Discussion Intel's past, present and future

[Win2012R2]

i have not seen a single EMT64 app in my life

It's a belated copy of AMD64, so you have been running a lot of them

Who is taking innovation with AVX 10.2/APX/AMX?

AMX impact = zero
APX - not even released, Intel's own estimated very modest
AVX 10.2 - well basically AVX512, but attempt to create yet another product differentiation

Next!

 

[511]

If it wasn't for AMD, Intel would have "innovated" us all into being stuck with Itanium.

Well I am glad that didn't happen

APX - not even released, Intel's own estimated very modest

It's a nice improvement and addition to ISA

AVX 10.2 - well basically AVX512, but attempt to create yet another product differentiation

it's basically sorted the issue with AVX/AV2/AVX-512 and actually forced there hand that they can't do like they have been doing it with AVX-512

AMX impact = zero

oh really?

 

[Win2012R2]

oh really?

Until very recently AVX-512 impact was near zero (outside HPC), only with AMD's excellent proper implementation we actually finally moved the needle here. AMX impact is zero because anybody who needs matrix stuff will use Nvidia GPU, plus it drops frequency like good old AVX-512, yeah, good luck having mixed code.

All these "innovations" by Intel had only one purpose - keep CPU buyers off AMD stuff because they were always a step behind on supposedly new stuff that will be super good.

x86 will certainly do fine with AMD taking the torch.

 

[511]

Until very recently AVX-512 impact was near zero (outside HPC), only with AMD's excellent proper implementation we actually finally moved the needle here. AMX impact is zero because anybody who needs matrix stuff will use Nvidia GPU, plus it drops frequency like good old AVX-512, yeah, good luck having mixed code

that's cope pure cope and GLC implementation is fine as well it doesn't do what og AVX-512 implementation did.

All these "innovations" by Intel had only one purpose - keep CPU buyers off AMD stuff because they were always a step behind on supposedly new stuff that will be super good.

another cope if AMD needs to take the torch they need to do daring stuff in doing stuff that benefits competition and the only recent thing is implementation of AVX-512 in client to force Intel to give AVX-512 in client.
They haven't done something like this since AMD64.

 

[Schmide]

that's cope pure cope and GLC implementation is fine as well it doesn't do what og AVX-512 implementation did.

another cope if AMD needs to take the torch they need to do daring stuff in doing stuff that benefits competition and the only recent thing is implementation of AVX-512 in client to force Intel to give AVX-512 in client.
They haven't done something like this since AMD64.

AMD only needs to provide the platform for the community to innovate. Intel kept all their tech behind a walled mote and even charged their big customers (on demand) for basic features. I wrote an avx512 fft for audacity way back in 2012-2014, but it couldn't get tested for lack of hardware and basically had no chance to usurp the sse/avx version.

An AVX-512 version of VPCLMULQDQ Vector Carry-Less Multiplication of Quadwords is making its way into the Linux Kernel and for Sapphire/Emerald Rappids provides a 3x uplift. AMD only a 2x. Intel still has the horrible 2s latency warmup for ZMM but still 3x throughput.

Imagine where we would be if it was released to the masses years ago.

 

[511]

AMD only needs to provide the platform for the community to innovate. Intel kept all their tech behind a walled mote and even charged their big customers (on demand) for basic features.

Well that is true let's not forget the SPR accelerator debacle they were charging to enable accelerators.

I wrote an avx512 fft for audacity way back in 2012-2014, but it couldn't get tested for lack of hardware and basically had no chance to usurp the sse/avx version.

damm

An AVX-512 version of VPCLMULQDQ Vector Carry-Less Multiplication of Quadwords is making its way into the Linux Kernel and for Sapphire/Emerald Rappids provides a 3x uplift. AMD only a 2x. Intel still has the horrible 2s latency warmup for ZMM but still 3x throughput.

Imagine where we would be if it was released to the masses years ago.

oh i agree with feature gating some of the stupidest things they do let's not forget Xeons without AVX-512 for one i want this gone also AMD would do the same if they were in Intel's position.

 

[jpiniero]

and even charged their big customers (on demand) for basic features

I am pretty sure AMD hasn't bothered to implement anything similar to the stuff that's used by OnDemand. It's pretty niche.

 

[jpiniero]

Well that is true let's not forget the SPR accelerator debacle they were charging to enable accelerators.

I still think it was a yield thing (and with Granite, you don't get the option to upgrade)

 

[511]

I still think it was a yield thing (and with Granite, you don't get the option to upgrade)

With GNR you get everything 4/4/4/4/

 

[jpiniero]

With GNR you get everything 4/4/4/4/

That's not true. There are models with it cut down.

 

[coercitiv]

Well that is true let's not forget the SPR accelerator debacle they were charging to enable accelerators.

Accelerators... I read this word next to Intel recently...

