just do it might as well tbh
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Nvidia GTC16 Webcast Live Stream@5pm/9am PDT *Jen looks great in leather.*
- Thread starter tommybhoy
- Start date
just do it might as well tbh
Will beat a 980ti.
Associate
highly doubtful might aswell get one now !
Caporegime
Will be better than F1 even if the cars dont make it to the end of the race.
Caporegime
Simon stuck your order in for the new card yet? Oh wait....
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
So, to sum up, Nvidia are the first to make a interposer chip with a giant core and 3d memory.. apparently, first of it's kind he says... lol. Also they 'launched' GP100 for the professional market, it will be coming 'soon', but the only actual date given was Q1 for availability to the big OEMs. They also for, from what I recall the first time ever, launched a card without showing the card even once.
To me this screams headline grabbing crap, if this was a proper launch with even a small but steady volume there would have been a card shown on stage and ooo-ing and aaahh-ing from the audience to go along with it. It felt showy, when you're going to be second to market, announce first, don't show the card and claim you were first.
Also he couldn't have said more often what Pascal is aimed at... not gaming. Very small single precision performance gain. FP16 is more for deep learning than anything else and FP64 will be for HPC. The two focuses were improving compute for the two areas that gaming isn't focused in. Minimal Fp32 shader increase for 85% more transistors, minimal FP32 performance gain. If this is what all the Pascal cards are based around, I wouldn't be expecting massive gaming performance gains. Now there is SOME potential for FP16 to be used for cheaper/easier gaming algorithms and to gain some performance that way but it's clearly not the focus.
To me this screams headline grabbing crap, if this was a proper launch with even a small but steady volume there would have been a card shown on stage and ooo-ing and aaahh-ing from the audience to go along with it. It felt showy, when you're going to be second to market, announce first, don't show the card and claim you were first.
Also he couldn't have said more often what Pascal is aimed at... not gaming. Very small single precision performance gain. FP16 is more for deep learning than anything else and FP64 will be for HPC. The two focuses were improving compute for the two areas that gaming isn't focused in. Minimal Fp32 shader increase for 85% more transistors, minimal FP32 performance gain. If this is what all the Pascal cards are based around, I wouldn't be expecting massive gaming performance gains. Now there is SOME potential for FP16 to be used for cheaper/easier gaming algorithms and to gain some performance that way but it's clearly not the focus.
Last edited:
They are being a bit clever with the wording :S while the cores might be in volume production the interposer and HBM2 side of it certainly isn't. Has very little implications for consumer Pascal though.
am i the only one to see the teslaP100 as bad news for pascal ?
the compute numbers doesnt stack up with the size of a 16nm 600mm² chip, if the size minimized the damage, they would have went to a compromise route size/performance at least for yields/cost/efficiency, and if the problem persists even at a smaller size, the consumer gpu should have the same issue then., am i reading this wrong ?
the compute numbers doesnt stack up with the size of a 16nm 600mm² chip, if the size minimized the damage, they would have went to a compromise route size/performance at least for yields/cost/efficiency, and if the problem persists even at a smaller size, the consumer gpu should have the same issue then., am i reading this wrong ?
Well, that was unexpected.
In what way?
True
Associate
- Joined
- 27 Aug 2011
- Posts
- 266
So the chips in the tesla p100 are GP100?
Associate
I'm sceptic myself. 15B transistors and only 2TFLOPS more than a Fury X.
HBM2 and Finfet, and still more TDP than a GM200
The cores are absolutely NOT in production. HBM2 will begin initial production soon (not volume). If they were stockpiling cores now, ready for Q1 '17 Tesla P100, the cores manufactured now as opposed to 9-12 months later would be obscenely more expensive. The only way they could ever justify doing that is if they forsee the design / process being so broken that it's not going to significantly improve in the next 9-12 months, meaning they have to start making them now to get enough non-defective units to launch a year out. That would be completely unheralded in the graphics industry, and probably bank breaking for them.
If they were in production he'd have showed test / chips cards.
They still can't even show GP104 chips / cards, let alone P100 Tesla.
Bunch of lies, again.
The ghost of Fermi says hello ... new memory controller, rushed design, new process, aimed at enterprise, huge, hot & lacking density.
It's looking far more understandable why AMD were looking so smug about the Fiji Pro Duo launch and why they seemed so confident even though it was so late, and seemed happy if anything not to have their HBM2 (Vega) cards until Q1 '17.
It'll take NVIDIA nearly a year to respond with what will likely prove the most expensive graphics chip ever, with only a modest improvement in performance outside of scenarios requiring DP and large amounts of RAM (in applications where it can't be pooled). The headline numbers look really bad for P100 ... I still don't see how it's going to be viable at such low density, with so much enterprise stuff baked into it, with what will probably be a bunch of cores disabled, at huge cost for Titan / 1080Ti. They can't make a different unit with all that stuff stripped out and smaller, or the binning process for Tesla would literally require them to chuck them in the bin.
Last edited: