mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2018-07-05, 12:49   #1
henryzz
Just call me Henry
 
henryzz's Avatar
 
"David"
Sep 2007
Cambridge (GMT)

163116 Posts
Default Alternative CPU architecture

Has anyone come across this: http://www.theregister.co.uk/2018/06...ge_windows_10/
henryzz is offline   Reply With Quote
Old 2018-07-05, 12:58   #2
mackerel
 
mackerel's Avatar
 
Feb 2016
UK

1011110102 Posts
Default

Wasn't aware of this, but will keep an eye on it and see if it goes anywhere in the years to come.
mackerel is offline   Reply With Quote
Old 2018-07-05, 13:37   #3
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

19×31 Posts
Default

Sounds interesting. I'm pretty ignorant so someone with a better understanding please correct me if this is gibberish. Am I right to understand that it's a traditional CPU (by that I mean a normal pipeline with rigid instructions) controlling an FPGA-like resource in a similar way to how you might use a GPU as a co-processor? The controller passes data, instructions and defines the resources to be used for a work-unit, then the computation is a black-box until it returns a result to the controller.
M344587487 is online now   Reply With Quote
Old 2018-07-05, 13:50   #4
henryzz
Just call me Henry
 
henryzz's Avatar
 
"David"
Sep 2007
Cambridge (GMT)

163116 Posts
Default

Quote:
Originally Posted by mackerel View Post
Wasn't aware of this, but will keep an eye on it and see if it goes anywhere in the years to come.
It might be dead and we will hear no more but we can hope.
henryzz is offline   Reply With Quote
Old 2018-07-05, 14:42   #5
mackerel
 
mackerel's Avatar
 
Feb 2016
UK

2·33·7 Posts
Default

M344587487, my quick reading of it was that it does work a bit differently from CPUs as we know them. Work is better assigned through the compiler, to allow this new CPU configuration to work on unrelated operations locally and not more globally, in the hopes it can get more done. Part of me wonders if this might suit more heavily threaded tasks as it doesn't sound like it would necessarily help sequential work, but I'm not sure I got the exact intent either.

The FPGA part is just what they used to build it out of, for testing. I assume it isn't sufficiently advanced for them to go through the cost of getting custom silicon made yet.
mackerel is offline   Reply With Quote
Old 2018-07-05, 17:04   #6
VictordeHolland
 
VictordeHolland's Avatar
 
"Victor de Hollander"
Aug 2011
the Netherlands

23×3×72 Posts
Default

We must not forget what happened to EPIC (a.k.a. Itanium).
POWER and SPARC are also in decline. Putting a new architecture on the market is a big risk.
VictordeHolland is offline   Reply With Quote
Old 2018-07-05, 18:03   #7
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

19×31 Posts
Default

Quote:
Originally Posted by mackerel View Post
...
The FPGA part is just what they used to build it out of, for testing. I assume it isn't sufficiently advanced for them to go through the cost of getting custom silicon made yet.
Ok I misunderstood. I'm sure I've read somewhere about how intel is testing integrating FPGA-like elements into their CPUs and assumed this was that. I've read the EDGE wiki (so I'm now an expert :P ), it still seems interesting but nothing like what I thought.

Quote:
Originally Posted by VictordeHolland View Post
We must not forget what happened to EPIC (a.k.a. Itanium).
POWER and SPARC are also in decline. Putting a new architecture on the market is a big risk.
I think this time may be different. Thanks to custom Arm designs being commonplace foundries are well suited to custom designs and the capacity is there. If they make their designs compatible with existing nodes and infrastructure risk could be minimised at least. Disclaimer I know bugger all about the realities of making chips, it just seems from an outside perspective that now is probably the least risky time it's ever been to try a new approach.
M344587487 is online now   Reply With Quote
Old 2018-07-05, 19:22   #8
rogue
 
rogue's Avatar
 
"Mark"
Apr 2003
Between here and the

2·5·7·83 Posts
Default

I would be curious to know what Intel would do if someone told it that they don't need to make the next generation CPU backward compatible with Windows. IMO, there appears to be a number of "features" in the x86 CPU that seem to only exist because Windows exists.

Note that any executable running on this other CPU Microsoft is testing with cannot run on x86 and vice versa. I would like to know if Microsoft modified the calling conventions when using this other CPU in order to simplify the calling conventions it uses on x86.
rogue is online now   Reply With Quote
Old 2018-07-06, 09:34   #9
mackerel
 
mackerel's Avatar
 
Feb 2016
UK

2×33×7 Posts
Default

IMO Microsoft are sacrificing their cash cow in the hopes of finding the next big thing. They already failed big time with Windows phone and the non-x86 Surfaces. For today, and likely for some time to come, people use Windows for x86 software compatibility. How many care about the so called universal apps?

I know they are looking at ARM, and I feel this is a bit different from before, in that there is a larger software ecosystem around it that might be leveraged in some way.

But yet another new CPU architecture? It would have to give some good reasons for existing. Can it do it at lower price? Performance per watt? Or just all out performance? For consumer markets it usually comes down to price, but if it would be used in other areas, performance per watt may be more interesting.

Even now, Intel and AMD are diverging a bit in their offerings. On one side you have much better FP performance, on the other, generally more cores for the price. Will this impact how near future software is written?
mackerel is offline   Reply With Quote
Old 2018-07-07, 10:57   #10
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

19×31 Posts
Default

Quote:
Originally Posted by mackerel View Post
...
But yet another new CPU architecture? It would have to give some good reasons for existing. Can it do it at lower price? Performance per watt? Or just all out performance? For consumer markets it usually comes down to price, but if it would be used in other areas, performance per watt may be more interesting.
...
As far as I can tell consumers are set for the foreseeable future. Density, efficiency and scalability are always desirable for non-GPU server compute. Complementary ways I see to do that:
  • Performance per die area for common compute tasks unsuitable for offloading onto GPU. It would need to be balanced with efficiency to make it possible to cool.
  • Tailor make a CPU from the ground up for high bandwidth MCM communication, allowing for potentially hundreds of dies to be tightly coupled and in a single node. This has the benefit of being scalable, similar to how AMD is using the same zen dies in 1, 2 and 4 die configs but on a grand scale and with cross-die communication in mind. Bonus points for integrating an L4 cache per die.
M344587487 is online now   Reply With Quote
Old 2018-07-07, 22:44   #11
NookieN
 
NookieN's Avatar
 
Aug 2002

2·33 Posts
Default

Never say never, however modern OoO, pipelined cores are good at keeping their transistors busy when they have data to work on. I didn't see much discussion in that article of how that design could be more immune to latency when it needs to load data into the global registers.

I am not a security researcher but I would also imagine that general purpose FPGA-based processors open up many new side-channel attack possibilities.
NookieN is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Alternative to LL paulunderwood Miscellaneous Math 36 2019-08-26 08:09
P1 on non-x86 architecture. Lorenzo Information & Answers 10 2016-03-16 07:42
The Mill CPU architecture ewmayer Hardware 5 2014-05-03 19:58
Optimizing for CPU architecture: unknown Intel almostfrugal Information & Answers 1 2012-09-19 14:22
RIP 32 bit architecture... chalsall Lone Mersenne Hunters 2 2010-09-27 19:18

All times are UTC. The time now is 11:22.

Tue Jul 14 11:22:30 UTC 2020 up 111 days, 8:55, 1 user, load averages: 1.67, 1.65, 1.65

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.