Production ARM cpu server -> 64 cores

Message boards : Number crunching : Production ARM cpu server -> 64 cores
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1975636 - Posted: 17 Jan 2019, 20:53:28 UTC

I wonder how a production ARM cpu would do in Seti?

Here is the website https://amperecomputing.com/

They are also looking for developers. Offering deals on 32 core developer rigs. The production machine is designed for communications, edge computing (whatever that is) etc. But it Turbo's to 3.3GHz which is as fast as most high core count Intel servers so it oughta be able to crunch data acceptably fast.

Anyone?

Tom
A proud member of the OFA (Old Farts Association).
ID: 1975636 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1975694 - Posted: 18 Jan 2019, 4:48:29 UTC - in response to Message 1975636.  

I wonder how a production ARM cpu would do in Seti?

Pretty poorly i'd expect.

Benchmarking Amazon's ARM Graviton CPU.
It's a cloud computing benchmark, but the comparison is between ARM & current Epyc & Xeon cloud instances which would give a good idea of what to expect.
ARM has a long way to go in the HPC (High Performance Computing) area.
Grant
Darwin NT
ID: 1975694 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1975697 - Posted: 18 Jan 2019, 5:05:01 UTC

Yep they're only meant for low level/low power server applications and that's it (and where they will shine well).

But once some brawn is required to move a load or heavy I.O. requests then look elsewhere as these will soon show their short comings. ;-)

Cheers.
ID: 1975697 · Report as offensive
Profile Gary Charpentier Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 25 Dec 00
Posts: 30608
Credit: 53,134,872
RAC: 32
United States
Message 1975700 - Posted: 18 Jan 2019, 5:39:00 UTC

There is a huge difference between the ARM instruction set and the AMD or Intel/X64 instruction sets. ARMs are by their very nature much slower because of it. About the ratio between the fastest multicore Intel chip and the least powerful GPU.

However what they lack in speed they might make up for in the power bill.

I don't know of any benchmarks that are published for cost per FLOP. That is of course what we are all interested in. Say cost to buy and run the box, including A/C for a year vs the FLOP's over the year.
ID: 1975700 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1975704 - Posted: 18 Jan 2019, 6:21:07 UTC - in response to Message 1975700.  

Phoronix.com usually does performance-per-watt benchmarks on all the hardware they test. But it requires actual physical tests. Since their Graviton tests were done on a cloud server instance, they couldn't hook it up to their normal WattsUp Pro USB meter for wattage readings. You might like to look at these two articles about ARM hardware they have tested.

https://www.phoronix.com/scan.php?page=article&item=ec2-graviton-performance&num=1

https://www.phoronix.com/scan.php?page=article&item=raspberry-pi-3&num=2
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1975704 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1975785 - Posted: 18 Jan 2019, 16:40:55 UTC
Last modified: 18 Jan 2019, 16:56:31 UTC

Apparently this cpu is a later generation than the benchmarks that have been posted.

https://amperecomputing.com/product/
https://www.theregister.co.uk/2018/09/18/ampere_shipping/

Supposedly this product is also aimed at Analytics tasks. If so, somehow, its got to be able to crunch numerical data at some kind of speed.

Tom
A proud member of the OFA (Old Farts Association).
ID: 1975785 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22158
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1975800 - Posted: 18 Jan 2019, 18:16:19 UTC

There are analytic tasks and analytic tasks - they have not really defined what sort of task they are talking about.
A few years ago one project I was involved in used an ARM processor as the synchronising controller for a nest of assorted Intel & TI processors and it did a very good job at that, but as a raw number cruncher it was pretty poor (orders of magnitude behind TI DSPs we were using).
OK, past behind:
To see if they would be a viable proposition you need to find someone to build a system (given the scaleability inherent in the ARM architecture a representative low core count system would do), port BOINC to the OS, port the current science applications to the OS & processor combination, debug both, then optimise the science application. You don't happen to know an ARM developer with loads of free time do you????
Then there's the "fun" of getting a suitable board design up and running (and you thought getting a mining board going was hard work).
Otherwise chuck a few hundred k dollars (or pounds or euros) and you might get a system working in a couple of years.....
And in reality I would doubt that the performance of a high ARM core count computer would get anywhere near that of a fairly middle of the road GPU.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1975800 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1975827 - Posted: 18 Jan 2019, 22:18:17 UTC

I could learn ARM

- if they had a card with enough CPUs to have 6400+ freely assignable cores and
- less than 180W TDP
- with multi Gb of high speed ram and
- enough cache for CPUs.
- a decent instruction set with low latency, short pipeline and a sufficient register file.

... or I could take a nap ... or something.
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1975827 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22158
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1975909 - Posted: 19 Jan 2019, 8:26:00 UTC

I've done some work with ARM processors - and I would strongly suggest that the nap is the best way forward.....
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1975909 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1975942 - Posted: 19 Jan 2019, 16:45:19 UTC

SNORE.....
A proud member of the OFA (Old Farts Association).
ID: 1975942 · Report as offensive

Message boards : Number crunching : Production ARM cpu server -> 64 cores


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.