PCTuning fórum

PET5 píše:Jasně, že to bude nová architektura. Maxwell přece není schopen výpočtů ve DP. Pascal snad bude a zase umožní konstrukci výpočetních karet Tesla.

HEAD píše:Ja sem si 100% jistej, ze pascal nebude jen maxwell s HBM, ale bude to nova architektura.
NV ma miliardy na vyzkum a urcite se ty 2 roky nestourali v nose a nic nedelali

The timeframe is why I think Pascal is just going to be what Nvidia promised Maxwell would originally be but with HBM.


Announced in mid 2014 shipping 2016. With Volta announced a year before.

Assuming that NVidia immediently started working on Pascal after announcing Volta that puts 3 years total for a middle-of-the-year 2016 release.

Shrink Maxwell, fix parts of Maxwell, figure HBM controller out.

Well, I imagine Volta was planned for 2016 previously. All that NVidia really had to do was to go back to Volta and tweak it, possibly make the parts larger than originally planned as process will be much more mature when it launches.


They can than mostly reuse Maxwell for a whole generation of GPUs called Pascal and get another 1.5-2.5 years of money from near zero R&D.

Designing a GPU uarch takes more than just 1½ yrs. to 2 yrs. from what I understand. Going from the initial design process all the way to volume production and market release is pretty much a 4+ year endeavor, right? So I don't see how they could instantly make changes that quickly without massively screwing things up. In which case AMD will have the distinct hardware superiority in DX12 & Vulkan and Nvidia will have to rely on their usual scummy PR and "friendly press" to do their dirty work of "correcting peoples' vision" to the truth that Pascal would be, actually, superior to GCN 2.0.

I suppose if they already had a design somewhat ready/totally ready and they decided to add in some async hardware, they could potentially do it without a complete re-work? But it would be more cash burned if they'd already taped out the original design and went for masks and early silicon and such for tweaking, then having to re-do it all. Not like it'd matter in the end though, besides them still having a rather embryonic design compared to AMD who has had a leg up in this regard for years already.

hnizdo píše:Jestli Pascal=Maxwell+HBM, tak to nedava zadny smysl. Maxwell bandwidth HBM nepotrebuje, dokonce HBM2. Nulovy benefit, velke naklady.
Takze hlasuju Pascal=nova architektura.

Krteq píše:Hlasování ohledně toho by se celkem šiklo Co vy na to modi?