SoylentNews is people
SoylentNews
Ampere's Product List: 80 Cores, up to 3.3 GHz at 250 W; 128 Core in Q4
The Ampere Altra range, as part of today's release, will offer parts from 32 cores up to 80 cores, up to 3.3 GHz, with a variety of TDPs up to 250 W. As we've described in our previous news items on the chip, this is an Arm v8.2 core with a few 8.3+8.5 features, offers support for FP16 and INT8, supports 8 channels of DDR4-3200 ECC at 2 DIMMs per channel, and up to 4 TiB of memory per socket in a 1P or 2P configuration. Each CPU will offer 128 PCIe 4.0 lanes, 32 of which can be used for socket-to-socket communications implemented with the CCIX protocol over PCIe. This means 50 GB/s in each direction, and 192 PCIe 4.0 lanes in a dual socket system for add-in cards. Each of the PCIe lanes can bifurcate down to x2.
[...] Previously Ampere had stated they were going for 80 cores at 3.0 GHz at 210 W, however the Q80-33 is pushing that frequency another 300 MHz for another 40 W, and we understand that the tapeout of silicon from TSMC performed better than expected, hence this new top processor.
[...] If that wasn't enough, Ampere dropped a sizeable nugget into our pre-announcement briefing. The company is set to launch a 128-core version of Altra later this year.
This will be a new silicon design, beyond Ampere's initial layout of 80 cores for Altra, however Ampere states that while they are using the same platform as the regular Altra, they have done extensive tweaking and optimizations within the mesh interconnect for Altra Max to hide the additional contention that might occur when using the same main memory speeds.
Altra Max will be socket and pin-compatible with Altra, also support dual socket deployments, and Ampere states that the silicon will be ready for early sampling with partners in Q4, and is looking to move into high volume in mid-2021.
Previously: Ampere Launches its First ARM-Based Server Processors in Challenge to Intel
80-Core Arm CPU To Bring Lower Power, Higher Density To A Rack Near You
Related: Amazon Announces 64-core Graviton2 Arm CPU
Marvell Announces ThunderX3, an ARM Server CPU With 96 Cores, 384 Threads
AMD and Intel Have a Formidable New Foe (Amazon)
(Score: 2) by takyon on Tuesday June 23 2020, @06:59PM (16 children)
You are thinking of the discontinued Xeon Phi, based on the 80-core Teraflops Research Chip, which was not a product.
https://en.wikipedia.org/wiki/Teraflops_Research_Chip [wikipedia.org]
https://en.wikipedia.org/wiki/Xeon_Phi [wikipedia.org]
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by JoeMerchant on Tuesday June 23 2020, @07:33PM (7 children)
Yep - they announced that in 2006/7 timeframe - I remember as a "coming to market soon" thing, but maybe that was the usual tech-press hype. Took until 2012 to ship the first product, still haven't quite reached 80 cores in the 2017 release but they do claim 4x hyperthreading.... thanks for the links.
🌻🌻 [google.com]
(Score: 2) by takyon on Tuesday June 23 2020, @08:03PM (6 children)
The funny thing is that Intel is creeping right back up there with Xeon core counts. According to this leak video [youtube.com], there will be 56 and possibly up to 72 cores (on a single die) for Sapphire Rapids Xeon by 2021-2022. Right now they are combining two 28-core dies to get to 56.
For AMD, my guess is that they will do 80-96 cores with Zen 4 Epyc (ten or twelve 8-core chiplets).
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by JoeMerchant on Tuesday June 23 2020, @09:59PM (5 children)
Fifteen years ago the thing was: what's a core, really? Does it count if it doesn't have 64bit multipliers? How far can your nerf a core and still call it a core?
Lots of these Xeons are for "business" applications that don't need 64 bit anything, what they need is the ability to partition and run 1000s of VMs on as little hardware as possible.
🌻🌻 [google.com]
(Score: 2) by takyon on Tuesday June 23 2020, @10:36PM (2 children)
AMD took those questions too far, and got hit with a class action lawsuit. Dark times.
Pretty much everything they make now is x64. Xeon Phi was, Atom is. They killed off 32-bit Intel Quark... with the exception of the Management Engine.
