Jump to content
  • Advertisement
Sign in to follow this  
ynm

Adapteva IC has 50+GFLOPS per watt, can graphics benefit from it?

This topic is 2055 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Hi,

Here is the link, Adapteva claimed 70 GFLOPS per watt, better than Nvidia or AMD flagship GPU:
http://www.adapteva.com/white-papers/ten-myths-debunked-by-the-epiphany-iv-64-core-accelerator-chip/

100 GFLOPS under 2 watts, imagine this IC integrated into mobile platform, then it will have a huge calculator. They can produce it at 199$, 64 core, 100 GFLOPs. Will this kind of IC help the graphics field?

Regards

Share this post


Link to post
Share on other sites
Advertisement
To not get carried away, that list of myths smells of marketing straw-man arguments...

That said, it looks cool. The idea of having lots of fast little CPU's, with no hardware caches or global RAM chips, and instead just a small dedicated bank of RAM directly connected to each core, is the same design used by the PS3's SPUs. IMHO, these kinds of designs are the future of parallel computing.
The PS3's GPU is known to be really, really outdated... but PS3 games continue to look pretty decent because programmers are able to implement a lot of graphical techniques on the SPU's. So yes, this kind of processor would be very, very useful.

I would really love it if every PC had one of these in it to accelerate heavy computation... but it's hard to market it. Old/existing software doesn't benefit from it, and new software has to be written specifically for it (and why would I bother writing my software for it, if no customers yet have one), so there's a real chicken-and-egg problem when trying to turn this into a product.

Share this post


Link to post
Share on other sites
Yeah, what I would really like to see is a modular accelerator card where you can just plug in more and more execution units as you get them, instead of buying whole new hardware every two/three years, kind of like how memory sticks work. That would be interesting.

Share this post


Link to post
Share on other sites

Yeah, what I would really like to see is a modular accelerator card where you can just plug in more and more execution units as you get them, instead of buying whole new hardware every two/three years, kind of like how memory sticks work. That would be interesting.


Yay for scale model of your minecraft house built using processor- and RAM-cubes!

Share this post


Link to post
Share on other sites

I would really love it if every PC had one of these in it to accelerate heavy computation... but it's hard to market it. Old/existing software doesn't benefit from it, and new software has to be written specifically for it (and why would I bother writing my software for it, if no customers yet have one), so there's a real chicken-and-egg problem when trying to turn this into a product.

Would something like OpenCL or OpenMP be able to resolve this (the issue of having to write specifically for it)? I haven't used either, so I have no clue.

Share this post


Link to post
Share on other sites
OpenCL would be able to use that thing pretty nicely but as far as I can tell that thing is a standalone device not a PC expansion. Software already running on it should at least be able to use OpenCL nicely.

Share this post


Link to post
Share on other sites

Yay for scale model of your minecraft house built using processor- and RAM-cubes!
I vote for having a voxel world in which each voxel runs on its own CPU :P

Share this post


Link to post
Share on other sites

[quote name='Hodgman' timestamp='1353822980' post='5003891']
I would really love it if every PC had one of these in it to accelerate heavy computation... but it's hard to market it. Old/existing software doesn't benefit from it, and new software has to be written specifically for it (and why would I bother writing my software for it, if no customers yet have one), so there's a real chicken-and-egg problem when trying to turn this into a product.

Would something like OpenCL or OpenMP be able to resolve this (the issue of having to write specifically for it)? I haven't used either, so I have no clue.
[/quote]"Compute" type languages would be a better fit than OpenMP, but yes, software written for either of them would have a better chance of being instantly portable. You're still facing the problem that most software isn't written using OpenCL though wink.png

Share this post


Link to post
Share on other sites
C'mon guys! With their architecture you get 800MHz * 64 cores which gives you a total of 51.2 GHz!!!! That's wayyyyy more than the puny 16 GHz you get from a GPU!

If that doesn't sell you on their tech, then this mind-blowing performance benchmark is sure to do the trick. Edited by MJP

Share this post


Link to post
Share on other sites

C'mon guys! With their architecture you get 800MHz * 64 cores which gives you a total of 51.2 GHz!!!! That's wayyyyy more than the puny 16 GHz you get from a GPU!

If that doesn't sell you on their tech, then this mind-blowing performance benchmark is sure to do the trick.


I'd like to see a comparison with a GPU implementation instead of a CPU implementation. Of course it's a lot faster then the CPU. Though it was actually less lightning fast then I'd expect. But then again, just 2W :)
I don't doubt their tech is cool and useful, but it's rarely fair to compare cores and "GHz" across different architecture types. Edited by Olof Hedman

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

We are the game development community.

Whether you are an indie, hobbyist, AAA developer, or just trying to learn, GameDev.net is the place for you to learn, share, and connect with the games industry. Learn more About Us or sign up!

Sign me up!