NVIDIA Pascal Thread

[antihelten]

Depends on your capabilities, I guess. We have a single machinist to support our small (half-dozen Engineer) business unit, and we have an EDM machine for mold making. They wouldn't be common in a hobbyist garage, but they're far from uncommon. If I were so motivated I could have a shroud made up with "490X" written on it and shaped like a giant penis, though the company might not find that to be a particularly wise use of resources.

That being said, it would be a dedicated troll. Props to him if it is.

The issue isn't so much about having the machinery to make the mold (you could do that on a CNC mill), but rather about having the machinery to use said mold for aluminium injection molding (i.e. not just the usual plastic injection molding).

 

[MrTeal]

there was a hope to begin with? they clearly stated vega would be an early 20178 card

Jesus. Guess I'm definitely going nVidia this time.

 

[96Firebird]

That being said, it would be a dedicated troll. Props to him if it is.

It would be, and they would be a smart troll leaving the second-to-last number out to allow for machining of the 7 or 8 or whatever number they decide to put there.

So, technically, the shroud is both molded and machined.

 

[Kenmitch]

Jesus. Guess I'm definitely going nVidia this time.

Not even sure if our dust will be around by then.

 

[antihelten]

Jesus. Guess I'm definitely going nVidia this time.

Don't worry, AMD will have a bunch of really nice rebrands to cover that 18,162 year gap.

 

[railven]

Don't worry, AMD will have a bunch of really nice rebrands to cover that 18,162 year gap.

Well, there is all this talk of modular GPUs. Polaris 10 Erector Set patent pending.

 

[Timmah!]

there was a hope to begin with? they clearly stated vega would be an early 20178 card

If i had to save cash for a new GPU until 20178, i suppose i could buy entire AMD by then

 

[R0H1T]

How long before we get a graphics card from AMD or Nvdia as powerful as a Gtx960 for under $100? The entry level video card market has been stagnant for far too long and Intel lighting fire under AMD's and Nvdia's arses hasn't led to anything.

Not anytime before 2017 at the earliest, besides what you're talking about is the notebook market & I doubt Intel can come even close to the 750Ti (or 950) with their best IGP two years from now.

there was a hope to begin with? they clearly stated vega would be an early 20178 card

I'm not sure humans would still exist on this earth by then, global warming & all that D:

 

[mohit9206]

Not anytime before 2017 at the earliest, besides what you're talking about is the notebook market & I doubt Intel can come even close to the 750Ti (or 950) with their best IGP two years from now.
I'm not sure humans would still exist on this earth by then, global warming & all that D:

I'm talking about desktop market. In USA $99 bought you hd4670 in 2008, hd6670 in 2011 and r7 250x in 2014 so im hopefully expecting r7 370 level performance by 2017.
Since you're indian, you would know in India $100 only buys you an r7 250/gt740 that is near useless for playing ps4 and xbox one ports. We need atleast 750ti level card for under $100.
Games are getting extremely demanding and entry level gpu is no longer enough to maintain even 30fps average.
Need a faster Pascal/Polaris card under $100.

 

[antihelten]

Well, there is all this talk of modular GPUs. Polaris 10 Erector Set patent pending.

Maybe AMD could do a tie in with the Voltron franchise, I hear that their marketing department is looking for good ideas.

 

[xorbe]

If i had to save cash for a new GPU until 20178, i suppose i could buy entire AMD by then

That's $82590 a year to reach $1.5B ... D:

 

[Termie]

I'm talking about desktop market. In USA $99 bought you hd4670 in 2008, hd6670 in 2011 and r7 250x in 2014 so im hopefully expecting r7 370 level performance by 2017.
Since you're indian, you would know in India $100 only buys you an r7 250/gt740 that is near useless for playing ps4 and xbox one ports. We need atleast 750ti level card for under $100.
Games are getting extremely demanding and entry level gpu is no longer enough to maintain even 30fps average.
Need a faster Pascal/Polaris card under $100.

