Nobody said he did, least of all him. Stop reading his posts then, at least he makes some sense when he posts.
-
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.
** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **
- Thread starter Boomstick777
- Start date
Nobody said he did, least of all him. Stop reading his posts then, at least he makes some sense when he posts.
Soldato
- Joined
- 11 Mar 2013
- Posts
- 5,470
I'm hoping for 1080 with a 30% performance boost over Titan X. Keeping my fingers crossed?
Not really a straight answer though is it.
Well maybe we should have a new thread with a poll.
Do you think NVidia will show a Pascal based discrete GPU at GTC and say that it will be available within 1-8 weeks?
1. yes
2. no
Nothing fancy, no ulterior motives, I would just like to see how many think it is likely and how many think it isn't.
Last edited:
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
They aren't, yields are the only issue, there is fundamentally no difference at all to building a 100mm^2 core and a 600mm^2 core. Wafers are made in 300mm diameters, they have a fixed cost. The yields go down as core size goes up. A 600mm^2 gpu compared to a 300mm^2 gpu will have lower yields, but generally not horrifically lower unless a mistake has been made, it will use the power that twice the transistor count will use... so what. Bigger chip uses more power. Low power processes don't mean anything, it's talking about per transistor performance, not chip performance.
Big cores will probably go away eventually in favour of modular approach, big chip cards won't go away, this will just allow higher yields and singular tape outs and then putting more or less together on an interposer. We aren't there yet.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
It was a straight forward answer, just not the one you were looking for.
Also your question appears to be limited there, do I think Nvidia will show Pascal when... do I think they can show it and it then be available in 1-8 weeks... sure, today, nope, July, probably.. April, unlikely.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
http://www.nvidia.com/object/drive-px.html
says second gen Drive PX coming in Fall.
http://www.coinspeaker.com/2016/01/...ve-px2-a-supercomputer-for-self-driving-cars/
I would rather go with the news direction from NVidia over some 3rd party news site I have never heard of before. NV said availability second quarter.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Sorry, Nvidia's own website is a third party news site you haven't heard of before and Nvidia are directly saying available in the fall... which is also what I heard from everywhere else. In fact you're the only one anywhere saying anything except samples are going out. So I'll take Nvidia's site, and Nvidia themselves over you saying it's being released to tier 1 customers in Q2.
Since when was coinspeaker.com run by NVidia?
The quote and link I gave you that you are arguing against is directly copied and pasted from NVidia announcement on what’s happening. The link you posted was a 3rd party news site.
“The DRIVE PX 2 development engine will be generally available in the fourth quarter of 2016. Availability to early access development partners will be in the second quarter. “ is what was officially said by NV directly from there site.
You said the announcement specifically stated sampling but I cannot find any reference to that. So I am putting that down to yet another thing you just made up.
Last edited:
Soldato
- Joined
- 11 Mar 2013
- Posts
- 5,470
Just ignore DM. Him and few others like to cause trouble. Don't let it bother you there not worth it.
Soldato
- Joined
- 11 Mar 2013
- Posts
- 5,470
The rumour atm is we'll see pascal at GTC with Nvidia speech April 4th.
Soldato
It won't be GP100 though, that won't show till mid next year at the earliest. More than likely GP106 will be shown, it is not like Nvidia have not tested a new process with low - mid range cards before.
Soldato
I'm still wondering if the Pascal GP104 will be 256bit or maybe 384bit, seeing as the new top end GP100 will be HBM2.
the sensible option would be 256bit, but that would make it a direct 970/980 replacement, in the pecking order (not in performance), where as there will be no 384bit parts if the top end goes HBM2 and the next one down is 256bit.
I know that these decisions would have been made a long time ago and they could not have known that HBM2 or even the GDDR5x would not be quite as available as it would seem (according to the rumours/reports). But it seems like a good idea to utilise the 384bit for the next part down from HBM2.
It could look somewhat like this.
1080ti/TitanXI HBM2
1070/1080 384bit
1060/1060ti 256bit
1050 and below 128bit
the sensible option would be 256bit, but that would make it a direct 970/980 replacement, in the pecking order (not in performance), where as there will be no 384bit parts if the top end goes HBM2 and the next one down is 256bit.
I know that these decisions would have been made a long time ago and they could not have known that HBM2 or even the GDDR5x would not be quite as available as it would seem (according to the rumours/reports). But it seems like a good idea to utilise the 384bit for the next part down from HBM2.
It could look somewhat like this.
1080ti/TitanXI HBM2
1070/1080 384bit
1060/1060ti 256bit
1050 and below 128bit
Soldato
It is a brand new process, the Polaris 10/11 and gp 104-106 will already be the largest dies tested on it at 200-300mm^2 let alone jumping up to 400+mm^2 with a huge monolithic design.
Ofcourse there will be test dies, but I highly doubt something that big is ready for the process just yet. But at least end of year or early next for big Polaris and mid next for Nvidia unless they try hard to pull it forwards on poor yields.