[SemiAccurate] AMD to differentiate cores

[naukkis]

Show me the benchmarks please.

Just after AMD(or Intel) releases first high performance ARM design. Apple might take that path sooner though....

 

[whm1974]

Just after AMD(or Intel) releases first high performance ARM design. Apple might take that path sooner though....

If AMD or Intel releases an ARM design to begin with. And I would bet that Apple comes out with one first, if I was a betting man that is.

 

[moinmoin]

I think we already had the AMD and ARM discussion somewhere before.

AMD would be foolish to focus on ARM instead x64 since for the latter AMD is and always will be a second source at worst. The ARM market on the other hand is highly competitive and if AMD entered it, they would be only one competitor among many, in a very volatile market.

Better pick the one area where said volatile market is boosting progress for everybody involved, and use this for their own products: The steady node advancements by the pure play foundries. In that regard I consider AMD moving to TSMC (and potentially SS) nearly as important as the introduction of the Zen architecture.

As for chasing low power, this is not strictly connected to ARM or the cheap market. In contrary, with the ever increasing amount of cores as well as amount and speed of I/O perfecting idle power usage and general power efficiency is highly important in data centers as well. And that's where most of the money is found.

 

[whm1974]

I think we already had the AMD and ARM discussion somewhere before.

AMD would be foolish to focus on ARM instead x64 since for the latter AMD is and always will be a second source at worst. The ARM market on the other hand is highly competitive and if AMD entered it, they would be only one competitor among many, in a very volatile market.

Better pick the one area where said volatile market is boosting progress for everybody involved, and use this for their own products: The steady node advancements by the pure play foundries. In that regard I consider AMD moving to TSMC (and potentially SS) nearly as important as the introduction of the Zen architecture.

As for chasing low power, this is not strictly connected to ARM or the cheap market. In contrary, with the ever increasing amount of cores as well as amount and speed of I/O perfecting idle power usage and general power efficiency is highly important in data centers as well. And that's where most of the money is found.

Thanks. I should have mention Data Centers as that is what, Intel's biggest market?

 

[NTMBK]

AMD doesn't have a modem, so they can't compete in the mobile market with Qualcomm. And in the data centre, the vast majority of software is already running on AMD64, so why torpedo their own ISA and pay a competitor a license fee?

 

[naukkis]

And in the data centre, the vast majority of software is already running on AMD64, so why torpedo their own ISA and pay a competitor a license fee?

Why is it self destructive to offer product with different ISAs? That was AMD's plan years ago, to offer product for same platform with x86 and arm64 support. x86 for legacy use, arm64 for better performance for those which aren't restricted to x86. They did abandon that because of financial reasons, they simply didn't have money to support both platforms. ARM64 benefits are still there, soon it will become hard to compete with x86 even against off the shelf arm products.

 

[whm1974]

Why is it self destructive to offer product with different ISAs? That was AMD's plan years ago, to offer product for same platform with x86 and arm64 support. x86 for legacy use, arm64 for better performance for those which aren't restricted to x86. They did abandon that because of financial reasons, they simply didn't have money to support both platforms. ARM64 benefits are still there, soon it will become hard to compete with x86 even against off the shelf arm products.

Neither AMD or Intel are standing still with x86-64. Yes there are areas where ARM64 excels in, but AMD64 processors are here to stay for a long while.

 

[NTMBK]

Why is it self destructive to offer product with different ISAs? That was AMD's plan years ago, to offer product for same platform with x86 and arm64 support. x86 for legacy use, arm64 for better performance for those which aren't restricted to x86. They did abandon that because of financial reasons, they simply didn't have money to support both platforms. ARM64 benefits are still there, soon it will become hard to compete with x86 even against off the shelf arm products.

Currently the dominant ecosystem is AMD64, and only two players can compete in it- Intel and AMD. But if AMD divert their attentions to ARM, that increases the viability and legitimacy of the rival platform. Engaging in ARM will actively undermine AMD64. Whereas if great competitive AMD64 CPUs keep coming out, and if all of your software runs on AMD64, the incentive to switch is greatly diminished.

 

[coercitiv]

x86 for legacy use, arm64 for better performance for those which aren't restricted to x86.

This claim was made by many, yet we still have to produce clear evidence that one or the other ISA is overall superior.

We do however have strong evidence that a mature software ecosystem built around one ISA is almost as important as the architecture itself. We do this weird dance of judging relative performance by looking at each ISA in it's "natural habitat", yet as soon as each one pushes towards enemy territory everything becomes a convoluted mess in terms of performance and efficiency assessment, always leading us to the same conclusion: both arm64 and amd64 are well entrenched in their market segment.

Last year Cloudflare loudly declared they were completely switching to ARM by Q4, only to realize just how much they really loved x86 performance and re-embrace "upgraded" Intel CPUs and SSDs by October 2018. And this came from a company that went through the trouble of writing optimized code for ARM and had specific workloads that favored the many-core approach of ARM server chips. One could say it all for show, but that wouldn't reflect well on the theory that arm64 offers better performance, would it?

