By: bakaneko (nyan.delete@this.hyan.wan), April 24, 2012 4:24 am
Room: Moderated Discussions
David Kanter (dkanter@realworldtech.com) on 4/23/12 wrote:
---------------------------
>Just finished up a deep dive article looking at the Ivy Bridge GPU, take a look:
>
>http://www.realworldtech.com/page.cfm?ArticleID=RWT042212225031
>
>David
Thanks for the treat.
It's annoying how much specialized hardware, conventions and
data formats are still needed in a GPU. I wish they would
get rid of all that and give us more general purpose number
crunching.
Having to read through 12 documents (for sandy bridge) for
the "ISA" is just painful. Yes, most of it isn't about
instructions themselves. But they are still all necessary
and full of conditions for when, how, what can be done,
which isn't exactly thrilling either. And most of it is
(special purpose)**3.
I'm already way beyond confusion after reading about some of
it which is so useless except for 1-2 cases but interacts in
funny ways with the more programmable parts. Which are ugly
to look at, too.
To me it feels like we took a huge step backward with trying
to program these things which seem to be bound to turn into
regular cores either way or could work better if tightly
coupled to them.
I hope it is just a one step back-two step forward in the
history of computing. Or is there some better trend which I
don't see and will replace traditional CPU cores?
---------------------------
>Just finished up a deep dive article looking at the Ivy Bridge GPU, take a look:
>
>http://www.realworldtech.com/page.cfm?ArticleID=RWT042212225031
>
>David
Thanks for the treat.
It's annoying how much specialized hardware, conventions and
data formats are still needed in a GPU. I wish they would
get rid of all that and give us more general purpose number
crunching.
Having to read through 12 documents (for sandy bridge) for
the "ISA" is just painful. Yes, most of it isn't about
instructions themselves. But they are still all necessary
and full of conditions for when, how, what can be done,
which isn't exactly thrilling either. And most of it is
(special purpose)**3.
I'm already way beyond confusion after reading about some of
it which is so useless except for 1-2 cases but interacts in
funny ways with the more programmable parts. Which are ugly
to look at, too.
To me it feels like we took a huge step backward with trying
to program these things which seem to be bound to turn into
regular cores either way or could work better if tightly
coupled to them.
I hope it is just a one step back-two step forward in the
history of computing. Or is there some better trend which I
don't see and will replace traditional CPU cores?
| Topic | Posted By | Date |
|---|---|---|
| Ivy Bridge GPU article | David Kanter | 04/23/12 08:27 AM |
| Ivy Bridge GPU article | Exophase | 04/23/12 09:08 AM |
| Ivy Bridge GPU article | ltcommander.data | 04/23/12 05:42 PM |
| Ivy Bridge GPU article | Arcadian | 04/24/12 09:03 PM |
| Ivy Bridge GPU article | kalelovil | 04/25/12 07:37 AM |
| Ivy Bridge GPU article | David Kanter | 04/25/12 11:15 AM |
| Ivy Bridge GPU article | Arcadian | 04/25/12 02:34 PM |
| Ivy Bridge GPU article | mik | 04/25/12 03:38 PM |
| Ivy Bridge GPU article | kalelovil | 04/25/12 06:41 PM |
| Ivy Bridge GPU article | David Kanter | 04/25/12 07:42 PM |
| Ivy Bridge GPU article | kalelovil | 04/26/12 01:05 AM |
| Ivy Bridge GPU article | David Kanter | 04/26/12 02:19 AM |
| Ivy Bridge GPU article | kalelovil | 04/26/12 04:19 AM |
| Ivy Bridge GPU article | David Kanter | 04/26/12 07:51 AM |
| Offtopic | EduardoS | 04/26/12 03:37 PM |
| Offtopic | Konrad Schwarz | 04/27/12 03:23 AM |
| Offtopic | none | 04/27/12 04:05 AM |
| Cortex-M4 has some DSP, I think (NT) | Paul A. Clayton | 04/27/12 09:53 AM |
| Offtopic | Anon | 04/28/12 02:28 PM |
| Ivy Bridge GPU article | Foo_ | 04/26/12 04:02 AM |
| Ivy Bridge GPU article | kalelovil | 04/26/12 04:14 AM |
| Ivy Bridge GPU article | mik | 04/26/12 04:53 AM |
| Ivy Bridge GPU article | kalelovil | 04/26/12 06:22 AM |
| Ivy Bridge GPU article | Arcadian | 04/26/12 07:06 AM |
| Ivy Bridge GPU article | mik | 04/26/12 03:52 PM |
| Ivy Bridge GPU article | egan | 04/23/12 10:50 AM |
| Ivy Bridge GPU article | Anon | 04/24/12 02:03 AM |
| Ivy Bridge GPU article | egan | 04/24/12 04:19 AM |
| Didn't you hear about "wheel of reincarnation"? (NT) | Michael S | 04/24/12 08:04 AM |
| Ivy Bridge GPU article | anon | 04/24/12 06:31 PM |
| Ivy Bridge GPU article | ltcommander.data | 04/23/12 06:44 PM |
| Ivy Bridge GPU article | bakaneko | 04/24/12 04:24 AM |



