NVIDIA Pascal Thread

[tential]

Nvidia never hypes their products so far ahead prior launch. Wait for April @GTC to learn more about the architecture, and then another press event shortly before launch. It worked before so why shouldn't they keep this format?

So if Nvidia never hypes products before launch you have no idea when it is launching....

 

[tweakboy]

Wow I wonder about the price on this. gl

 

[3DVagabond]

And Polaris so far is just lots of hot air produced by desperation.

No. Polaris actually exists. Been shown working. Pascal, on the other hand, we've only seen fake mock ups.

 

[Pinstripe]

No. Polaris actually exists. Been shown working. Pascal, on the other hand, we've only seen fake mock ups.

That doesn't make me wrong about my previous statement that Nvidia keeps it's treasure close to the chest before they launch. Pascal is real and it's coming, and it will take the wind out of AMD's sails (again).

 

[Silverforce11]

That doesn't make me wrong about my previous statement that Nvidia keeps it's treasure close to the chest before they launch. Pascal is real and it's coming, and it will take the wind out of AMD's sails (again).

Cool, hopefully its real and coming soon. I would like some cheap Polaris SKUs if possible, don't wanna pay jacked up prices due to no competition.

 

[Krteq]

That doesn't make me wrong about my previous statement that Nvidia keeps it's treasure close to the chest before they launch. Pascal is real and it's coming, and it will take the wind out of AMD's sails (again).

Hmm, I would like to share your optimism.

 

[3DVagabond]

That doesn't make me wrong about my previous statement that Nvidia keeps it's treasure close to the chest before they launch. Pascal is real and it's coming, and it will take the wind out of AMD's sails (again).

You said Polaris is just hot air. So, it does make you wrong.

 

[caswow]

That doesn't make me wrong about my previous statement that Nvidia keeps it's treasure close to the chest before they launch. Pascal is real and it's coming, and it will take the wind out of AMD's sails (again).

might place 200 pascals in your trunk thats how powerfull 1 pascal titan Y will be D:

 

[Pinstripe]

Cool, hopefully its real and coming soon. I would like some cheap Polaris SKUs if possible, don't wanna pay jacked up prices due to no competition.

Who said Polaris will be cheap? Unless AMD wants to lose money again.

 

[myocardia]

You said Polaris is just hot air. So, it does make you wrong.

No, he said "So far, Polaris is just hot air", which it is, along with Pascal...unless you happen to know where we can all go buy ourselves a Polaris card?

 

[NTMBK]

Bench-life just leaked some juicy info about Pascal (GP104). Unfortunately Chrome's translation is poor, so I would appreciate if someone could shed some light here:

NVIDIA GeForce GTX 1080 launching in May, GDDR5 / GDDR5X instead of HBM2



https://benchlife.info/gp104-aka-nvidia-geforce-gtx-1080-will-ship-in-may-and-no-hbm2-031112016

- Pascal products being announced in April @ GTC
- TSCM 16nm FinFET
- Geforce 'GTX 1080' debuting in May
- Reference model(?) Geforce 'GTX 1080' including PCIe 8 PIN, DisplayPort x 2, HDMI x 1 shipping/launching May 27th(?)
- Custom OEM models available after July
- GTX 1080 sports 8GB of GDDR5 / GDDR5X
- GP100 is using HBM2 as expected
- NVIDIA using Samsung's HBM2 (Hynix for AMD)

If it's launching in May, then it's GDDR5 not GDDR5X. GDDR5X isn't going into volume production until the summer: https://www.micron.com/about/blogs/2016/february/gddr5x-has-arrived

 

[Silverforce11]

No, he said "So far, Polaris is just hot air", which it is, along with Pascal...unless you happen to know where we can all go buy ourselves a Polaris card?

Did you see the followup comment? The guy claims Polaris is just hot air, but Pascal is REAL, it's coming soon! For real!

 

[caswow]

No, he said "So far, Polaris is just hot air", which it is, along with Pascal...unless you happen to know where we can all go buy ourselves a Polaris card?

i rather take a closed showing of an actual chip that was proved by journalists than a lying ceo holding up a mockup AGAIN claiming it ist pascal what they are seeing D:

 

[jpiniero]

If it's launching in May, then it's GDDR5 not GDDR5X. GDDR5X isn't going into volume production until the summer: https://www.micron.com/about/blogs/2016/february/gddr5x-has-arrived

Yeah which has me wondering how legit the info is, although it's possible that it could ship with GDDR5X later.

