Overclock.net banner

[kitg, semia] Soft Machines unveils VISC CPU architecture

5K views 48 replies 20 participants last post by  mtcn77 
#1 ·
Source: Kitguru
Quote:
Soft Machines, a startup that develops microprocessor technologies, has unveiled its own VISC processor architecture. The developer claims that VISC can increase processor's average number of instructions executed per each clock cycle (instructions per cycle, IPC) by three or four times, thus increasing performance performance-per-watt by two or four times in single- and multi-threaded applications compared to existing microprocessors.

There is a number of ways to boost performance of a central processing unit (CPU), but the most obvious are: to boost its frequency, to increase the number of processing cores, to improve IPC execution. Amplified clock-rates advance performance in single-threaded applications and have little effect on multi-threaded applications; increased amount of cores boosts performance in multi-threaded programs, but has no effect on single-threaded apps; improved IPC brings benefits to all programs, but in the recent years the growth of IPC has been very slow. Tangible improvements of IPC requires addition of new hardware into every core of a multi-core CPU, which makes designs bigger and more expensive.

Soft Machines has no plans to produce VISC processors itself, but it intends to license technology to those, who can use it.

Soft Machines was established around 2007 and since then has got $125 million from various companies and organizations, including AMD, GlobalFoundries, Samsung as well as govern-ment investment funds from Abu Dhabi (Mubadala), Russia (Rusnano and RVC), and Saudi Arabia (KACST and Taqnia). Sanjay Jha, chief executive officer of GlobalFoundries, is the chairman of Soft Machines. At present Soft Machines has more than 250 employees and 75+ issued patents.

At present VISC processor proof-of-concept prototype can run Linux operating system and boot Android.


Source: Semiaccurate
Quote:
So how does it do in practice? Since Soft Machines has a prototype chip they should have numbers, right? The company gave a demo of that part running the EEMBC RGB-CMYK DENBENCH benchmark against a 2C A15@1.7GHz (Exynos 5250), a 1C Hasswell@1GHz(3550M), and a Baytrail Atom@1.86GHz (Z3470). The Soft Machines Dual V-Core Test chip had 1MB of L2 and won the benchmark as you can see below. Test times are in minutes, lower is of course better.

At least on this benchmark, the Soft Machines architecture works well. The company claims it will draw between 1/4 and 1/2 the power or run at ~2x the performance of competing architectures across a variety of benchmarks. They showed a wide range of numbers on one slide including SPEC 2000, SPEC 2006, EEMBC DE, Kraken, and of course Dhrystone all against an A15. Needless to say they won by large margin on the low-end, a large multiple in a few cases.

The important thing to note is that the Soft Machines core has a much higher IPC than the others, and by much we mean a multiple. If you look at the test results above, the slowest of the competition is running at 1GHz, the VISC chip was running at a mere 350MHz and finished significantly faster. If only the Soft Machines core could run at clocks of four digits, that would be really impressive.


VISC Architecture Tech briefing
We need this for more competition.
yessir.gif
 
See less See more
3
#6 ·
Quote:
Originally Posted by Ultracarpet View Post

I would buttttttt, big Maxwell may be replacing the graphics card as well....
tongue.gif


Depends on how good the r9 3xx series is. For some reason I feel like it's going to be underwhelming.
Lol, yeah okay you can always have whatever you want
smile.gif


On Topic:
If VISC appears in Windows PC, you might have to write "Sorry Intel" as well in upcoming years.
tongue.gif
 
#7 ·
Quote:
Originally Posted by sumitlian View Post

Lol, yeah okay you can always have whatever you want
smile.gif


On Topic:
If VISC appears in Windows PC, you might have to write "Sorry Intel" as well in upcoming years.
tongue.gif
nah I would never apologize to Intel... the dark side never gets an apology.

lol seriously though, I wonder if AMD has already been working with them for their ARM core considering they were one of the big investors named.
 
#8 ·
Quote:
Originally Posted by ruonboothb View Post

Their first customer in line is AMD lol
rNe201
If early VISC sample can run Linux then AMD might be the first to use it on PC, You must keep this in mind that AMD isn't providing funding for no reason. And Samsung might be the one to use it for their Android mobiles.
I suspect it to arrive on 2016-17. What I don't understand why AMD is funding this when they already working on Zen to be arrive in 2016. May be VISC is being followed by this Zen arch, who knows ?
 
#9 ·
Quote:
Originally Posted by sumitlian View Post

If early VISC sample can run Linux then AMD might be the first to use it on PC, You must keep this in mind that AMD isn't providing funding for no reason. And Samsung might be the one to use it for their Android mobiles.
I suspect it to arrive on 2016-17. What I don't understand why AMD is funding this when they already working on Zen to be arrive in 2016. May be "Zen" is being followed by this VISC arch, who knows ?
Or maybe Zen has some VISC arch in it!!!!
thumb.gif
 
#10 ·
Quote:
Originally Posted by Ultracarpet View Post

lol seriously though, I wonder if AMD has already been working with them for their ARM core considering they were one of the big investors named.
And I think reviewers have sorted names of investors by alphabet. AMD, GlobalFoundries, Samsung. LOL
redface.gif

AMD probably is not the biggest investor considering they have very little money compared to other investors.
Well It doesn't matter who's the biggest investor
tongue.gif

If any one of them comes among investors, they indeed are going to use it in real world. And it matters
wink.gif

...for all of us.
thumb.gif
 
#13 ·
Even mobile computing seems to moving to 64-bit, so I wonder how long it would take to update their prototype to 64-bit. I'm not sure the required changes to their "converter layer" would be a simple thing or require a few more years development.
 
#14 ·
Is there a document explaining juicy details about what's happening in the hardware? I looked at that .pdf with the slides for the presentation and it describes what they are saying that it can do, but not how it actually does that, so that's basically only marketing drivel.

This document here is a lot more interesting (second half of page 2 and page 3): http://softmachines.com/wp-content/uploads/2014/10/MPR-11303.pdf

While that's interesting, it does not really have anything that hints to why this should be a lot different and better performance compared to normal out-of-order architectures like in Intel's and AMD's CPUs, or why this should be better than what NVIDIA does in their Denver stuff.
 
#19 ·
This is brilliant, I hope this makes it to consumer hardware eventually. So this essentially splits the transistors into cores depending on the program's threading. So a single threaded application allows it to use all the transistors as one large core, then a multithreaded application allows it to split into the amount of threads it needs while still using all the transistors. Is that how this works?
 
#20 ·
Quote:
Originally Posted by Tsumi View Post

Moore's "law" wasn't about performance...
And as you indicate by the " " it wasn't a real law of nature either. Just an observation without any consistency to make it an actual law. If all people on earth die the "law" no longer applies thus it is not a real law.
 
#22 ·
Finally, new cpu architecture made by a company other than Intel & AMD
biggrin.gif
Looks promosing, really looking forward to more details.
 
#24 ·
Quote:
Originally Posted by Ultracarpet View Post

lol seriously though, I wonder if AMD has already been working with them for their ARM core considering they were one of the big investors named.
Wouldnt surprise me, since Nvidia just proved that using this sort of conversion layer to move to a much more advanced design is actually doable and the performance can be very good.
 
#26 ·
Quote:
Originally Posted by MapRef41N93W View Post

Who? How did they get an x86 license? And why are they named after Robert Wyatt's old band?
I'm pretty sure their chip is ARM. But that doesn't mean this virtual core technology couldn't be implemented into a x86/x64 processor given access to the patents.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top