Intel QAT Accelerators Being Demoted On Linux By FSCRYPT: Bug Prone & Slow​

Google engineer Eric Biggers basically found the accelerators like QAT with Xeon Emerald Rapids to be a big headache for FSCRYPT file encryption and the performance to be much worse than the AVX-512 VAES optimized code path. Eric Biggers is responsible for a lot of AVX-512 optimizations and other optimizations for different CPU ISA extensions to the Linux kernel cryptography code. In the context of FSCRYPT, using QAT accelerators leads to horrible performance that even the plain C code can outpace but that the AVX-512 VAES path puts everything else to shame on Intel Xeon Emerald Rapids

 

[511]

That's not true. There are models with it cut down.

Well that is due to yield not them locking them down 🙂

 

[DrMrLordX]

Accelerators... I read this word next to Intel recently...

Intel QAT Accelerators Being Demoted On Linux By FSCRYPT: Bug Prone & Slow​

That is sad. Kinda makes you wonder, is this solely a driver issue? Is the underlying hardware just flawed?

 

[511]

That is sad. Kinda makes you wonder, is this solely a driver issue? Is the underlying hardware just flawed?

Can't say at this point

 

[DavidC1]

Most x86 advances are made by Intel. AMD helps, but despite that, is not a really big player. And despite AMD efforts, x86 would be stagnant for a good chunk of time if Intel starts to be screwed up.

Same situation with Windows since depends on x86 the most.

It isn't so much the x86 that Intel is responsible for, but for all the other things they do, which is way more important.

The support they give for the reference devices to be made on is seriously impressive. AMD barely does anything in that regard. They make CPUs and they make GPUs, that's about it. It eases developing new computers and form factors on Intel-based hardware(which includes AMD). They developed everything around their chips. The cornerstone for PC is Intel for this reason.

If Intel goes away tomorrow, all that will go to zero.

 

[Win2012R2]

If Intel goes away tomorrow, all that will go to zero.

Nah, AMD will have sales doubled and will re-hire.

Chances are Govt will force buyout and new company will be called AMDel

 

[511]

Nah, AMD will have sales doubled and will re-hire.

AMD doesn't have the competency to run fabs

Chances are Govt will force buyout and new company will be called AMDel

Chances are goverment will save Intel and their Oregon R&D along with Intel

 

[DavidC1]

Everyone thinks that contra revenue was just Intel giving rebates for using their chip, misses the amount of what they were doing.

If say you wanted to make a new company based on Intel's new Tablet offerings, you'd call up Intel and the easiest thing would be to rebrand their reference design, which there were multiple versions. Full hardware from circuitry to the device itself all ready to go, with your logo ready to be stamped. And they would give you full support on that thing.

This is the primary reason why they were losing so much money. All the new never heard of Chinese brands were multiple Intel designs rebranded.

PC market has essentially been Intel for 3 decades now. It's like if Apple instead of making iPhone and iPad, they gave that design away for any vendor to come and freely rebrand it to be their own.

Of course they don't do everything in PC, many vendors do enough modifications or even roll their own. But it's extremely easy to do so if you don't.

Nah, AMD will have sales doubled and will re-hire.

Chances are Govt will force buyout and new company will be called AMDel

This is a joke right?

 

[Win2012R2]

AMD doesn't have the competency to run fabs

They won't run fabs - they will re-hire handful of people who were doing USB work and other minor stuff

This is a joke right?

Which part?

If say you wanted to make a new company based on Intel's new Tablet offerings

Now that's a joke - what tablet offerings from Intel? This market is owned by Apple

This is the primary reason why they were losing so much money

Nah, they are losing money because they are setup to run as a monopoly with 95% market share, plus the market for computing devices shifted massively towards GPUs which can be fed by ARM CPUs, and of course they lost the fab race, and designs are ****

 

[DZero]

Nah, AMD will have sales doubled and will re-hire.

Chances are Govt will force buyout and new company will be called AMDel

Not so fast. Apple will use that void to fill it with their products. On server side ARM will hit hard and Power takes some of the quota too.

AMD has a BIG problem and is production capability, something Apple don't have.

 

[511]

AMD has a BIG problem and is production capability, something Apple don't have.

and software even AMD have said getting Enterprise share from Intel is difficult

 

[Win2012R2]

On server side ARM will hit hard

It's hitting hard only with hyperscalers who control software stack and use to reduce prices from Intel/AMD.

AMD has a BIG problem and is production capability, something Apple don't have.

They both fab in the same place - TSMC, with enough demand AMD can certainly shift some cheap server chiplet production to Samsung, it's all doable if they stand to get extra 15-20 bln sales per year, with fixed costs being more or less the same that would make them super profitable.

 

[DZero]

and software even AMD have said getting Enterprise share from Intel is difficult

Actually software side AMD is OK, the issue is the capability of production. AMD has a very good uArch but not enough production.

 

[DZero]

They both fab in the same place - TSMC, with enough demand AMD can certainly shift some cheap server chiplet production to Samsung, it's all doable if they stand to get extra 15-20 bln sales per year, with fixed costs being more or less the same that would make them super profitable.

Let's see if AMD is going with Samsung despite the current situation.

 

[Win2012R2]

Let's see if AMD is going with Samsung despite the current situation.

Why would they go now? They want to get premium market customers who want the best, so that's TSMC N2.

But if Intel falls then CPU prices will certainly go up, and market opportunity will appear to use cheaper nodes that likely satisfy a lot of customers (especially if the alternative is pricey).