By: someone (someone.delete@this.somewhere.com), July 25, 2012 8:58 am
Room: Moderated Discussions
David Kanter (dkanter.delete@this.realworldtech.com) on July 25, 2012 1:37 am wrote:
> New computational efficiency data shows GPUs with a clear edge over CPUs, but
> the gap is narrowing as CPUs adopt wide vectors (e.g. AVX). Surprisingly, a
> throughput CPU is the most energy efficient processor, offering hope for future
> architectures. Our data also shows some advantages of AMD's Bulldozer, and the
> overhead associated with highly scalable server CPUs.
>
> Comments and feedback
> welcome!
>
> David
Calling FLOPS/W and FLOPs/mm2 efficiency is highly misleading because it has no
concept of effective FLOPs while doing something useful. The FP functional units
of a general purpose MPU is a tiny fraction of device area and power budget. Why?
Everything else there SUPPORTS feeding those units over a huge spectrum of usage
in terms of data access and control complexity without demanding quite unreasonable
effort and methods for programming. GPUs have less silicon overhead per FP unit
because they are very less generally useful. For HPC algorithms with complex data
access and control paths GPUs are hugely inefficient and can only approach a tiny
fraction of their theoretical peak FLOPs. Why hasn't anyone run published a SPECfp
score running entirely on a GPU yet? :-D
In a modern process I could tile a 200 mm2 die with nothing but FMACs and clock
and power distribution and blow away everything on your graph but it would not be
capable of anything useful. But hey, what "efficiency" woot!
> New computational efficiency data shows GPUs with a clear edge over CPUs, but
> the gap is narrowing as CPUs adopt wide vectors (e.g. AVX). Surprisingly, a
> throughput CPU is the most energy efficient processor, offering hope for future
> architectures. Our data also shows some advantages of AMD's Bulldozer, and the
> overhead associated with highly scalable server CPUs.
>
> Comments and feedback
> welcome!
>
> David
Calling FLOPS/W and FLOPs/mm2 efficiency is highly misleading because it has no
concept of effective FLOPs while doing something useful. The FP functional units
of a general purpose MPU is a tiny fraction of device area and power budget. Why?
Everything else there SUPPORTS feeding those units over a huge spectrum of usage
in terms of data access and control complexity without demanding quite unreasonable
effort and methods for programming. GPUs have less silicon overhead per FP unit
because they are very less generally useful. For HPC algorithms with complex data
access and control paths GPUs are hugely inefficient and can only approach a tiny
fraction of their theoretical peak FLOPs. Why hasn't anyone run published a SPECfp
score running entirely on a GPU yet? :-D
In a modern process I could tile a 200 mm2 die with nothing but FMACs and clock
and power distribution and blow away everything on your graph but it would not be
capable of anything useful. But hey, what "efficiency" woot!
Topic | Posted By | Date |
---|---|---|
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/25 12:37 AM |
New Article: Compute Efficiency 2012 | SHK | 2012/07/25 01:31 AM |
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/25 01:42 AM |
New Article: Compute Efficiency 2012 | none | 2012/07/25 02:18 AM |
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/25 10:25 AM |
GCN (NT) | EBFE | 2012/07/25 02:25 AM |
GCN - TFLOP DP | jp | 2012/08/09 12:58 PM |
GCN - TFLOP DP | David Kanter | 2012/08/09 02:32 PM |
GCN - TFLOP DP | Kevin G | 2012/08/11 04:22 PM |
GCN - TFLOP DP | Eric | 2012/08/09 04:12 PM |
GCN - TFLOP DP | jp | 2012/08/10 12:23 AM |
GCN - TFLOP DP | EBFE | 2012/08/12 07:27 PM |
