# Inside Pascal: NVIDIA's Newest Computing Platform **URL:** \ **Category:** Technical Blog\ **Created:** [April 5, 2016, 5:23pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541 "2016-04-05T17:23:55Z")\ **Posts on this page:** 20\ **Page:** 2
**Author:** ![anon92356358](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon92356358](https://forums.developer.nvidia.com/u/anon92356358)\ **Post date:** [April 6, 2016, 10:21pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/21 "2016-04-06T22:21:48Z")
Its probably because of the high rate (1/2 FP32) double precision performance. The consumer variant will probably do 1/3rd FP32 DP by running at reduced clocks thereby saving 50 watts or so. Also the card quite literally churns 2x compute than Maxwell. The 16nm node halves the transistor size and they put twice the same - law of conservation of die size :P ---
**Author:** ![anon92356358](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon92356358](https://forums.developer.nvidia.com/u/anon92356358)\ **Post date:** [April 6, 2016, 10:33pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/22 "2016-04-06T22:33:30Z")
They might as well lack all the FP64 as has been the usual case with past architectures except P100's gaming brethren of course. ---
**Author:** ![anon92356358](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon92356358](https://forums.developer.nvidia.com/u/anon92356358)\ **Post date:** [April 6, 2016, 10:36pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/23 "2016-04-06T22:36:07Z")
I was expecting more information regarding the new pre-emption support and about how this chip might perform asynchronous compute tasks... ---
**Author:** ![anon20251040](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon20251040](https://forums.developer.nvidia.com/u/anon20251040)\ **Post date:** [April 6, 2016, 10:51pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/24 "2016-04-06T22:51:43Z")
Consumer variants of Pascal will more likely resemble Maxwell more than GP100. Although they will have the architecture improvements Pascal brings, just they will lack the dedicated DP shader units. dedicated DP shader units are bulky and waste die space on a consumer part as they are not used in 3D graphics. The DP shaders are larger than a single SP shader. So they will instead just cut them out of the architecture all together, making the consumer parts far smaller. And if they make a large consumer part, it will end up with more SP shaders and higher SP performance than GP100 due to not having DP shaders. ---
**Author:** ![anon92356358](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon92356358](https://forums.developer.nvidia.com/u/anon92356358)\ **Post date:** [April 6, 2016, 11:03pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/25 "2016-04-06T23:03:34Z")
My opinion is they'll release a titan black like $1000 consumer GPU that will probably have all the HPC centric advancements of P100 except the ECC and nv-link which have no use for consumers. ---
**Author:** ![anon20251040](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon20251040](https://forums.developer.nvidia.com/u/anon20251040)\ **Post date:** [April 6, 2016, 11:13pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/26 "2016-04-06T23:13:46Z")
It was mentioned in an article somewhere that when questioned on if we will see GP100 in the consumer space, nvidia said no. ---
**Author:** ![anon81564343](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon81564343](https://forums.developer.nvidia.com/u/anon81564343)\ **Post date:** [April 7, 2016, 6:57am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/27 "2016-04-07T06:57:27Z")
there is another site that said this card was like 129k... ---
**Author:** ![anon81564343](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon81564343](https://forums.developer.nvidia.com/u/anon81564343)\ **Post date:** [April 7, 2016, 6:57am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/28 "2016-04-07T06:57:48Z")
im not exactly sure but im guessing about 500k with one... if you were to hook 8 of them up together i think you could run everyones solitare at once :D ---
**Author:** ![anon81564343](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon81564343](https://forums.developer.nvidia.com/u/anon81564343)\ **Post date:** [April 7, 2016, 6:58am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/29 "2016-04-07T06:58:34Z")
i think pascel my be able to but that would be like the $1600 cards ---
**Author:** ![anon81564343](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon81564343](https://forums.developer.nvidia.com/u/anon81564343)\ **Post date:** [April 7, 2016, 7:03am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/30 "2016-04-07T07:03:10Z")
Yes they do but they hevent Revealed themn yet because they are saving the best for last ---
**Author:** ![anon20251040](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon20251040](https://forums.developer.nvidia.com/u/anon20251040)\ **Post date:** [April 7, 2016, 7:36am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/31 "2016-04-07T07:36:44Z")
Not the card, an entire server with 8 P100 tesla cards in it. They said price would be 129k at gtc. ---
**Author:** ![anon92356358](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon92356358](https://forums.developer.nvidia.com/u/anon92356358)\ **Post date:** [April 7, 2016, 10:14am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/32 "2016-04-07T10:14:54Z")
That's probably because the P100 is not the full chip. By the time they think of releasing titan grade products the node will mature allowing them to incorporate a revised GP100 with all 60 SMs. ---
**Author:** ![anon20251040](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon20251040](https://forums.developer.nvidia.com/u/anon20251040)\ **Post date:** [April 7, 2016, 10:35am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/33 "2016-04-07T10:35:20Z")
Also it is a very expensive part and a waste in the consumer space, 1/3 of the die is just Double precision shaders that will never be used on the desktop. They will more than likely use specific parts for consumers that lack dedicated DP shaders. ---
**Author:** ![anon51531848](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon51531848](https://forums.developer.nvidia.com/u/anon51531848)\ **Post date:** [April 7, 2016, 12:56pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/34 "2016-04-07T12:56:57Z")
NVlink It is compatible with pci express 3.0? .\_. ---
**Author:** ![anon79358340](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon79358340](https://forums.developer.nvidia.com/u/anon79358340)\ **Post date:** [April 7, 2016, 6:16pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/35 "2016-04-07T18:16:10Z")
Next year????? ---
**Author:** ![anon46393664](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon46393664](https://forums.developer.nvidia.com/u/anon46393664)\ **Post date:** [April 8, 2016, 1:12am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/36 "2016-04-08T01:12:44Z")
Where are the GPU'S for gaming? u.u ---
**Author:** ![anon29032965](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon29032965](https://forums.developer.nvidia.com/u/anon29032965)\ **Post date:** [April 8, 2016, 6:30pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/37 "2016-04-08T18:30:19Z")
HBM is invention of AMD btw... ---
**Author:** ![anon57564592](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon57564592](https://forums.developer.nvidia.com/u/anon57564592)\ **Post date:** [April 10, 2016, 3:13pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/38 "2016-04-10T15:13:17Z")
Aye, Volta ---
**Author:** ![anon72267818](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon72267818](https://forums.developer.nvidia.com/u/anon72267818)\ **Post date:** [April 12, 2016, 7:36am UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/39 "2016-04-12T07:36:04Z")
I have a doubt. Considering that each processing block withing a SM only has 8 LD/ST units, wouldn't a load take 4 cycles to be performed? ---
**Author:** ![anon29175476](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@anon29175476](https://forums.developer.nvidia.com/u/anon29175476)\ **Post date:** [April 19, 2016, 5:39pm UTC](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541/40 "2016-04-19T17:39:34Z")
Any plans for low/no FP64 server cards? [Previous page](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541.md?page=1) [Next page](https://forums.developer.nvidia.com/t/inside-pascal-nvidias-newest-computing-platform/148541.md?page=3)