NVIDIA: GP104 is in full production, yields are good (custom VGAs in early June)

[3DVagabond]

He wasn't asked anything that required an answer direct enough to be called out on. Pascal is in full production and yields are good. Are there specific numbers you have to hit to make that statement? No. Just vagaries that nobody can be held to anything about.

 

[Abwx]

Is it as effective as hardware support?

No, otherwise AMD wouldnt have implemented a hardware solution.

Correct. If directly asked on a conference call and yields are bad, they will either avoid the question, be vague (not good), or answer truthfully and directly.

When you have something in quantities you ll try to sell as much as possible and not do a strange launch with reduced availability.

Was there a founder edition for the previous GPUs like Maxwell?..

No, because it was 28nm and readily available in huge quantities..

 

[exar333]

No, otherwise AMD wouldnt have implemented a hardware solution.


When you have something in quantities you ll try to sell as much as possible and not do a strange launch with reduced availability.

Was there a founder edition for the previous GPUs like Maxwell?..

No, because it was 28nm and readily available in huge quantities..

Agree on async shaders...

On availability. What can you provide in terms of proof that quantities are limited? The founder's edition alone isn't enough to justify if those are limited. Custom acrds will be available 1st/2nd week of June. 16/14nm are NOT new production nodes for the industry, and NV obviously selected only one SKU to use GDDRX. Even that production has been rumored to be pretty successful.

Every new NV GPU has been relatively tough to get for years. Pick one up on launch morning or wait a few weeks as availability is generally 'limited' as stock pops in and out. It is just supply and demand. No different than AMD top-end offering IMHO.

 

[tviceman]

So glad I sold my rig during my house move; GP104 hype is starting to get to me. By the time I am ready to move, Vega 11 should be close and the leaks should be rolling. I'm holding out until then to see where prices are.

 

[maddie]

He wasn't asked anything that required an answer direct enough to be called out on. Pascal is in full production and yields are good. Are there specific numbers you have to hit to make that statement? No. Just vagaries that nobody can be held to anything about.

Exactly.

What in the world is a good yield. It most probably changes over time anyhow and cannot be pinned down to a single precise number. All those claiming such honesty and truthfulness when talking to investors should try to do some basic reasoning. I say yields are good, who are you to dispute me. What is good in absolute, non relative terms anyway?

I challenge anyone here to give me the actual yield value if they are being described to investors as good, by Mr Jen-Hsun Huang.

 

[Sweepr]

Agree on async shaders...

On availability. What can you provide in terms of proof that quantities are limited?

He can't. This is FUD based on the fact that NVIDIA managed to launch FinFET products before the competition.

 

[Creig]

So another confirmation that Pascal has Async Compute. Good

Actually, it doesn't sound good at all.

For DirectX 12 and VR, the term Asynchronous Compute was thrown around, especially since AMD Radeon-based cards were beating Nvidia GeForce cards in DirectX 12 titles such as Ashes of The Singularity and Rise of the Tomb Raider. We were told that the Pascal architecture doesn’t have Asynchronous Compute, but that there are some aspects of this feature which qualified the card for ‘direct12_1’ feature set.

http://vrworld.com/2016/05/10/pascal-secrets-nvidia-geforce-gtx-1080/

In other words, while Pascal may have some Asynchronous Compute components, it won't have the full abilities that Hawaii has or that Polaris & Vega will have.

 

[Sweepr]

The same VRWorld that was sure about GP104 having 1.920 SPs? Better wait for reviews and solid info instead.

 

[Glo.]

The same VRWorld that was sure about GP104 having 1.920 SPs? Better wait for reviews and solid info instead.

If GP104 from GTX 1070 will have 1920 CUDA cores, will that count as they were correct?

 

[Face2Face]

Respectfully disagree. The 4870/4850 was a MASSIVE shift in perf/$ and cut the legs out from under the much more expensive Geforce 280 and 270. The GP104 chips don't really do much for perf/$, they do provide more performance but with an almost directly proportional price increase.

It just caused NVIDIA to release the 9800 GTX+ and release the GTX 260 Core 216. Both with were on par with AMD's GPUs.

