Comments on: A Look at NVIDIA’s Kepler-based Tesla K-Series GPU Accelerators http://techgage.com/article/a-look-at-nvidias-kepler-based-tesla-k-series-gpu-accelerators/ PC enthusiasts one-stop resource for high-quality reviews, articles and current technology news. Mon, 03 Aug 2015 20:35:00 +0000 hourly 1 By: Marfig http://techgage.com/article/a-look-at-nvidias-kepler-based-tesla-k-series-gpu-accelerators/#comment-108 Tue, 13 Nov 2012 15:17:00 +0000 http://techgage.com/?post_type=article&p=17698#comment-108 An interesting accusation. I’m sure you are right. I would just love to hear of some evidence to this major news piece everyone else has failed to report yet.

]]>
By: DarkStarr http://techgage.com/article/a-look-at-nvidias-kepler-based-tesla-k-series-gpu-accelerators/#comment-104 Tue, 13 Nov 2012 09:57:00 +0000 http://techgage.com/?post_type=article&p=17698#comment-104 Too bad its outclassed MASSIVELY by AMDs new S10000 Fire Pro with over 1TFLOP of DP power. IT’S OVER 9000!!! :D Oh and BTW Marfig, Nvidia handicaps the GeForce cards to keep theme from being competitive. AFAIK AMD does not.

]]>
By: Rob Williams http://techgage.com/article/a-look-at-nvidias-kepler-based-tesla-k-series-gpu-accelerators/#comment-99 Mon, 12 Nov 2012 22:06:00 +0000 http://techgage.com/?post_type=article&p=17698#comment-99 Any way you look at it, this level of performance on either side is just amazing. So amazing, that it almost makes me wish I was involved in some of these fields, or actually had a need for such performance outside of F@h. Things are getting better and better all the time. It’s exciting.

]]>
By: Marfig http://techgage.com/article/a-look-at-nvidias-kepler-based-tesla-k-series-gpu-accelerators/#comment-98 Mon, 12 Nov 2012 22:03:00 +0000 http://techgage.com/?post_type=article&p=17698#comment-98 Consumer-grade cards do tend to catch-up quickly with Nvidia’s GPGPU line. Here’s the FP performance for the GTX580 at AccelerEyes: http://forums.accelereyes.com/forums/viewtopic.php?f=7&t=1633

The author compares it to the Tesla C2050 at the start of the post. Kinda makes one wonder why not just wait a bit more and go with a consumer card for those tasty TFLOPS. ;) That thought however is destroyed once we think on both cards power consumption difference.

]]>
By: Rob Williams http://techgage.com/article/a-look-at-nvidias-kepler-based-tesla-k-series-gpu-accelerators/#comment-96 Mon, 12 Nov 2012 21:35:00 +0000 http://techgage.com/?post_type=article&p=17698#comment-96 This is something that’s difficult to figure out, because both AMD and NVIDIA are so vague with their answers. To my knowledge, while a typical GPU may be able to have its performance “unlocked”, it wouldn’t match Tesla-level performance. Both sets of cards are optimized for a certain task, both through drivers and hardware at some level.

That said, if my desktop GPU could match this sort of Tesla performance if it were “unlocked”, it’s kind of depressing to think about. Folding@home, for example, would majorly benefit from unlocked performance. Given NVIDIA’s major support for F@h, you’d almost imagine that there’d be a rule written in the driver to unlock the performance for just that app. I can only imagine how much more efficient and useful the GPUs could be for that purpose then.

]]>
By: Marfig http://techgage.com/article/a-look-at-nvidias-kepler-based-tesla-k-series-gpu-accelerators/#comment-94 Mon, 12 Nov 2012 19:24:00 +0000 http://techgage.com/?post_type=article&p=17698#comment-94 Fantastic Article, Rob! The first time I see someone addressing these serious cards in an accessible manner.

I’d like to make a point to our readers about these cards capabilities when compared with the consumer-grade cards. Contrary to what you may hear being said, cards like your GForce 500 are also capable of true double-precision floating point arithmetic. When it is said Tesla cards “unlock the double precisions floating point performance”, what this means is that indeed the performance level was unlocked, not the ability to perform such operations. So far, Tesla cards can perform a double-precision floating point operation in one GPU cycle, as opposed to the GForce 500 series which require 4 cycles (meaning they are 4 times slower for the same operation).

]]>