NVIDIA Pascal Thread

[showb1z]

Yeah...I recently hosted a poll on this site and the majority of the people on this forum, per the poll, are actually "AMD fans." Very few identified as fans of NVIDIA.

Though they easily make up for this numerical disadvantage in sheer trolling output.

 

[Sweepr]

So the HBM2-based Vega is a 2017 product. Will AMD have a GP104 competitor this year or Polaris is aimed at lower price points?

 

[Kenmitch]

So the HBM2-based Vega is a 2017 product. Will AMD have a GP104 competitor this year or Polaris is aimed at lower price points?

Were can I buy a GP104?

 

[Sweepr]

Were can I buy a GP104?

When it launches, sometime after Computex according to the current rumours. Is Polaris 10 the competitor?

 

[Kenmitch]

After it launches, sometime after Computex according to the current rumours. Is Polaris 10 the competitor?

You were suppost to say you can't.

Then I'd have replied with that's my point.

Not sure....I guess once them there gpus are in the hands of reviewers we'll find out.

 

[Sweepr]

Looking more and more likely that the first Pascal/Polaris GPUs are small chips with clear focus on perf/watt rather than moving past Fiji/GM200 raw performance.

 

[3DVagabond]

Looking more and more likely that the first Pascal/Polaris GPUs are small chips with clear focus on perf/watt rather than moving past Fiji/GM200 raw performance.

In the end perf/W is the ultimate limiting factor for raw performance.

 

[xorbe]

In the end perf/W is the ultimate limiting factor for raw performance.

The weakest link will be the limiting factor, which could be a great number of things. You can certainly bottleneck before hitting thermal limits.

 

[Sweepr]

Pascal-based GM204 & GM206 successors set to launch shortly after Computex

As the focus of the launch of Pascal is expected to be on laptops, it's probably not about GP100 - the next monster circuit with an assumed number of transistors on the whole 17 billion . Instead, it is probably an outright replacement for GM204 and GM206, which is used by the Geforce GTX 965M up to the GTX 980th

www.sweclockers.com/nyhet/21783-nvidia-geforce-pascal-till-computex-2016

If this late February rumour is right, 2016 is all about GP104/GP106 vs Polaris 10/11 as far as gaming GPUs are concerned. Sadly we will have to wait till April for more info on NVIDIA's side.

 

[3DVagabond]

The weakest link will be the limiting factor, which could be a great number of things. You can certainly bottleneck before hitting thermal limits.

I didn't say the only limiting factor. If you eliminate all the other factors then the thermal limit "ultimately" will be the limiting factor.

 

[3DVagabond]

www.sweclockers.com/nyhet/21783-nvidia-geforce-pascal-till-computex-2016

If this late February rumour is right, 2016 is all about GP104/GP106 vs Polaris 10/11 as far as gaming GPUs are concerned. Sadly we will have to wait till April for more info on NVIDIA's side.

That roadmap is from a year ago.

 

[Sweepr]

http://cdn.wccftech.com/wp-content/uploads/2015/03/Screenshot-91.png

That roadmap is from a year ago.

Illustrative purpose, pretty sure that's not where the info comes from.

 

[Sweepr]

Guru3D - Nvidia GeForce X80 and X80 Ti Pascal Specs?

www.guru3d.com/news-story/nvidia-geforce-x80-and-x80-ti-pascal-specs.html

 

[Kenmitch]

Guru3D - Nvidia GeForce X80 and X80 Ti Pascal Specs?

www.guru3d.com/news-story/nvidia-geforce-x80-and-x80-ti-pascal-specs.html

As stated it's most likely a fake....Time will tell in the end I guess.

 

[MrTeal]

Is 8GT/s on the memory even feasible at this point in a shipping GPU using GDDR5? I didn't think most people were able to hit that speed on the 980Ti even with a good boost in voltage.

Of course, besides that minor point there's the pretty major red flag of GP100 running both HBM2 and GDDR5.

 

[ShintaiDK]

It does look fake. GP100 should be HBM2 only as well. The mix makes no sense. And I doubt GP104 will be more than 3000SP or something.

 

[Det0x]

www.sweclockers.com/nyhet/21783-nvidia-geforce-pascal-till-computex-2016

If this late February rumour is right, 2016 is all about GP104/GP106 vs Polaris 10/11 as far as gaming GPUs are concerned. Sadly we will have to wait till April for more info on NVIDIA's side.

Let me help you translate some of the text from sweclockers

There is whipping up a storm on the graphics card market. SWEclockers experience that Nvidia had plans to release Pascal early in the summer, but problems may force the launch to be postponed.

..focus mainly will be on gaming notebooks.

The plans may have to change. According to the same data, there is lack of clarity around when Pascal is ready, because Nvidia have encountered problems. The aim was to launch in mid-June, but the likelihood that the launch window will be moved to a later date, are described as large.

As the focus of the launch of Pascal is expected to be on laptops... ...it is probably an outright replacement for GM204 or GM206, which is used by the Geforce GTX 965M up to the GTX 980th

 

[Sweepr]

Looking at AMD's strategy with Polaris 10 I don't think NVIDIA is going crazy with GP104 (4096 CUDA cores), but I like to be surprised. GDDR5 (not GDDR5X) and HBM2 for GP100 sounds unlikely.

Anyway, if they at manage to match/beat Geforce GTX980 Ti they will have the fastest card till GP100/Vega roll out.

 

[MrTeal]

It would be unfortunate if GP104 is delayed, as that's really (along with GP100) the part that most people here are waiting for.

Given the 8TFLOPS figure given for Drive PX2, it seems reasonable that a single GP106 might come in somewhere around the performance of GM204 (4.6TFLOPS).

 

[poofyhairguy]

As stated it's most likely a fake....Time will tell in the end I guess.

I for one hope its fake. Could you imagine the prices of cards with that many shaders? The Titan would be a $2k card.

 

[xorbe]

I find it a little hard that they would mix memory types like that. I'd be down for an X80 though, if true. Supposedly all specs above Titan X, except 6GB, that works for me.

 

[96Firebird]

Guys, it has to be real. It says "NVIDIA Confidential" on the bottom!

 

[NTMBK]

Guru3D - Nvidia GeForce X80 and X80 Ti Pascal Specs?

www.guru3d.com/news-story/nvidia-geforce-x80-and-x80-ti-pascal-specs.html

I'll repost what I said in the other thread...

"Those specs are BS. I massively doubt that we will see a single GPU which can use both GDDR5 and HBM. The interface is completely and utterly different, you would need to have both a full set of GDDR5 controllers and HBM controllers, with a pinout which somehow has interposer-compatible microbumps plus PCB-compatible full-sized bumps... I don't see how you could make that work. Unless you had a "dummy" interposer even for the GDDR5 part, which just passed through pins to the PCB...
"

 

[Glo.]

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

 

[Stuka87]

I'll repost what I said in the other thread...

"Those specs are BS. I massively doubt that we will see a single GPU which can use both GDDR5 and HBM. The interface is completely and utterly different, you would need to have both a full set of GDDR5 controllers and HBM controllers, with a pinout which somehow has interposer-compatible microbumps plus PCB-compatible full-sized bumps... I don't see how you could make that work. Unless you had a "dummy" interposer even for the GDDR5 part, which just passed through pins to the PCB...
"

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