 

[crisium]

It just caused NVIDIA to release the 9800 GTX+ and release the GTX 260 Core 216. Both with were on par with AMD's GPUs.

4870/4850 didn't "just" do that. It was a "massive" improvement in price-to-performance, as he said. Within days Nvidia cut the GTX 280 from $650 to $500 and the GTX 260 from $400 to $300, in addition to being revised to be faster. You can see how this release doesn't quite have the same flavor.

 

[96Firebird]

It doesn't really matter how this launch goes, the negative nancies in this thread will spin it whichever way suits them best.

If GP104 cards have plenty of stock due to good yields, they will claim no one wants them and they are just sitting on the shelves...

If GP104 cards sell out due to high demand, they will claim yields aren't good and it is a paper launch...

All you can do is smile and nod.

 

[airfathaaaaa]

The same VRWorld that was sure about GP104 having 1.920 SPs? Better wait for reviews and solid info instead.

if nvidia had any sort of async compute capabilities they would have already spammed the world with it..
instead we got only ONE slide today from them saying pretty much nothing

"first ever" like no one since 2011 had done this already..

you see something similiar here? :sneaky:
and the GLORIOUS official async slide

its pretty clear they are going to brute force it they didnt had time to rework the entire maxwell v3.0 to introduce any sort of async capabilities

 

[Face2Face]

4870/4850 didn't "just" do that. It was a "massive" improvement in price-to-performance, as he said. Within days Nvidia cut the GTX 280 from $650 to $500 and the GTX 260 from $400 to $300, in addition to being revised to be faster. You can see how this release doesn't quite have the same flavor.

Correct. I meant on top of the aggressive price drops. NVIDIA has to drop their prices and release new SKUs to compete against AMD.

 

[Sweepr]

if nvidia had any sort of async compute capabilities they would have already spammed the world with it..
instead we got only ONE slide today from them saying pretty much nothing

Funny, because your conclusion is...

its pretty clear they are going to brute force it they didnt had time to rework the entire maxwell v3.0 to introduce any sort of async capabilities

I guess 'nothing' is more than enough for your negative NVIDIA predictions.


Threadcrapping and trolling are not allowed
Markfw900

 

[airfathaaaaa]

Funny, because your conclusion is...



I guess 'nothing' is more than enough for your negative NVIDIA predictions.

no i actually love being fed lies and lies


Threadcrapping and trolling are not allowed
Markfw900

 

[renderstate]

if nvidia had any sort of async compute capabilities they would have already spammed the world with it..

instead we got only ONE slide today from them saying pretty much nothing

"first ever" like no one since 2011 had done this already..

You are confused. Async compute and preemption are *completely* different things.

Also NVIDIA is right, they have the first gfx pre-emption that stops processing at pixel level. All other HW out there, AFAIK, works at coarser granularity.

 

[Phynaz]

no i actually love being fed lies and lies

This line of yours is getting old.

 

[airfathaaaaa]

You are confused. Async compute and preemption are *completely* different things.

Also NVIDIA is right, they have the first gfx pre-emption that stops processing at pixel level. All other HW out there, AFAIK, works at coarser granularity.

Im well aware of what i said...thats why i put the third slide

also afaik as per zlatan
http://forums.anandtech.com/showpost.php?p=37520764&postcount=9

(also just a heads up that whole "pixel level pre emption" is a qualcomm idea)

 

[maddie]

4870/4850 didn't "just" do that. It was a "massive" improvement in price-to-performance, as he said. Within days Nvidia cut the GTX 280 from $650 to $500 and the GTX 260 from $400 to $300, in addition to being revised to be faster. You can see how this release doesn't quite have the same flavor.

Correct. I meant on top of the aggressive price drops. NVIDIA has to drop their prices and release new SKUs to compete against AMD.

Not just cut prices, they actually sent refunds to early buyers.

 

[Keysplayr]

Im well aware of what i said...thats why i put the third slide

also afaik as per zlatan
http://forums.anandtech.com/showpost.php?p=37520764&postcount=9

(also just a heads up that whole "pixel level pre emption" is a qualcomm idea)

So nothing good then can be garnered from this topics information then?

Is it really all badness and lies and deception as you claim?