Intel should adopt a chiplet approach and use it to put a bunch of small Atom cores together. As in hundreds of them. The largest they have is 24-core Atom P5962B, as far as I can tell. Make 256-core Atoms, put a few of those near each other, and you've got a compelling box.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by JoeMerchant on Tuesday June 23 2020, @11:46PM (1 child)
I played the options market once, started out with a $100 stake, doubled it, took that and doubled it again, next go round I got more lucky and tripled from $400 to $1200 - I think one of those plays was Amazon circa 2005... forgot the others, anyway... final play was AMD calls in early 2006ish - yeah, bust. They were looking really good with their low power and 64 bit consumer parts advantages back then, but it all came crashing down while I was holding calls.
I guess, when you think about it, even if your VM is only 32 bit, the host machine wants to be 64 bit so they can run tons of 'em.
Also back in the 2006 timeframe we had some IBM sales suits come through and pitch their "flex core" lease model to us. Didn't make sense for us since we intended to actually 100% use whatever processors we had access to, but for businesses who were unsure of when/if they would scale up their IT needs IBM basically proposed a rack system with massively excess capacity, but you only pay for what you are using - when your needs increase, the capacity is there and you start paying for it after it gets switched on. I actually ended up pissing off one of the business guys in that company, he had visions of a rack of 100 Mac Pros feverishly calculating away - we did a code review of his algorithm and identified a 100x speedup - now he only needs one Mac Pro for acceptable performance... no more big impressive shiny "supercomputer" to go in his system, he was crushed.
🌻🌻 [google.com]
(Score: 0) by Anonymous Coward on Wednesday June 24 2020, @04:12AM
Amd64 has a ton more registers than i386, so all other things being equal, will probably be faster (since fewer memory moves). 64 bit also provides a huge address space that makes attacks against ASLR impractical that are practical on 32 bit.
(Score: 0) by Anonymous Coward on Wednesday June 24 2020, @03:12AM
It's what's left when you take away the Apple.
Ugh. Sorry.
(Score: 2) by DannyB on Wednesday June 24 2020, @01:21PM
Marketing says that even if it has only one dumb register it can be called a core.
The lower I set my standards the more accomplishments I have.
(Score: 3, Funny) by DannyB on Tuesday June 23 2020, @08:00PM (5 children)
I still want my dream of enough CPU cores, on a desktop, so that GPUs simply become unnecessary. Imagine the software simplicity of so many general purpose processing elements -- supported by languages and frameworks that make it eas{y|ier} to exploit these cores. And I want a pony too!
The lower I set my standards the more accomplishments I have.
(Score: 0) by Anonymous Coward on Tuesday June 23 2020, @08:19PM (1 child)
video/audio processing is a different beast to general purpose tasks. Dedicated GPU will always outperform array of general purpose cores.
(Score: 0) by Anonymous Coward on Wednesday June 24 2020, @03:15AM
Yeah, whatever happened to the term "vector processor"?
(Score: 2) by takyon on Tuesday June 23 2020, @08:35PM (2 children)
It's a moving target if you consider gaming. New games push up polygon counts, rays traced, etc. and there is a general push for higher resolution and frame rates, 4K @ 60-120 Hz within the next couple years, ultimately stopping somewhere around 8K-16K @ 240-1000 Hz in my opinion. GPUs are hanging out in the 5000-10000 core range, and multi-chip modules could drive that up.
Elsewhere, GPUs or purpose-built AI accelerators will dominate if they continue to improve, although high core count CPUs don't do as bad as you might think [anandtech.com]. But massively parallel neuromorphic architectures will be preferred over CPUs for trying to make brain-like AI happen.
I have the opposite dream. I want to see 3D SoCs with lots of CPU cores, GPU cores, ML/tensor cores, neuromorphic memristors "brains", quantum qubits, and universal memory, all in the same chip, able to communicate with each other in nanoseconds. And then stick that in a nice 1 liter cubic form factor with a bunch of ports and sensors.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Tuesday June 23 2020, @08:45PM (1 child)
"a nice 1 liter cubic form factor ..."
Shaped like a pony.
(Score: 2) by takyon on Tuesday June 23 2020, @09:13PM
Dead and cubic.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Wednesday June 24 2020, @12:32PM (1 child)
But were those independent cores, or just independent ALU's? IIRC they were part of Intel's massively-parallel GPU projects, which means that the cores could not execture different program structures independently -- just execute the same instruction on a wide range of memory locations. Vector programming on steroids, massively parallel data streams instead of massively parallel instruction streams.
Anyway, Sun did it first [wikipedia.org].
(Score: 2) by takyon on Wednesday June 24 2020, @12:40PM
They were Atom cores, but modified from the Atom cores we know today.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]