The answer is you will never see this. Neither AMD nor Nvidia has shown any interest in this market since 2012. Bringing out new tech at <$100 makes no business sense, and the last brand-new card to debut at $100 or less was the GT 740. Not going to happen ever again given inflation and the shrinking low-cost discrete GPU market.

 

[Kenmitch]

That's $82590 a year to reach $1.5B ... D:

It will be called SEMD by then....I'll let you ponder what it stands for.

 

[maddie]

I'm talking about desktop market. In USA $99 bought you hd4670 in 2008, hd6670 in 2011 and r7 250x in 2014 so im hopefully expecting r7 370 level performance by 2017.
Since you're indian, you would know in India $100 only buys you an r7 250/gt740 that is near useless for playing ps4 and xbox one ports. We need atleast 750ti level card for under $100.
Games are getting extremely demanding and entry level gpu is no longer enough to maintain even 30fps average.
Need a faster Pascal/Polaris card under $100.

Since all of the next gen designs are finalized as of now, your best bet would be that Polaris 11 shocks us with very good performance and low price. Nvidia will have to respond. With everything happening worldwide, it might happen.

 

[swilli89]

Since all of the next gen designs are finalized as of now, your best bet would be that Polaris 11 shocks us with very good performance and low price. Nvidia will have to respond. With everything happening worldwide, it might happen.

Wonder if this cooling shroud will be the next design faux pas? Possibly overdoing it on the angles.. which kinda reminds me of..

At least those screws are definitely metallic and not wood!

 

[dacostafilipe]

No idea if already posted.

Pascal architecture whitepaper : http://images.nvidia.com/content/pdf/tesla/whitepaper/pascal-architecture-whitepaper.pdf

Before Pascal, on
systems where compute and display tasks were run on the same GPU, long-running compute kernels
could cause the OS and other visual applications to become unresponsive and non-interactive until the
kernel timed out.

... with Compute Preemption in GP100,
those applications can now run as long as they need when processing large datasets or waiting for
specific conditions to occur, while visual applications remain smooth and interactive ...

 

[Silverforce11]

@NeoLuxembourg

I posted about that change already a few times.

Pascal finally has PROPER and FUNCTIONAL preemption. NVIDIA has been lying about Maxwell's capability in this aspect, where they tell devs yes they have it... except it doesn't work, even their priority preemption queues can get stuck behind a draw call. -_-

Pascal will have half of what's needed for proper Async Compute. Basically it will no longer require a slow context switch for graphics <-> compute queues. If there's a graphics draw call in the pipe, a priority compute queue can immediately suspend that graphics call and gets processed.

This is what they refer to as fine-grained preemption. It also means Pascal will not suffer like Maxwell and Kepler when games use a lot of compute, since it can switch much faster, there's no idle time between graphics/compute tasks that causes a performance loss the more compute is used.

It's a great change for modern games and VR. I think they will finally be able to get motion to photon latency below their current best of 25ms and it will be heavily advertised for VR.

 

[tviceman]

I don't think GP100 will ever get a consumer release. With so much more emphasis in DP (proportional to the chip) vs. GK110 and GF114, coupled with dedicated transistors to new features specific to HPC, I think the GP102 rumors will prove true. I never believed the GK102/112 rumors, either. A mythical GP102 could be 50% more than GP104 in cores and memory bus, come in at 450mm2, and end up faster than GP100 in graphics.

 

[Silverforce11]

I don't think GP100 will ever get a consumer release. With so much more emphasis in DP (proportional to the chip) vs. GK110 and GF114, coupled with dedicated transistors to new features specific to HPC, I think the GP102 rumors will prove true. I never believed the GK102/112 rumors, either. A mythical GP102 could be 50% more than GP104 in cores and memory bus, come in at 450mm2, and end up faster than GP100 in graphics.

