Linux 5.13 To Add Perf Support For Intel Alder Lake

[andreano]

This is the start of Intel un_f***ing itself.

I think so too, but for different reasons:

1. Die size: If early ARM big.LITTLE designs (A7/A15) are anything to compare with, you can have 8 little cores for the same die area as 1 big core. In order to keep adding cores, AV1 accelerators, SIMD/AI instructions and what not after Moore's law is dead (as it has been for Intel lately), it is either this or adding chiplets. Hint: The latter option is fine if you don't care about power.
2. As we get to absurdly many cores, it makes less and less sense for all of them to be fully performant, because software threads just aren't equally loaded. Hyperthreads were the solution to this, but I think won't make sense anymore once you have plenty of small cores, which makes sense to have, both for efficiency and because they give that much more bang for the buck.

[andreano]

the country

An utterance I'm slightly tired of in discussions of global matters.

[Jumbotron]

I think so too, but for different reasons:

1. Die size: If early ARM big.LITTLE designs (A7/A15) are anything to compare with, you can have 8 little cores for the same die area as 1 big core. In order to keep adding cores, AV1 accelerators, SIMD/AI instructions and what not after Moore's law is dead (as it has been for Intel lately), it is either this or adding chiplets. Hint: The latter option is fine if you don't care about power.
2. As we get to absurdly many cores, it makes less and less sense for all of them to be fully performant, because software threads just aren't equally loaded. Hyperthreads were the solution to this, but I think won't make sense anymore after adding a ton of small cores with much more bang for the buck.

Bang...spot on.

[Guest]

This is the start of Intel un_f***ing itself. It will take an additional 3 years, but the market is already behind Pat Gelsinger in a way that no CEO of Intel has had since Andy Grove, so that will give Pat and Intel some cover for the next 3 years or less to get 7nm out the door at global scale.

I say this as an inveterate AMD fanbois. But Pat Gelsinger is the man for Intel as this time. No one else could have come close. For all who don't know, but Pat Gelsinger is the "Lisa Su" of Intel. He was hired by Intel at the age of 18 with only an Associate Degree from a small technical college. He went on to co-design the 386 and ACTUALLY co-wrote the book on 386 programming. After that, Intel made him chief architect for the 486 and after that Intel made Pat the very first Chief Technical Officer of Intel. He was the CEO of EMC and the CEO of VMWare.

I say all is this as a warning and prophecy to AMD. Intel has their Lisa Su now. You have 3 years to secure your future for the rest of the decade. It will not be enough time.

#1. Drop ROCm now and embrace OneAPI and make it a BETTER OneAPI for AMD CPUs nd GPUs.

#2. Give free hardware and software including AMD optimized OneAPI to every single Comp Sci and Comp Engineering school in the country

#3. Buy a BIOS company. Doesn't matter which. Just buy one. Optimize it for AMD products only. Then FORCE vendors to ONLY use AMD BIOS on ALL AMD BOARDS BOTH OEM AND 3RD PARTY BOARD MAKERS WITH EVERY OPTION AVAIALBLE AND TURNED ON FULLY SUPPORTED.

#4. Immediately...like TODAY...dust off your ARM license and start competing with Apple and Ampere and Fujitsu in the ARM consumer and server space. You WILL NOT defeat Intel in x86 long term. But you CAN carve a VERY BIG slice off the ARM market from Nvidia. DO NOT LET NVIDIA WITH THEIR ARM BASED GRACE CPUs and NVIDIA GPUs TAKE CONTROL OF THAT MARKET !! An AMD ARM SiP with HBM, IF, CCIX, AMD CDNA GPU, ARM DSP, NUERAL NET ACCELERATOR and an on board Xilinx FPGA would be KILLER !! Do...THIS....NOW !!

If you don't, Lisa, then AMD by 2030 will be a byword in the industry and you'll be right back where you found AMD in 2015.

Yeah, ROCm sucks. Honestly, they should just do OpenCL and Vulkan Compute. Forget everything else.

[Guest]

Yes but they haven’t been very successful in laptops yet have they? Only Apple has made it work.

??? ARM Chromebooks have been around for a while now, and they seem to be doing well.

[jayN]

The hipSYCL presentation yesterday described a great deal of work that appears to be supporting an AMD version of oneAPI. The presentation will be available Apr 28.

https://www.iwocl.org/iwocl-2021/conference-program/

[cynical]

??? ARM Chromebooks have been around for a while now, and they seem to be doing well.

ChromeOS is a web browser that runs web apps, or Android apps. That severely limits what you can actually do with it. (to the point that it didn’t even register to me that it would count as a laptop)

Apple is the first to get ARM working well on a real, general purpose OS. Qualcomm tried to do it on Windows with one of the Microsoft laptops (a premium device) and they failed really badly.

[deppman]

> Apple is the first to get ARM working well on a real, general purpose OS.

IMO, the OS isn't the problem, it's proprietary apps. I have the Tegra K1 (circa 2014) in both a Chromebook and a Jetson TK1. The Chromebook runs very well and is usually indistinguishable from an x86 device. The Jetson + Ubuntu + KDE also ran very well as did the ports of almost all the OSS apps like Libre office, and that was 5 years ago. What it lacked were proprietary ports like Chrome, Dropbox, Flash, Skype, etc.

And ARM SoCs have gotten a lot better since then, but I don't see much movement either on Windows or Linux on the apps (maybe RasPi has improved that but I haven't looked in a while).

Apple can convince devs to port in a way almost no other company can, and that is perhaps why they are having more success.