Nvidia Pascal Announcement

Wow, after quite a bit of silence, the 1060 seems closer than expected. So much for "We believe we are months ahead in this transition".

If it manages to reach 480X performance with a smaller die, a lower number of DRAMs, and lower production cost, the pipedream/storyline of AMD taking over the mainstream performance segment may run into some issues as well.

Well..given that GP102 and GP106 are basically GP104 (±) 50%..it's basically a Ctrl+C & Ctrl+V job for the engineers! (I kid of course..but compared to GP104..the additional engineering required is relatively trivial). Both GP102 and GP106 will be out next quarter according to my information..with GP107/8 being delayed a bit.

I believe GP106 should be able to match Polaris 10 pretty easily (while being <200 mm2). The huge clock speed disadvantage is really going to hurt AMD this gen.
 
GM204 is 75% bigger than GM206. If GP104 is only 50% bigger than GP106, it could be good times.

What's scary is that would probably put it at a smaller die, lower power consumption and higher performance than Polaris 10. Triple ouch.
 
I call for BS since there's no 14FF process available at TSMC *runs for his life*

Erinyes,
Are you sure? I heard one fart over 200 for 106?

Maybe they have a secret 14FF process and thats why Pascal is so fast?? *exits room and waits for it to show up at Videocardz...*

Yep that's what I've heard..a bit below. If GP106 is 50% of GP104..it shouldn't be more than 200 anyway.
GM204 is 75% bigger than GM206. If GP104 is only 50% bigger than GP106, it could be good times.

What's scary is that would probably put it at a smaller die, lower power consumption and higher performance than Polaris 10. Triple ouch.

Well I didn't mean GP104 is 50% bigger than GP106.I meant GP106 = ~50% of GP104's resources. Now ideally this would mean 50% of the die size..i.e. 167 mm2 but you have fixed function hardware, caches, phy, etc so of course it dosen't scale linearly like that.
 
Last edited:
Well I didn't mean GP104 is 50% bigger than GP106.I meant GP106 = ~50% of GP104's resources. Now ideally this would mean 50% of the die size..i.e. 167 mm2 but you have fixed function hardware, caches, phy, etc so of course it dosen't scale linearly like that.

I was going off the 200 + fart die size estimate :)
 
GM204 is 75% bigger than GM206. If GP104 is only 50% bigger than GP106, it could be good times.
GM206 was 57% the die size of GM204. I would expect GP106 to do slightly worse against GP104 size-wise (because GM206, while featuring updated and possibly larger video blocks, was literally half a GM204, but GP106 is supposedly having 3/4 the memory interface of GP104).
But even at a bit over 60% the die size of GP104 (314mm^2) it would still be just below 200mm^2. In any case even if it's not it should still be smaller than Polaris 10.
On a positive note, with a memory interface of 192bit nvidia can go for cheaper 7gb/s gddr5 and still have the same alu/bandwidth ratio as a 1080 (even while increasing core clocks a little).

What's scary is that would probably put it at a smaller die, lower power consumption and higher performance than Polaris 10. Triple ouch.
The maybe max ~20% advantage a GP106 might have in die size isn't too worrying imho, it's not the same fab so we don't (or at least I don't...) really know how that compares anyway. AMD should be a LOT more worried about the efficiency difference which is seemingly much larger (on the desktop something like 100W vs. 150W might not be all _that_ important but how are you supposed to get into notebooks with a huge efficiency deficit - well maybe at least Polaris 11 fares better...).
 
Now ideally this would mean 50% of the die size..i.e. 167 mm2 but you have fixed function hardware, caches, phy, etc so of course it dosen't scale linearly like that.
With smaller process geometries, the worse scaling of things like display engines, PCIe- and memory controllers compared to logic and memories (i.e. ALUs and caches) would make them take up proportionally more die space in addition to especially dispplay engines probably becoming more and more complex.

Does anyone have an idea whether or not ditching native analogue output (DVI-I and DSub) actually would help process scaling for the display engine or was it done mainly because most people wouldn't use it anymore anyway?
 
GeForce GTX 1060 Launch Imminent .....
However new info arrived on the web, the 1060 is now expected to be launched on the 7th of July, 2016. Actual availability is expected to follow a week later, on 14th July. That's way ahead of schedule as the 1060 was expected to launch later in the Summer of 2016. Here are the rumored specifications of the GeForce GTX 1060 as posted on Benchlife.info, the source for this information.

  • GPU: NVIDIA GeForce GTX 1060
  • ASIC: GP106-400-A1 and GP106-300-A1
  • 16 nm FinFET process
  • 120W TDP
  • 1280 CUDA cores, spread across 10 streaming multiprocessors
  • 80 TMUs and 48 ROP units
  • 192-bit GDDR5 memory interface
  • 6 GB and 3 GB models
http://www.guru3d.com/news-story/geforce-gtx-1060-launch-imminent.html
 
Nice! This could actually be the card to buy for VR given that unlike Polaris it does have s lot of new HW to accelerate VR apps. If it has 48 ROPs I'd expect it to pull away from the 480 at higher res/VR even if it has a lower physical memory bandwidth due to better color compression.
 

I'm surprised by the existence of the 3 GB model TBH. For a 2016 mid-range part..it seems too low. Also for marketing purposes, 6GB sounds better vs a 4 GB Rx480. The ALU/ROP ratio is also interesting..and is the opposite of AMD's approach. This should also easily beat the 970/RX 480 if its clocked similar to GP104.
Nice! This could actually be the card to buy for VR given that unlike Polaris it does have s lot of new HW to accelerate VR apps. If it has 48 ROPs I'd expect it to pull away from the 480 at higher res/VR even if it has a lower physical memory bandwidth due to better color compression.

I dont think 1060 has any shortage of bandwidth. If we compare Rx480 to GTX 1070..even though they have the same bandwidth..the 1070 is about 50% faster. Now if 1060 has 8 Gbps memory, its actually in a better position since it has 75% of the bandwidth of 1070 but only ~66% the ALU/tex. Now they could use 7 Gbps GDDR5 in which case it would only be ~65% of the bandwidth so that's a possibility as well.
 
http://videocardz.com/61753/nvidia-geforce-gtx-1060-specifications-leaked-faster-than-rx-480

1060 slides

NVIDIA-GeForce-GTX-1060-vs-Radeon-RX-480-performance-1-900x467.jpg


hmm typical Marketing slide.
 
usually they don't, but the rx480 does fill in a big hole in the $200 range which nV has no answer for till the 1060.
 
I can see nVidia doing lots of comparisons to the RX 480--especially when it comes to power draw. I think they will take full advantage of the bad publicity the power issues are getting.
 
Back
Top