** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[Orangey]

IIRC the 9800 was still 65nm and 9800+ was 55.

 

[TheMorningStar]

GDDR5X looks to be an interesting stop gap while the problems with HBM1/2 are still being addressed and there do still seem to be problems given supply of Fiji based cards still lacking, unless of course that's simply down to the size of the silicone and the losses in production likely higher due to this. I do kind of suspect this given the none x version seems to be in better supply, but who knows how many of those cores are actually damaged. I'd hope that AMD are also not putting their eggs in one basket and have a fallback plan if it looks like HBM2 may be having issues outstanding during next gen release.

 

[Seanspeed]

Might not be a supply problem, but cost. Which would explain the surprisingly high prices of Fiji.

Could also be a bit of both.

Anyways, if true, it would put a slight damper on my upgrade plans later in 2016, although there's always still AMD. My current 970 is a stop gap card, but I want my next upgrade to last me a few years.

 

[P.B]

When do people think these will be released, would it pay me to by another titanx or will I regret it ?

 

[tbh]

Just hope HBM isn't the ideal excuse some companies would like to put prices ever higher.

The only excuse they need is "people are willing to pay it", which many are; some would even welcome it, as it would reinforce the feelings of exclusivity and elitism that paying such high prices encourages.

 

[Jono8]

When do people think these will be released, would it pay me to by another titanx or will I regret it ?

Around April/May 2016

 

[Rroff]

When do people think these will be released, would it pay me to by another titanx or will I regret it ?

nVidia's bigger releases typically tend to be around the middle of June or start of November give or take a couple of weeks or so - given that mid-sized 16nm FF+ devices only taped out around July I don't think we'll be seeing volume production of next generation GPUs hit retail this November (nVidia do have a release around November I believe but that is probably something like a X2 card or something).

 

[FoxEye]

nVidia's bigger releases typically tend to be around the middle of June or start of November give or take a couple of weeks or so - given that mid-sized 16nm FF+ devices only taped out around July I don't think we'll be seeing volume production of next generation GPUs hit retail this November (nVidia do have a release around November I believe but that is probably something like a X2 card or something).

Call me optimistic, but I'm hoping for Feb/March for mid-range and mid-high parts. HBM2 won't be ready by then, but I don't think the 1070/1080 will be sporting HBM2 anyhow. Probably just the Titan Next.

In fact HBM2 cards probably will come even later than June, since they will only start mass producing it some time in Q2 (barring delays).

 

[mahius]

I wonder if they have prototype test units by now? The NX has dev kits going out and that makes me sad since it's likely the graphics are on the current 28nm process (XB1/PS4 already use that). I was looking forward to how consoles could utilise the die shrink chips, but it's a damn shame they are shying away from the newest technology. Even with GDDR5, if would be a massive boost to their performance. Meh, I wasn't expect consoles to be ground breaking anyway, just got to rely on good old PC for that. Bring on the almighty 1080ti!

 

[slake7]

Any idea when the new GPU's will be getting released?

 

[Dicehunter]

Any idea when the new GPU's will be getting released?

IF it's anything like previous releases we'll see the next Titan around march and the "Ti" version around June/July.

 

[mahius]

IF it's anything like previous releases we'll see the next Titan around march and the "Ti" version around June/July.

That's only if the 1070/1080 release this year though, unless you meant 2017. I expect the 1070/1080 next year and the Ti version at the end of 2016 or in 2017. Titan version in between of course along with a 1060.

 

[Dicehunter]

That's only if the 1070/1080 release this year though, unless you meant 2017. I expect the 1070/1080 next year and the Ti version at the end of 2016 or in 2017. Titan version in between of course along with a 1060.

No clue where you got 2017 from

But yeah I expect the Titan around March 2016 and the rest around June/July.

 

[FoxEye]

No clue where you got 2017 from

But yeah I expect the Titan around March 2016 and the rest around June/July.

Why do you expect the Titan first? HBM2 isn't available until 2H 2016, is it... Even GDDR5X isn't coming until 2H 2016.

So the only cards likely to come in 1H will be the mid-range ones, surely?

 

[Dicehunter]

Why do you expect the Titan first? HBM2 isn't available until 2H 2016, is it... Even GDDR5X isn't coming until 2H 2016.

So the only cards likely to come in 1H will be the mid-range ones, surely?

Like everyone here it's just complete guess work based on previous release schedules, No one knows anything unless you're working for Nvidia.

 

[JediFragger]

Like everyone here it's just complete guess work based on previous release schedules, No one knows anything unless you're working for Nvidia.

Prevous release schedules? Dude, the last 2 gens then nVidia have released the smaller chip 5/6 months before the larger one! Both kepler AND maxwell. with it being a new process in SUCH a long time then no one knows wtf to expect tbh

 

[Rroff]

Why do you expect the Titan first? HBM2 isn't available until 2H 2016, is it... Even GDDR5X isn't coming until 2H 2016.

So the only cards likely to come in 1H will be the mid-range ones, surely?

nVidia have commercial obligations that the compute cut Maxwell parts can't fulfil and Kepler is getting long in the tooth which tends to suggest they might focus on larger cores earlier than later - which is kind of how Titan came into being. Would likely be a double edged sword though if the cards are compute bias over gaming performance even with the higher focus on compute in DX12, etc.

 

[FoxEye]

nVidia have commercial obligations that the compute cut Maxwell parts can't fulfil and Kepler is getting long in the tooth which tends to suggest they might focus on larger cores earlier than later - which is kind of how Titan came into being. Would likely be a double edged sword though if the cards are compute bias over gaming performance even with the higher focus on compute in DX12, etc.

But GP100 and GP104 are basically different products, and potentially GP104 is a more conventional chip without dependencies on HBM2, etc.

In that case why delay GP104 if it's ready? If Titan can't be made until 2H but the mid-range cards can?

 

[drunkenmaster]

nVidia have commercial obligations that the compute cut Maxwell parts can't fulfil and Kepler is getting long in the tooth which tends to suggest they might focus on larger cores earlier than later - which is kind of how Titan came into being. Would likely be a double edged sword though if the cards are compute bias over gaming performance even with the higher focus on compute in DX12, etc.

They are doing the midrange cores first because of production issues, nothing more or less. 350mm^2 or larger cores have been a problem on every new process since 80nm, Nvidia still did it at 65, 55 and 40nm with at each stage lower yields, more respins and more problems until the 40nm disasters. 28nm and forward they gave up making the big core first for a very good reason. Also the only reason Titan came out months and months after the commercial product is due to yields, cut down parts, lower clock speeds and production volume/yields. poor earlier yields and lower availability suits a ludicrously high margin market with lower volume and doesn't suit the gaming market.

With drastically higher wafer/production costs then big wafers would in theory be much less viable with each new process node now. At some stage it wouldn't surprise me if AMD/Nvidia make use of smaller cores and multi core on package setups to produce bigger compute cards rather than monolithic dies above 400mm^2. With interposers being a reality now, that strategy comes into play. Rather like the less technical dual/quad/octo/etc core cpus which were just two cpus stuck together, interposers will bring a way to stick with smaller higher yield dies and turn them into effectively larger gpus on interposers. Not sure we'll see that with many 14/16nm parts, but by 10nm I think it will be a fairly standard way to get around problems with large dies, simply stop making them.

 

[shamus21]

With this being pascal first gen and on a die shrink small cores are more likely. I am thinking 2017 before Pascal Gpu`s are worth buying into.