 

[whm1974]

This claim was made by many, yet we still have to produce clear evidence that one or the other ISA is overall superior.

We do however have strong evidence that a mature software ecosystem built around one ISA is almost as important as the architecture itself. We do this weird dance of judging relative performance by looking at each ISA in it's "natural habitat", yet as soon as each one pushes towards enemy territory everything becomes a convoluted mess in terms of performance and efficiency assessment, always leading us to the same conclusion: both arm64 and amd64 are well entrenched in their market segment.

Last year Cloudflare loudly declared they were completely switching to ARM by Q4, only to realize just how much they really loved x86 performance and re-embrace "upgraded" Intel CPUs and SSDs by October 2018. And this came from a company that went through the trouble of writing optimized code for ARM and had specific workloads that favored the many-core approach of ARM server chips. One could say it all for show, but that wouldn't reflect well on the theory that arm64 offers better performance, would it?

I would argue that a mature software ecosystem is more important then the Architecture/ISA itself as long as the performance is good enough.

 

[beginner99]

Last year Cloudflare loudly declared they were completely switching to ARM by Q4, only to realize just how much they really loved x86 performance and re-embrace "upgraded" Intel CPUs and SSDs by October 2018

To be fair they wanted to go with Qualcomm centriq which was later on cancelled due to some stupid internal politics / bean counter issues.

 

[moinmoin]

Why is it self destructive to offer product with different ISAs? (...) They did abandon that because of financial reasons, they simply didn't have money to support both platforms. (...)

You answered your own question there. AMD by all measures is a midge in the markets it's already active in (all kinds of x64 CPU and GPU markets except mobile SoCs). AMD is well advised to focus on serving and supporting those markets well or better before adding any more markets not covered already.

 

[whm1974]

You answered your own question there. AMD by all measures is a midge in the markets it's already active in (all kinds of x64 CPU and GPU markets except mobile SoCs). AMD is well advised to focus on serving and supporting those markets well or better before adding any more markets not covered already.

This+. AMD can not afford to waste resources on R&D for products other than their core ones.

 

[scannall]

This claim was made by many, yet we still have to produce clear evidence that one or the other ISA is overall superior.

We do however have strong evidence that a mature software ecosystem built around one ISA is almost as important as the architecture itself. We do this weird dance of judging relative performance by looking at each ISA in it's "natural habitat", yet as soon as each one pushes towards enemy territory everything becomes a convoluted mess in terms of performance and efficiency assessment, always leading us to the same conclusion: both arm64 and amd64 are well entrenched in their market segment.

Last year Cloudflare loudly declared they were completely switching to ARM by Q4, only to realize just how much they really loved x86 performance and re-embrace "upgraded" Intel CPUs and SSDs by October 2018. And this came from a company that went through the trouble of writing optimized code for ARM and had specific workloads that favored the many-core approach of ARM server chips. One could say it all for show, but that wouldn't reflect well on the theory that arm64 offers better performance, would it?

I would argue that while ARM V8.x is the better ISA by a fair margin, x86 is pretty much locked in for quite a while yet. Thanks to Microsoft mostly. Can x86 be chipped away at? Sure, and it will be. But that's a very large rock to knock down a few handfuls of gravel at a time.

 

[whm1974]

I would argue that while ARM V8.x is the better ISA by a fair margin, x86 is pretty much locked in for quite a while yet. Thanks to Microsoft mostly. Can x86 be chipped away at? Sure, and it will be. But that's a very large rock to knock down a few handfuls of gravel at a time.

Well x86 is heavily entrenched by the very large base of Applications that people are required to use in order get needed functionality. Which the vast majority are not going be available for ARM v8.x platforms.

 

[scannall]

Well x86 is heavily entrenched by the very large base of Applications that people are required to use in order get needed functionality. Which the vast majority are not going be available for ARM v8.x platforms.

Well yeah, that's what I was trying to get at. At some point in time x86 may become obsolete. But that point in time isn't close at hand. It will be interesting to see what happens after all the Patents expire though. That will happen sooner than x86 going away.

 

[whm1974]

Well yeah, that's what I was trying to get at. At some point in time x86 may become obsolete. But that point in time isn't close at hand. It will be interesting to see what happens after all the Patents expire though. That will happen sooner than x86 going away.

Really good emulation maybe? Even then native software will have the performance advantage.

 

[NostaSeronx]

I think Zen is perfectly fine. It clearly can mimic Core M to Core X ranges of the Core cores from Intel. Copying Intel is the best thing for Zen. However, I think making a small Zen or a bigger Zen defeats Zen's purpose of being a superior Core copy. Zen was meant to aggregate the two architectures plan ULP(cat) and HPC(bul) into one. Breaking it up will add to multiply the costs of the shrinking nodes.

I think the best option is not to differentiate between HPC<->ULP, but rather differentiate more between high cost<->low cost.

So, high-cost solution should always be Zen. With it being similar enough in design to re-use Intel Core optimizations.

