Nvidia GTC16 Webcast Live Stream@5pm/9am PDT *Jen looks great in leather.*

[Rroff]

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.

Fraid you are wrong on this (well sort of). Even if he was being crafty with the wording. Don't forget that HBM2 isn't the only configuration they are using Pascal for.

 

[queamin]

I'm sceptic myself. 15B transistors and only 2TFLOPS more than a Fury X.
HBM2 and Finfet, and still more TDP than a GM200

Well adding what was cut out of maxwell, cost die space and with a new process and hbm2 as well they could be struggling a bit going so big.

 

[Gabbatek]

so pretty much, 980ti ftw for a new more years still

 

[Rroff]

so pretty much, 980ti ftw for a new more years still

This doesn't have a huge amount of implication on consumer cards - 16GB HBM2 setups were never coming before the end of 2016 anyhow.

 

[pmc25]

Fraid you are wrong on this (well sort of). Even if he was being crafty with the wording. Don't forget that HBM2 isn't the only configuration they are using Pascal for.

But this is all they've 'shown'. A render of it, and some headline specs. They (NVIDIA) just claimed to be in production, availability 'soon', but then go on to say that it will be in products in Q1 '17. The only credible statement among those three is Q1 '17. That necessitates that the other two are false. The headline specs sound really bad.

GP104 won't have any of the enterprise specific stuff, will be launching sooner (they hope), and won't have HBM2. They still haven't been able to show it, despite AMD doing so in early December for the first time with their competing products; P10 & P11.

 

[layte]

I'm sceptic myself. 15B transistors and only 2TFLOPS more than a Fury X.
HBM2 and Finfet, and still more TDP than a GM200

At least compare like for like, professional parts are always down clocked in comparison to gaming parts. The FirePro S9300 offers 6.95TF FP32 and 0.4TF FP64 per Fiji chip. So that's an order of magnitude more FP64 and over 50% more FP32 performance for the P100. Plus these are the numbers for what appears to be a part specifically designed for maximum density.

 

[tommybhoy]

Spouting Pascal Q1 we were spot on, they never mentioned what year...

5870/480 re run.

 

[Rroff]

But this is all they've 'shown'. A render of it, and some headline specs. They claim to be in production, availability 'soon', but then go on to say that it will be in products in Q1 '17. The only credible statement among those three is Q1 '17. That necessitates that the other two are false. The headline specs sound really bad.

GP104 won't have any of the enterprise specific stuff, will be launching sooner (they hope), and won't have HBM2. They still haven't been able to show it, despite AMD doing so in early December for the first time, with P10 & P11.

Sorry kind of mis-read your post above - nVidia "technically" have 16nm products in volume production but in relation to P100 its stretching the truth.

 

[pmc25]

Sorry kind of mis-read your post above - nVidia "technically" have 16nm products in volume production but in relation to P100 its stretching the truth.

I don't think the Drive stuff is in production either. The card in the glass case had unmarked heatsinks with Tegra decals on them, to prevent identification of what it was (prior showing was quickly identified as Maxwell, despite them claiming it to be Pascal). Almost certainly means they still don't have Drive PX2 test chips, even if it is going to be the first Pascal product to reach production.

 

[Simon Belmont]

Ordered 2.

June ain't far off.

 

[Jono8]

so pretty much, 980ti ftw for a new more years still

No. It was unlikely Nvidia would reveal pascal gp104 today if it was being released in June .

I dont remember Nvidia releasing any info on maxwell 970/980 until release

 

[Telecaster]

So then, all these orders for an X70 at +30% 980ti performance for £200 must be flying in, oh wait...

 

[Ayahuasca]

Ordered 2.

June ain't far off.

Ah so it's June now, fair play

Coming soon since November 2015

 

[Calo]

Well, that was a nice show. Bit disappointed to have no firm dates for consumer pascal but hey, that's also made my own little GPU selection easier now too.

How about that RoboRace eh? A nice trophy piece for Nvidia.

 

[Jono8]

So then, all these orders for an X70 at +30% 980ti performance for £200 must be flying in, oh wait...

I thought everyone was expecting these end of May/June anyway? It was wishful thinking for a release or even loads of info on gp104 today.

Some info or a tease would have been nive. However, Nvidia wont want to murder sales of maxwell if release is still a couple of months away.

 

[Rroff]

I thought everyone was expecting these end of May/June anyway? It was wishful thinking for a release or even loads of info on gp104 today.

Some info or a tease would have been nive. However, Nvidia wont want to murder sales of maxwell if release is still a couple of months away.

Been saying it all along they won't release much info people want to see today.

 

[pmc25]

I thought everyone was expecting these end of May/June anyway? It was wishful thinking for a release or even loads of info on gp104 today.

Some info or a tease would have been nive. However, Nvidia wont want to murder sales of maxwell if release is still a couple of months away.

There is no possible way that GP104 will be available to buy in June (or July).

 

[Mtom]

At least compare like for like, professional parts are always down clocked in comparison to gaming parts. The FirePro S9300 offers 6.95TF FP32 and 0.4TF FP64 per Fiji chip. So that's an order of magnitude more FP64 and over 50% more FP32 performance for the P100. Plus these are the numbers for what appears to be a part specifically designed for maximum density.

S9300 was specifically advertised as not for double precision. Also is the 50% more FP32 over Fiji is still not impress me from a newer architecture, and twice the transistor count.

Firepro S9170 with the "ancient" Hawaii chip has 5.24TFLOPS SP and 2.62TFLOPS DP. They managed to double that on a new process, with almost 3 times the transistor count.
Not impressed.

 

[Telecaster]

I thought everyone was expecting these end of May/June anyway?

lol really? half the folks in the Pascal thread were creaming themselves over an imminent X70/X80 launch, with HBM, x2 980ti performance and all for £300 lol.

 

[melmac]

On 28nm the professional cards launched months before the consumer cards did.

No they didn't. Nvidia were late with Tesla cards. They missed their contract date with Oakridge for their super computer, only managing to fill the order in September 2012, 5 months after the release of the 680/670.

In the interim they released the K10 which was around May 2012, I think.

The Quadro line of cards didn't appear until October 2012.