1x8pin also mean lower power consumption than GTX970/GTX980. Tho you can find a GTX970 Mini with 1x8pin.

That's probably true but according to AT's review the ref 980 is 2x6-pin. So if this is legit the power consumption is likely comparable but not lower. Which does make sense.

 

[PPB]

Lol, a lot of 970s are 6+6 or 8 pin, dont let his BS get through anybody

 

[Silverforce11]

8 pin is 150W, add MB power and max 225W. Same as 2x 6 pins.

 

[Sweepr]

If it's launching in May, then it's GDDR5 not GDDR5X. GDDR5X isn't going into volume production until the summer: https://www.micron.com/about/blogs/2016/february/gddr5x-has-arrived

Honestly I'm not 100% sure if the translation is right. Maybe what their saying here is shipping in late May, with lots of talk / demonstrations @ Computex and launch at a later date (from July onwards?).

Would 256-bit with the fastest GDDR5 available be a problem if it's launching in May? Maybe they're planning a higher-clocked full GP104 dGPU with GDDR5X at a later date (H2-2016). Sounds more realistic than a 512-bit bus for the mainstream (and likely small) GP104.

 

[moonbogg]

It looks like this will be the mid range card. I will not buy.

 

[tential]

It looks like this will be the mid range card. I will not buy.

What? Why not? You don't wanna get this chip, then the Titan then the 1080ti?

 

[Sweepr]

Would you guy be interested in something like this?

Geforce 'GTX 1080'
- +$500
- Full GP104 die
- Relatively small die and significant OC headroom (>20%)
- 15-25% faster than Geforce GTX980 Ti
- 256-bit / 512-bit GDDR5, new model @ H2 with GDDR5X

Geforce 'GTX 1070'
- +$400
- Relatively small die and significant OC headroom (>20%)
- Slightly faster than Geforce GTX980 Ti
- 256-bit / 512-bit GDDR5, new model @ H2 with GDDR5X

 

[Glo.]

Has ever first GPUs from Nvidia on new node had good OC headroom? Or only on mature process with mature architecture?

Lets not forget that this is 16 nm, a node after tradition for Nvidia, and that usually does not go well for them.

However, I expect doubling on performance of Nvidia GPUs on 16 compared to Maxwell. So around 300 mm2 GTX 1080 with 8 GB of RAM GDDR5, and 4096 CUDA cores.

 

[hawtdawg]

This is about what I've been expected. I assumed they would do what they did with Kepler, and milk it over a longer period of time since A: they made a buttload of cash and B. No one knows how long we'll be on 16nm.

 

[tential]

Would you guy be interested in something like this?

Geforce 'GTX 1080'
- +$500
- Full GP104 die
- Relatively small die and significant OC headroom (>20%)
- 15-25% faster than Geforce GTX980 Ti
- 256-bit / 512-bit GDDR5, new model @ H2 with GDDR5X

Geforce 'GTX 1070'
- +$400
- Relatively small die and significant OC headroom (>20%)
- Slightly faster than Geforce GTX980 Ti
- 256-bit / 512-bit GDDR5, new model @ H2 with GDDR5X

You think they'll double release the 1080/1070 That close together?(gddr5 version then 5x version. That's not the Nvidia I know..

 

[hawtdawg]

Has ever first GPUs from Nvidia on new node had good OC headroom? Or only on mature process with mature architecture?

Lets not forget that this is 16 nm, a node after tradition for Nvidia, and that usually does not go well for them.

However, I expect doubling on performance of Nvidia GPUs on 16 compared to Maxwell. So around 300 mm2 GTX 1080 with 8 GB of RAM GDDR5, and 4096 CUDA cores.

Kepler OC'd decently well. usually by +200 to +300 mhz. Not as good as Maxwell necessarily, but still pretty good.

 

[Sweepr]

I think ~50% faster than Geforce GTX980 with 256-bit GDDR5 wouldn't be bad at all. Doubling performance with their first 16nm products sounds a bit too optimistic.

You think they'll double release the 1080/1070 That close together?(gddr5 version then 5x version. That's not the Nvidia I know..

What do you suggest? Current rumours point to ~June(ish) launch and according to Micron GDDR5X is not read for prime time. Unless GP104 is not bandwidth limited at all, then might as well save GDDR5X for future products.