Low-cost can be niche because of being diversified from the high-end/high-cost solution. So, bring in the CMT-successor/revolutionary design on 12FDX from Excavator on 28nm. Supposedly, the 28nm to 12FDX price crossover is happening this next quarter. The reason to use CMT over CMP is performance related. Low-cost CMP will always have low frequency, while low-cost CMT can have high frequencies.

 

[TheELF]

This claim was made by many, yet we still have to produce clear evidence that one or the other ISA is overall superior.

You can easily emulate ARM on a desktop PC and run anything you want at native speed,you can emulate a tiny bit of x86 on ARM and it is slow as nobodies business.

Ix86 is pretty much locked in for quite a while yet. Thanks to Microsoft mostly.

Microsoft is trying their damnedest to make windows usable on ARM but windows needs too much processing power and only the most expensive ARM chips come anywhere near.

ARM is an awesome arch but it is only competitive at low power where x86 falls apart,at high power x86 can stretch it's legs and nothing else can come close to it.

 

[DrMrLordX]

One could say it all for show, but that wouldn't reflect well on the theory that arm64 offers better performance, would it?

One could also say that Intel is still good at putting money hats on the right heads. It's all about them discounts.

 

[scannall]

You can easily emulate ARM on a desktop PC and run anything you want at native speed,you can emulate a tiny bit of x86 on ARM and it is slow as nobodies business.

Microsoft is trying their damnedest to make windows usable on ARM but windows needs too much processing power and only the most expensive ARM chips come anywhere near.

ARM is an awesome arch but it is only competitive at low power where x86 falls apart,at high power x86 can stretch it's legs and nothing else can come close to it.

It's more of a Chicken or the Egg problem. Currently, the only place to sell ARM CPU's is in the low power mobile market. Again, Apple is the outlier there. The performance they get from the A series processors is incredible. But each iteration gets closer and closer to being a full desktop design.

Until there is a consumer facing product that could use, and does use a high powered ARM CPU, it won't get designed. And until one exists, then Microsoft will have a tough time getting Windows to run on it well.

Apple will likely be the one that breaks ARM out of the 'Mobile Box'. Though for that to trickle out to the rest of the industry will take time.

 

[naukkis]

You answered your own question there. AMD by all measures is a midge in the markets it's already active in (all kinds of x64 CPU and GPU markets except mobile SoCs). AMD is well advised to focus on serving and supporting those markets well or better before adding any more markets not covered already.

AMD did make one cpu design to support both ARM and X86 instruction sets. Why would they abandon that path in later versions, as I see offering both high performance x86 and ARM versions of their high performance core is brilliant idea. Mass producing die at top-notch production lines is what needs much funding, for cpu design itself AMD has them already done. And there's more use cases for high performance ARM-core, now AMD semi-custom chips are limited to x86 but how about doing semi-custom mobile SOC for some phone maker which itself just can't do it or want more performance that in-shelf ARM cores would have to compete against Apple, Samsung etc.....

 

[whm1974]

You can easily emulate ARM on a desktop PC and run anything you want at native speed,you can emulate a tiny bit of x86 on ARM and it is slow as nobodies business.

Microsoft is trying their damnedest to make windows usable on ARM but windows needs too much processing power and only the most expensive ARM chips come anywhere near.

ARM is an awesome arch but it is only competitive at low power where x86 falls apart,at high power x86 can stretch it's legs and nothing else can come close to it.

Of course I would have to get some PC type ARM hardware to be fair about this, but far as I can tell ChromeOS and Linux are far more suitable to run on such hardware then Windows is.

A bit off topic, but I can't wait to get the PineBook Pro and put it through it's paces.
www.Pine64.org

Edit:
https://forum.pine64.org/showthread.php?tid=7339

 

[TheELF]

It's more of a Chicken or the Egg problem. Currently, the only place to sell ARM CPU's is in the low power mobile market. Again, Apple is the outlier there. The performance they get from the A series processors is incredible. But each iteration gets closer and closer to being a full desktop design.

Until there is a consumer facing product that could use, and does use a high powered ARM CPU, it won't get designed. And until one exists, then Microsoft will have a tough time getting Windows to run on it well.

Apple will likely be the one that breaks ARM out of the 'Mobile Box'. Though for that to trickle out to the rest of the industry will take time.

ARM started out for desktop in the archimedes line and later as a replacement for the 68000 found in the Amiga and Aplle PCs of the time it made it until around the year 2000 where it became obvious that the technology can't scale upwards enough to remain relevant for desktop use.
The same problems remain today,sure technology came a long way and ARM are much more capable than back then but X86 didn't sleep all this time either.
ARM doesn't scale up with power and x86 doesn't scale down with power that's just the way it is,it might change at some point in the future but it's still a long time until then.

 

[TheELF]

but far as I can tell ChromeOS and Linux are far more suitable to run on such hardware then Windows is.

Because they do a lot less just like windows RT,they are build to only use instructions that ARM are good at and just drop everything else.