NVIDIA Pascal Thread

[NTMBK]

Even if X80Ti is wrong, and I assume it is, however X80, and Titan might be reality.

Why would Nvidia put out a spec sheet where a third of the specs are fabricated, but the rest are real? Either it's all real, or it's a fake.

 

[Glo.]

Why would Nvidia put out a spec sheet where a third of the specs are fabricated, but the rest are real? Either it's all real, or it's a fake.

Its not what I meant . I think the guy, who put it all up, unintentionally got it right in few places. I genuinely believe that GP100 is 6144 CUDA cores and GP104 is 4096. Rest is meaningless, in this case.

 

[jpiniero]

That is totally fake. Guru3D needs to up their fake specs for clickbait game.

You won't see a full GP100 Titan any time soon, likely not until the end of 2017. It'll be cut, probably severely. I'd put the O/U at 4668 cores.

I'd put the O/U for GP104 at 3200. Between that, higher stock clocks, and 128 ROPs would be enough to get enough of a distance between it and the 980 Ti.

 

[Kris194]

These specs are faker than anything I saw regarding new GPUs.

 

[CakeMonster]

So if you are implying that this chip is the one they'll be using for 2 years, just enabling more of it... then at least the initial chips would probably be running very cool and quiet. I can at least appreciate that.

 

[xorbe]

I agree, its either GDDR5 or HBM. It cannot support more than one memory type without a die change.

Package / uncore engineers are clever -- it could. But it seems unlikely.

 

[Fallen Kell]

Package / uncore engineers are clever -- it could. But it seems unlikely.

I agree. This is no different than Intel having a chipset that supported DDR and DDR2 back in the day, as well as DDR3 and DDR4 right now in the CPU. Nothing prevents you from putting the needed controller together that supports multiple specs except for the space required by it and the cost of debugging it. That being said, the GDDR5 controller is already fully designed, so the only part really being debugged is the new HBM2 and the method to set which feature-set to use.

 

[Sweepr]

Pascal entries @ 3DMark 11?

Nvidia Pascal 3DMark 11 Performance Entries Spotted – Potential GTX 1080, GTX 1070 & GTX 1060 Benchmarks

http://www.3dmark.com/3dm11/10969217
http://www.3dmark.com/3dm11/10985139
http://www.3dmark.com/3dm11/11061015

Discuss.

 

[airfathaaaaa]

love the memory on the last one sound like a dejavu

 

[ShintaiDK]

Pascal entries @ 3DMark 11?

Nvidia Pascal 3DMark 11 Performance Entries Spotted – Potential GTX 1080, GTX 1070 & GTX 1060 Benchmarks

http://www.3dmark.com/3dm11/10969217
http://www.3dmark.com/3dm11/10985139
http://www.3dmark.com/3dm11/11061015

Discuss.

Looks good if true. Very interesting indeed.

 

[Head1985]

Epic fail if true
19K GPu score is on GTX980 level.My GTX970 1500/8000 have 18k in 3dmark 11 GPU score.
OC GTX980 is above 20k.
This cant be 1070 or its mega Fail.

 

[jpiniero]

Looks good if true. Very interesting indeed.

Not really. A 980 Ti gets like 20k unless I am missing something. Mobile part perhaps?

 

[IllogicalGlory]

Uses desktop motherboards, and apparently only 2GB of system RAM (on the two "1060s")?

The 980 Ti-esque score, if this is true, is most likely the 1070 based on its 7680 MB (cut GP104 with ROPs, cache and memory bus disabled in the vein of the 970, 660Ti and 570) of memory. Not bad at $400.

 

[Head1985]

Pretty bad for 400USD.And its not at 980TI level.Its on GTX980 level.OC 980 have above 20k score.

 

[crisium]

480-bit memory? Whoa.

 

[iiiankiii]

Looks good if true. Very interesting indeed.

What looks good? Serious question.

 

[MrTeal]

You need to look at the graphics score. The overall score is dragged way down by the i3-2100.

 

[casiofx]

That card with full 8GB (1080?) gets same score with 980?
Is this downclocked engineering sample or it isn't 1080 but 1070. Scores kinda disappointing

 

[ShintaiDK]

Not really. A 980 Ti gets like 20k unless I am missing something. Mobile part perhaps?

Its low clocked core at 540Mhz. I expect retail to be at least twice that.

 

[Erenhardt]

maybe it is 8gb gtx980...

 

[jpiniero]

Its low clocked core at 540Mhz. I expect retail to be at least twice that.

OK, I missed that. So it could be an GP204 ES.

 

[casiofx]

Its low clocked core at 540Mhz. I expect retail to be at least twice that.

No such thing, it is a bug by 3dmark for unidentified cards. Same case as last time for unreleased cards.

Sorry to shatter your hope for a card that'll double the performance, lol

 

[IllogicalGlory]

Yeah, here's another Generic VGA (actually Iris 6100) that displays core clock as 300: www.3dmark.com/3dm11/10428118

That is the base/lowest clock state for that particular chip, but it boosts up to 1100.

 

[ShintaiDK]

Because 3Dmark reports base clocks I guess.

http://www.3dmark.com/3dm11/9879250
http://www.3dmark.com/3dm11/9924726
http://www.3dmark.com/fs/6724606

 

[casiofx]

Because 3Dmark reports base clocks.

http://www.3dmark.com/3dm11/9879250
http://www.3dmark.com/3dm11/9924726

And? What does the link have to do with 540mhz?

Check this engineering sample gtx970 instead

http://www.3dmark.com/3dm11/10729272