GCN - TFLOP DP | jp | 2012/08/13 01:02 AM |
GCN - TFLOP DP | EBFE | 2012/08/13 06:45 PM |
GCN - TFLOP DP | jp | 2012/08/14 12:21 AM |
New Article: Compute Efficiency 2012 | Adrian | 2012/07/25 03:39 AM |
New Article: Compute Efficiency 2012 | EBFE | 2012/07/25 08:33 AM |
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/25 10:11 AM |
New Article: Compute Efficiency 2012 | sf | 2012/07/25 05:46 AM |
New Article: Compute Efficiency 2012 | aaron spink | 2012/07/25 08:08 AM |
New Article: Compute Efficiency 2012 | someone | 2012/07/25 09:06 AM |
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/25 10:14 AM |
New Article: Compute Efficiency 2012 | EBFE | 2012/07/26 01:27 AM |
BG/Q | David Kanter | 2012/07/26 08:31 AM |
VR-ZONE KNC B0 leak, poor number? | EBFE | 2012/08/03 12:57 AM |
VR-ZONE KNC B0 leak, poor number? | Eric | 2012/08/03 06:59 AM |
VR-ZONE KNC B0 leak, poor number? | EBFE | 2012/08/04 05:37 AM |
VR-ZONE KNC B0 leak, poor number? | aaron spink | 2012/08/04 05:51 PM |
Leaks != products | David Kanter | 2012/08/05 02:19 AM |
Leaks != products | EBFE | 2012/08/06 01:49 AM |
VR-ZONE KNC B0 leak, poor number? | Eric | 2012/08/05 09:37 AM |
VR-ZONE KNC B0 leak, poor number? | EBFE | 2012/08/06 02:09 AM |
VR-ZONE KNC B0 leak, poor number? | aaron spink | 2012/08/06 03:33 AM |
VR-ZONE KNC B0 leak, poor number? | jp | 2012/08/07 02:08 AM |
VR-ZONE KNC B0 leak, poor number? | Eric | 2012/08/07 03:58 AM |
VR-ZONE KNC B0 leak, poor number? | jp | 2012/08/07 04:17 AM |
VR-ZONE KNC B0 leak, poor number? | Eric | 2012/08/07 04:22 AM |
VR-ZONE KNC B0 leak, poor number? | anonymou5 | 2012/08/07 08:43 AM |
VR-ZONE KNC B0 leak, poor number? | jp | 2012/08/07 04:23 AM |
VR-ZONE KNC B0 leak, poor number? | aaron spink | 2012/08/07 06:24 AM |
VR-ZONE KNC B0 leak, poor number? | aaron spink | 2012/08/07 06:20 AM |
VR-ZONE KNC B0 leak, poor number? | jp | 2012/08/07 10:22 AM |
VR-ZONE KNC B0 leak, poor number? | EduardoS | 2012/08/07 02:15 PM |
KNC has FMA | David Kanter | 2012/08/07 08:17 AM |
New Article: Compute Efficiency 2012 | forestlaughing | 2012/07/25 07:51 AM |
New Article: Compute Efficiency 2012 | Eric | 2012/07/27 04:12 AM |
New Article: Compute Efficiency 2012 | hobold | 2012/07/27 10:53 AM |
New Article: Compute Efficiency 2012 | Eric | 2012/07/27 11:51 AM |
New Article: Compute Efficiency 2012 | hobold | 2012/07/27 01:48 PM |
New Article: Compute Efficiency 2012 | Eric | 2012/07/27 02:29 PM |
New Article: Compute Efficiency 2012 | anon | 2012/07/29 01:25 AM |
New Article: Compute Efficiency 2012 | hobold | 2012/07/29 10:53 AM |
Efficiency? No, lack of highly useful features | someone | 2012/07/25 08:58 AM |
Best case for GPUs | David Kanter | 2012/07/25 10:28 AM |
Best case for GPUs | franzliszt | 2012/07/25 12:39 PM |
Best case for GPUs | Chuck | 2012/07/25 07:13 PM |
Best case for GPUs | David Kanter | 2012/07/25 08:45 PM |
Best case for GPUs | Eric | 2012/07/27 04:51 AM |
Silverthorn data point | Michael S | 2012/07/25 01:45 PM |
Silverthorn data point | David Kanter | 2012/07/25 03:06 PM |
New Article: Compute Efficiency 2012 | Unununium | 2012/07/25 04:55 PM |
New Article: Compute Efficiency 2012 | EduardoS | 2012/07/25 07:12 PM |
Ops... I'm wrong... | EduardoS | 2012/07/25 07:14 PM |
New Article: Compute Efficiency 2012 | TacoBell | 2012/07/25 07:36 PM |
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/25 08:49 PM |
New Article: Compute Efficiency 2012 | Michael S | 2012/07/26 01:33 AM |
Line and factor | Moritz | 2012/07/26 12:34 AM |
Line and factor | Peter Boyle | 2012/07/27 06:57 AM |
not entirely | Moritz | 2012/07/27 11:22 AM |
Line and factor | EduardoS | 2012/07/27 04:24 PM |
Line and factor | Moritz | 2012/07/28 11:52 AM |
tables | Michael S | 2012/07/26 01:39 AM |
