numerous people here thinks that porting == emulating.
 



"GCN was designed from scratch and it's roughly like an x86 processor with 2048bit SSE support."

I never heard it described like that before. Interesting. More like PHI.
 

er... not exactly. they're trying to branch into the ARM bubble that's reached it's saturation pressure.
while i agree with arm having the capability to supplant x86 in the long run, the hierarchy in ARM ecosystem will be quite hostile towards amd, both in high volume sector and high margin sector unless amd has something very attractive to offer. right now, they don't. they licensed vanilla arm cores instead of uarch(understandable, considering amd's funds and deadlines), their flagship gpu uarch isn't scalable down to ultramobile level to compete with the likes of adreno (qualcomm) and powervr(imagination tech.)... or even mali - without shrinking and tuning for 20nm, at least.
this isn't even the worst part. that starts when arm finds a foothold in the high margin sector with their a64 uarch and successors, then starts to raise their fees. even right now, smaller vendors quickly fall off of arm competition due to pressure from established players.
 


I think there is an important difference. ARM doesn't makes chips, Intel makes chips. Intel want the less possible competition and has been comdended in court by monopolistic/illegal practiques against competitors such as AMD.

ARM want the more open and competitive ecosystem to sell more and more licenses.



Yes the fight doesn't improve AMD position. I have tried to say precisely this with my example about a hypothetical excavator-like Opteron. Even if AMD beats Intel with a superb x86 server product, it will be crushed by ARM. That is the reason which AMD is migrating to ARM servers, with the new ambidextrous strategy.



The ARM64 ISA is more modern, efficient, and easy to implement than x64.

Intel is trying a brute force approach to minimize the limitations of the x86 arch with new processors, but there is no miracles in the horizon.
 


You are bashing AMD with false statements. You take an article about Kaveri performance using ordinary CPU workloads and you claim it is using "APU optimized software". You take an article that doesn't claim a 100% profit (at contrary) and you claim it says 100% profit. You avoid technical discussion and pretend there is only marketing hype. You take another article with benchmarks and claim that are FAKE, because you pretend that Kaveri /Steamroller is 36% slower than Bulldozer... And lots of anti-AMD BS as that.
 


It also run some few others (e.g. CE) but it is not officially supported.

Also, check also last slide of

https://archive.fosdem.org/2013/schedule/event/wine_arm/attachments/slides/233/export/events/attachments/wine_arm/slides/233/Wine_on_ARM_FOSDEM2013.pdf

and this

http://www.winehq.org/pipermail/wine-devel/2011-April/089638.html

for future plans.



GIMP is already ported to ARM and can be used in an ARM desktop machine (if you have one powerful enough). Photoshop doesn't port to ARM by the same reason that they don't port to any archs/OSs and offer the "Not supported on this architecture" error to some Mac desktop users.

The ARM servers will run an ARM OS that is virtually identical to the x86 version in features. Neither AMD nor someone else will be migrating today to ARM to miss desktop/laptop markets. The migration will be made when those markets are ready for the change, not before.

I already explained this migration is not something that will happen today nor tomorrow. I also offered a possible schedule, with desktops being the last in migrating.



Nope. They will release GCN accelerators for ARM chips. I am said (but cannot verify it) that AMD Seattle is already HSA compatible.

Note also that Nvidia has already ported CUDA to ARM. CUDA is not for playing Android games. It is for High-Performance-Computing.



Intel is already moving the next Xeon Phi to socket package for blocking customers to use FirePro/Teslas with Xeon CPUs.



I didn't say "only", true?

 


Yes, which is why GCN > VLIW4 in raw compute capability. It's also far easier to code for if you're running compute functions. That's likely why the consoles went the route they did, the GPU can run lots of compute and so they can utilize far more resources with a close to the metal approach than most people realize.

The same approach would have been far more difficult with some sort of CUDA setup...

@juanrga:

Considering GIMP is freeware on Linux platforms, it would probably run on anything you could get to emulate, closely enough, x86.

Though, blackkstar's point is valid. If you are looking for this timeline to happen in the next...oh...20 years. You're losing your mind.

Everyone said PPC was going to destroy x86 all the way back 20 years ago. Look at PPC now...not doing so hot in the consumer space right? x86 is far stronger than you suspect...like any other ISA, it has it's own sets of issues. Though, those are not insurmountable challenges.
 
Status
Not open for further replies.