I agree. With its main competitors in the field also using ARM, the ecosystem will standardise around that. Increasing amounts of workload will run on ARM, even on x86 servers. I think that is a good thing. With Xe HP(C) also using an actual GPU architecture, instead of trying to use x86 like Xeon Phi did, and an actual foundry service, Intel seems to be committing a lot more to make good products and sales where they can, rather than to rely on vendor lock in and economics of scale alone.
Totally agree with you on Xeon Phi, it was such a bad idea to think that a massive amount of x86 cores (P54C derivatives if memory serves) are better suited than an architecture designed from the ground up for GPU workloads. Peak Intel hubris.
Now that they let go of it, they can suddenly make products that seem to actually be competitive. Seem to, because nobody has tested them yet. But at least it looks as if they have all the right ingredients: the right architecture, up-to-date manufacturing, etc.
Because I (apparently) love writing things down so I can later be officially, on-the-record wrong, I kind of think these will remain pretty specialized, and lose out in favor of (90%) putting accelerators and small cores in servers but without this tight coupling and (10%) smarter but not revolutionary networking/storage gear.
Seems what distinguishes this hardware from any old mini-computer on an AIC is 1) high-bandwidth, fast communication between components on a SoC and 2) special-purpose accelerators and lots of small cores.
But servers have an embarrassment of bandwidth that's expected to grow, and in the long term you can put accelerators and efficiency cores in them without bundling them into a second mini-computer like this. Meanwhile form the other end, it's possible to keep making incrementally smarter NVMe devices or NICs without inventing a whole new category.
Maybe these end up priced to sell in volume, or the software and ecosystem around them is so good that they're an irresistible drop-in speedup for lots of datacenter apps. If they remain only economical for people who have otherwise maxed out a server and need to cram more power in the box, or for a few network/storage-centric uses, then for most of us it's gonna be another trend that comes and goes.
(I should have said "storage devices"--I don't think NVMe as we know it could integrate compression offload, for example, and its crypto support is pretty specific to a single full-disk key I think, but you could imagine a storage device that could do more!)
I think a theory like that depends on the larger issue of "who lands up providing for the data center"?
- nV doesn't have the CPU to provide the sort of total package you have in mind.
- Intel and AMD both seem uninterested in truly grasping the implications of SoC and providing *total* packages. Or to put it differently, both have become so addicted to data center money that they see the low-end as something that can be cobbled together from high-end scraps; that might seem like a sensible decision, but it means they're missing the sorts of tight cross-device integration we've seen on SoCs for years.
- so who's left? AMZ (and Altera and whoever plans their own cloud chips) may well see things your way. They have the problem of limited resources, but they are willing to be sensible bout buying what they don't have internally.
- almost everything I see here (and in the early session) I see already on Apple's SoC. UMA is of course the consistent single virtual address space Intel talked about. Likewise the tagging of every NoC transaction for various telemetry/performance monitoring/optimization purposes like ARM N2's MPAM. As I've pointed out before, Apple's DMA fabric does most of the tasks of these accelerators, from crypto and compression to bulk movement to TCP offload, along with the matrix multiply accelerator that everyone now feels they need. (Sort is an interesting IBM accelerator that I haven't seen in Apple or anywhere else.)
So Apple seems closest to the vision you have of designing this stuff from the ground up as a unity, not as a collection of separate chips that hopefully work together. Of course Apple don't sell server or data center chips... But there's more than one way to skin a cat...
But to circle back to your point, as long as data center equipment is seen as extensions to AMD and INTC CPUs, the current situation will continue. So WILL data center equipment continue that way? Amazon is providing one track to shake things up. A second even more abstracted track could be Apple Cloud Services?
Why would nVidia need its own CPU cores? They currently license ARM cores, and ARM has a plethora of cores for most imaginable scenarios. Plus, they are trying (perhaps unsuccessfully) to buy ARM.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
12 Comments
Back to Article
OreoCookie - Monday, August 23, 2021 - link
Is that Intel‘s first modern (i. e. post XScale sale) ARM product? That seems quite significant.autarchprinceps - Tuesday, August 24, 2021 - link
I agree. With its main competitors in the field also using ARM, the ecosystem will standardise around that. Increasing amounts of workload will run on ARM, even on x86 servers.I think that is a good thing. With Xe HP(C) also using an actual GPU architecture, instead of trying to use x86 like Xeon Phi did, and an actual foundry service, Intel seems to be committing a lot more to make good products and sales where they can, rather than to rely on vendor lock in and economics of scale alone.