Interlagos L2+L3 | Rana | 2012/07/26 02:13 AM |
Interlagos L2+L3 | Rana | 2012/07/26 02:13 AM |
Interlagos L2+L3 | David Kanter | 2012/07/26 08:21 AM |
SP vs DP & performance metrics | jp | 2012/07/27 06:08 AM |
SP vs DP & performance metrics | Eric | 2012/07/27 06:57 AM |
SP vs DP & performance metrics | jp | 2012/07/27 08:18 AM |
SP vs DP & performance metrics | aaron spink | 2012/07/27 08:36 AM |
SP vs DP & performance metrics | jp | 2012/07/27 08:47 AM |
"Global" --> system | Paul A. Clayton | 2012/07/27 09:31 AM |
"Global" --> system | jp | 2012/07/27 02:55 PM |
"Global" --> system | aaron spink | 2012/07/27 06:33 PM |
"Global" --> system | jp | 2012/07/28 01:00 AM |
"Global" --> system | aaron spink | 2012/07/28 05:54 AM |
"Global" --> system | jp | 2012/07/29 01:12 AM |
"Global" --> system | aaron spink | 2012/07/29 04:03 AM |
"Global" --> system | none | 2012/07/29 08:05 AM |
"Global" --> system | EduardoS | 2012/07/29 09:26 AM |
"Global" --> system | jp | 2012/07/30 01:24 AM |
"Global" --> system | aaron spink | 2012/07/30 02:05 AM |
"Global" --> system | aaron spink | 2012/07/30 02:03 AM |
daxpy is STREAM TRIAD | Paul A. Clayton | 2012/07/30 05:10 AM |
SP vs DP & performance metrics | aaron spink | 2012/07/27 06:25 PM |
SP vs DP & performance metrics | Emil Briggs | 2012/07/28 05:40 AM |
SP vs DP & performance metrics | aaron spink | 2012/07/28 06:05 AM |
SP vs DP & performance metrics | jp | 2012/07/28 10:04 AM |
SP vs DP & performance metrics | Brett | 2012/07/28 02:32 PM |
SP vs DP & performance metrics | Emil Briggs | 2012/07/28 05:11 PM |
SP vs DP & performance metrics | anon | 2012/07/29 01:53 AM |
SP vs DP & performance metrics | aaron spink | 2012/07/29 04:39 AM |
Coherency for discretes | Rohit | 2012/07/29 08:24 AM |
SP vs DP & performance metrics | anon | 2012/07/29 10:09 AM |
SP vs DP & performance metrics | Eric | 2012/07/29 12:08 PM |
SP vs DP & performance metrics | aaron spink | 2012/07/27 08:25 AM |
Regular updates? | Joe | 2012/07/27 08:35 AM |
New Article: Compute Efficiency 2012 | 309 | 2012/07/27 09:34 PM |
New Article: Compute Efficiency 2012 | Ingeneer | 2012/07/30 08:01 AM |
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/30 12:11 PM |
New Article: Compute Efficiency 2012 | Ingeneer | 2012/07/30 07:04 PM |
New Article: Compute Efficiency 2012 | David Kanter | 2012/07/30 08:32 PM |
Memory power and bandwidth? | Iain McClatchie | 2012/08/03 03:35 PM |
Memory power and bandwidth? | David Kanter | 2012/08/04 10:22 AM |
Memory power and bandwidth? | Michael S | 2012/08/04 01:36 PM |
Memory power and bandwidth? | Iain McClatchie | 2012/08/06 01:09 PM |
Memory power and bandwidth? | Eric | 2012/08/07 05:28 PM |
Workloads | David Kanter | 2012/08/08 09:49 AM |
Workloads | Eric | 2012/08/09 04:21 PM |
Latency and bandwidth bottlenecks | Paul A. Clayton | 2012/08/08 03:02 PM |
Latency and bandwidth bottlenecks | Eric | 2012/08/09 04:32 PM |
Latency and bandwidth bottlenecks | none | 2012/08/10 05:06 AM |
Latency and bandwidth bottlenecks -> BDP | ajensen | 2012/08/11 02:21 PM |
Memory power and bandwidth? | Ingeneer | 2012/08/06 10:26 AM |
NV aims for 1.8+ TFLOPS DP ? | jp | 2012/08/11 12:21 PM |
NV aims for 1.8+ TFLOPS DP ? | David Kanter | 2012/08/11 08:25 PM |
NV aims for 1.8+ TFLOPS DP ? | jp | 2012/08/12 01:45 AM |
NV aims for 1.8+ TFLOPS DP ? | EBFE | 2012/08/12 09:02 PM |
NV aims for 1.8+ TFLOPS DP ? | jp | 2012/08/13 12:54 AM |
NV aims for 1.8+ TFLOPS DP ? | Gabriele Svelto | 2012/08/13 08:16 AM |
NV aims for 1.8+ TFLOPS DP ? | Vincent Diepeveen | 2012/08/14 02:04 AM |
NV aims for 1.8+ TFLOPS DP ? | David Kanter | 2012/08/13 08:50 AM |
NV aims for 1.8+ TFLOPS DP ? | jp | 2012/08/13 10:17 AM |
NV aims for 1.8+ TFLOPS DP ? | EduardoS | 2012/08/13 05:45 AM |