GP102 can exist when 16nm FF is more mature and volume is even higher. TSMC in their recent financial conference said only this year will they hit their ~20% revenue target from 16nm FF.

Last year, they were saying, they are on track to be ~30% by this year. So we can see they overestimate, all the time. That's their track record.

Now reality hits and every major mobile and network chip makers are competing with NVIDIA for 16nm wafers.

Under this wafer supply limited scenario, what makes more sense?

1. Use most of the wafers you get to make GP100, sell 8x (harvested chip even!) chips for $127K.

2. Use some wafers for GP102 that's gaming focused and sell each of those as $1K Titan SKUs, the rest harvested for $799 and $599 GTX SKU?

A major point to consider:

GP104 has a good FP32 performance, with it's smaller die size, it will yield much better than GP102 will. If a Deep Learning Tesla based only on FP32 is required, GP104 will be the better choice for profits. A 2x GP104 Tesla could also squeeze into the 300W limit and offer amazing FP32 performance for the HPC market that requires it.

Maybe we will see GP102 sometime next year.

 

[tviceman]

Q4 for GP102, if it exists. Full version will be the next Titan, and cut down version will be the next x80 TI. As gddr5x ramps up, I expect Pascal to be refreshed quickly with many parts having significantly faster memory and slightly higher clocks for an easy refresh with noticeable performance bumps for said parts.

 

[DooKey]

@NeoLuxembourg

I posted about that change already a few times.

Pascal finally has PROPER and FUNCTIONAL preemption. NVIDIA has been lying about Maxwell's capability in this aspect, where they tell devs yes they have it... except it doesn't work, even their priority preemption queues can get stuck behind a draw call. -_-

Pascal will have half of what's needed for proper Async Compute. Basically it will no longer require a slow context switch for graphics <-> compute queues. If there's a graphics draw call in the pipe, a priority compute queue can immediately suspend that graphics call and gets processed.

This is what they refer to as fine-grained preemption. It also means Pascal will not suffer like Maxwell and Kepler when games use a lot of compute, since it can switch much faster, there's no idle time between graphics/compute tasks that causes a performance loss the more compute is used.

It's a great change for modern games and VR. I think they will finally be able to get motion to photon latency below their current best of 25ms and it will be heavily advertised for VR.

Supposedly cuda works just fine. It may be that NV wants to sell pascal at the same rate they sold maxwell so they aren't exposing capability in directx. I'm not defending that from a consumer POV, but it may be the case since it does make good business sense when they have such a dominant market position.

 

[Silverforce11]

Supposedly cuda works just fine. It may be that NV wants to sell pascal at the same rate they sold maxwell so they aren't exposing capability in directx. I'm not defending that from a consumer POV, but it may be the case since it does make good business sense when they have such a dominant market position.

That's the rumor, no credible independent review has actually tested graphics + compute parallel workloads even in CUDA.

Either way, Pascal resolves one of the important hurdles, with fine-grained preemption possible, it's a big step up for VR and any compute heavy games.

 

[poofyhairguy]

The answer is you will never see this. Neither AMD nor Nvidia has shown any interest in this market since 2012. Bringing out new tech at <$100 makes no business sense, and the last brand-new card to debut at $100 or less was the GT 740. Not going to happen ever again given inflation and the shrinking low-cost discrete GPU market.

You have to hope eventually the 750 ti just lives below $100, maybe rebranded as a 1040 or something.

 

[Sweepr]

Zolkorn says the cards are ready, should be launched next month (reference models), while custom/manufacturer models will be shown at Computex. They also hint that AMD postponed their cards in order to be more competitive. Help with the translation would be appreciated.

NVIDIA Pascal GTX 1080 and GTX 1070 could come sooner than expected. See you next month!

www.zolkorn.com/news/nvidia-might-launch-geforce-gtx-1000-series-soon

 

[Kris194]

Now I care only about one thing, when reviews will be released.