OreoCookie - Tuesday, August 24, 2021 - link
Totally agree with you on Xeon Phi, it was such a bad idea to think that a massive amount of x86 cores (P54C derivatives if memory serves) are better suited than an architecture designed from the ground up for GPU workloads. Peak Intel hubris.Now that they let go of it, they can suddenly make products that seem to actually be competitive. Seem to, because nobody has tested them yet. But at least it looks as if they have all the right ingredients: the right architecture, up-to-date manufacturing, etc.
Ian Cutress - Thursday, August 26, 2021 - link
Phi eventually went Silvermont in KNL.jeremyshaw - Tuesday, August 24, 2021 - link
Also plenty of arm products from the FPGA side of the business. One of which is quite popular for the MiSTer retro hardware (silicon logic) project.Intel FPGA are about as inhouse as Intel >10GbE network chips, so I think the FPGA business counts.
watersb - Tuesday, August 24, 2021 - link
Seems as if Hot Chips this year is really huge! Perhaps things deferred in 2020?twotwotwo - Tuesday, August 24, 2021 - link
I think it's mostly that Ian decided to go all-out covering it this year!Ian Cutress - Thursday, August 26, 2021 - link
I did 18 live blogs last year as well! This year I decided to coalesce several blogs from one section into singular piecestwotwotwo - Tuesday, August 24, 2021 - link
Because I (apparently) love writing things down so I can later be officially, on-the-record wrong, I kind of think these will remain pretty specialized, and lose out in favor of (90%) putting accelerators and small cores in servers but without this tight coupling and (10%) smarter but not revolutionary networking/storage gear.Seems what distinguishes this hardware from any old mini-computer on an AIC is 1) high-bandwidth, fast communication between components on a SoC and 2) special-purpose accelerators and lots of small cores.
But servers have an embarrassment of bandwidth that's expected to grow, and in the long term you can put accelerators and efficiency cores in them without bundling them into a second mini-computer like this. Meanwhile form the other end, it's possible to keep making incrementally smarter NVMe devices or NICs without inventing a whole new category.
Maybe these end up priced to sell in volume, or the software and ecosystem around them is so good that they're an irresistible drop-in speedup for lots of datacenter apps. If they remain only economical for people who have otherwise maxed out a server and need to cram more power in the box, or for a few network/storage-centric uses, then for most of us it's gonna be another trend that comes and goes.
twotwotwo - Tuesday, August 24, 2021 - link
(I should have said "storage devices"--I don't think NVMe as we know it could integrate compression offload, for example, and its crypto support is pretty specific to a single full-disk key I think, but you could imagine a storage device that could do more!)name99 - Tuesday, August 24, 2021 - link
I think a theory like that depends on the larger issue of "who lands up providing for the data center"?- nV doesn't have the CPU to provide the sort of total package you have in mind.
- Intel and AMD both seem uninterested in truly grasping the implications of SoC and providing *total* packages. Or to put it differently, both have become so addicted to data center money that they see the low-end as something that can be cobbled together from high-end scraps; that might seem like a sensible decision, but it means they're missing the sorts of tight cross-device integration we've seen on SoCs for years.
- so who's left? AMZ (and Altera and whoever plans their own cloud chips) may well see things your way. They have the problem of limited resources, but they are willing to be sensible bout buying what they don't have internally.
- almost everything I see here (and in the early session) I see already on Apple's SoC.
UMA is of course the consistent single virtual address space Intel talked about.
Likewise the tagging of every NoC transaction for various telemetry/performance monitoring/optimization purposes like ARM N2's MPAM.
As I've pointed out before, Apple's DMA fabric does most of the tasks of these accelerators, from crypto and compression to bulk movement to TCP offload, along with the matrix multiply accelerator that everyone now feels they need. (Sort is an interesting IBM accelerator that I haven't seen in Apple or anywhere else.)
So Apple seems closest to the vision you have of designing this stuff from the ground up as a unity, not as a collection of separate chips that hopefully work together. Of course Apple don't sell server or data center chips... But there's more than one way to skin a cat...
But to circle back to your point, as long as data center equipment is seen as extensions to AMD and INTC CPUs, the current situation will continue. So WILL data center equipment continue that way? Amazon is providing one track to shake things up. A second even more abstracted track could be Apple Cloud Services?
OreoCookie - Wednesday, August 25, 2021 - link
Why would nVidia need its own CPU cores? They currently license ARM cores, and ARM has a plethora of cores for most imaginable scenarios. Plus, they are trying (perhaps unsuccessfully) to buy ARM.