Aces High Bulletin Board

General Forums => Hardware and Software => Topic started by: Chalenge on June 25, 2016, 11:33:33 AM

Title: Tesla P100 Q4'16
Post by: Chalenge on June 25, 2016, 11:33:33 AM
Now this is what I have been waiting on. The Pascal GPU with HBM2 memory due later this year.

http://www.anandtech.com/show/10433/nvidia-announces-pci-express-tesla-p100
Title: Re: Tesla P100 Q4'16
Post by: morfiend on June 26, 2016, 05:04:20 PM
Been waiting myself!  I see they have a PCIe version so it should be just a plug and play type thing,correct?  This Nvlink had me  :headscratch:  wondering what I would need to use this but as I check further I see it's for multi card setup and specific computer use,in other words not the every day gammers thing.

  Chllenge,any price point yet?   Been saving my pennies for some time so the piggy bank is quite fat...... :devil



 I have a 2700K running 3.9,could use some more ram but since I only play AH on this machine I havent upgraded anything for awhile. So I'm wondering if just this card would do or will I need to upgrade more? Not sure if this older cpu will be a bottle neck or not?


    :salute
Title: Re: Tesla P100 Q4'16
Post by: Chalenge on June 26, 2016, 05:31:22 PM
This is (I think) the commercial, content-creation type card, so it will probably cost in the multi-thousands. It's a good sign that the HBM2 gaming versions may follow as early as Spring, though anything in the third to sixth month would make me happy. Could come sooner obviously.
Title: Re: Tesla P100 Q4'16
Post by: Easyscor on June 26, 2016, 08:17:54 PM
Since this type of issue was raised...

I have the i7 2600K and a single PCIe 2.0 x 16 mode slot. I'm told these new cards should be backward compatible and without too much lose in frame rate. I've been looking at the 1070 as the best bang for the buck. What do you guys think?

Am I kidding myself?
Title: Re: Tesla P100 Q4'16
Post by: Chalenge on June 27, 2016, 12:27:11 AM
Yes, the 1070 should be the best bang per buck for a long time to come. I can't answer about AH3 requirements, but the direction the game is going you would probably be safe with the 1070 or 1080.

The HBM2 cards that may show up one day will offer greater bandwidth and perhaps greater resolution. We will have to wait and see.
Title: Re: Tesla P100 Q4'16
Post by: Easyscor on June 27, 2016, 02:16:01 AM
Thanks, but will it run in that 2.0 slot?
Title: Re: Tesla P100 Q4'16
Post by: artik on June 27, 2016, 02:52:29 AM
Tesla isn't designed for gamers but rather for GPU computing, it is monster in comparison to any gaming card
Title: Re: Tesla P100 Q4'16
Post by: Chalenge on June 27, 2016, 11:41:42 AM
Actually, graphics acceleration, which is great for rendering but not games (although it could help the bang/buck return is negative).
Title: Re: Tesla P100 Q4'16
Post by: Pudgie on June 27, 2016, 06:02:49 PM
Thanks, but will it run in that 2.0 slot?

Yep, should be just fine as long as the PCI-E 2.0 slot that it is put in is set up for 16 lanes..........

 :salute
Title: Re: Tesla P100 Q4'16
Post by: BoilerDown on July 08, 2016, 08:45:20 AM
Everytime I read the thread title, I think its about a car.

Just saying.
Title: Re: Tesla P100 Q4'16
Post by: Easyscor on July 08, 2016, 05:49:01 PM
Thanks Pudgie, I haven't been back in this thread since that post.
Title: Re: Tesla P100 Q4'16
Post by: Chalenge on July 08, 2016, 06:02:01 PM
Everytime I read the thread title, I think its about a car.

Just saying.

Probably the most common use of accelerator cards like this is in automobile design prototyping (artist renderings). They are also used in accident visualizations for courtroom demonstrations, and rendering farms for movies and animations.
Title: Re: Tesla P100 Q4'16
Post by: Pudgie on July 09, 2016, 05:04:04 AM
Thanks Pudgie, I haven't been back in this thread since that post.

No problem.

PCI-E 2.0 lane thruput spec is 500Mb\s. At 16 lanes (16x) the total thruput is 8Gb\s which is more than enough for most games even today as more games are making more use of the GPU vs CPU to draw, render & display graphics so as long as a modern vid card has enough dedicated onboard mem and mem bus bandwidth to keep the vid card GPU from having to use the swap file created in system mem for frame buffers (which will increase the amount of data sent back and forth across the PCI-E lanes between the vid card and system mem thru DMA controller which is much slower than GDDR5 mem\vid card mem bus) and using DDR3 1066 mem (8.5Gb\s transfer rate) or higher there shouldn't be any "bottlenecking", ie, slowdowns. Due to this, IMO a modern vid card w\ a min of 4Gb GDDR5 mem is the min spec for PCI-E 2.0 16x to prevent frame buffer swapping. More onboard mem obviously is better.
At 8 lanes (8x) the thruput is cut in half to 4Gb\s and this is where most modern vid cards will be hampered in a PCI-E 2.0 spec mobo (equal to PCI-E 1.0 16x thruput).

Hope this helps.

 :salute