** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[Gregster]

Agreed DM, bad nVidia for whatever they do!

 

[drunkenmaster]

using existing parts as a dev kit for something not yet available is pretty standard also. Sometimes you show a working pre production model, sometimes you show a dev kit set to emulate the final product.

Can you not read or something? Is it not a fact that Nvidia have repeatedly lied about their products on stage, yes. So it could be that, it could also be a dev kit emulating the final product which you may or may not notice but I said is entirely standard. The difference comes in what you claim you're holding when you show it around. If you're honest, it's fine because that is how the industry works, you can't prepare for new products if people can't have any idea what it will do.

Most console dev kits don't have final hardware, they are usually PC's with appropriate toolkits and hardware to emulate the rough performance you will expect on the console.

I'm saying everything Nvidia are doing is bad by saying it's standard and fine, and mentioning the other possibility make the post Nvidia bashing?

 

[bru]

I would suggest you stop focusing on the 8TF and look at the other things this can do compared to a TitanX.

Screen grab of the presentation.

 

[Gregster]

@DM It is the way you go about posting. AMD are fantastic and Nvidia are bad and that is what we always get. Please feel free to carry on though but I and many others enjoy a healthy debate about Pascal/Greenland and sadly, no new info at CES this time, so I suspect GTC in April will be when we hear this and that about it.

Just chill fella, it is all good and just hardware for our enjoyment

 

[Kaapstad]

Can you not read or something? Is it not a fact that Nvidia have repeatedly lied about their products on stage, yes. So it could be that, it could also be a dev kit emulating the final product which you may or may not notice but I said is entirely standard. The difference comes in what you claim you're holding when you show it around. If you're honest, it's fine because that is how the industry works, you can't prepare for new products if people can't have any idea what it will do.

Most console dev kits don't have final hardware, they are usually PC's with appropriate toolkits and hardware to emulate the rough performance you will expect on the console.

I'm saying everything Nvidia are doing is bad by saying it's standard and fine, and mentioning the other possibility make the post Nvidia bashing?

Could be worse

NVidia could have said their cards were "an overclockers dream".

If any manufacturer had made that claim it would be silly now wouldn't it.

 

[drunkenmaster]

I would suggest you stop focusing on the 8TF and look at the other things this can do compared to a TitanX.

Screen grab of the presentation.

I don't know if you noticed, but that image doesn't show anything it can do that TitanX can't, however the discussion was if they were real Pascal gpus or not, nothing more less and 8TF is still a low number for 250W if that can be achieved on 28nm.

@DM It is the way you go about posting. AMD are fantastic and Nvidia are bad and that is what we always get. Please feel free to carry on though but I and many others enjoy a healthy debate about Pascal/Greenland and sadly, no new info at CES this time, so I suspect GTC in April will be when we hear this and that about it.

Just chill fella, it is all good and just hardware for our enjoyment

Just chill, really, who was it that got all upset at nothing. The discussion moved to if they were real Pascal GPUs, I neither brought this up, others mentioned the fake 480gtx and woodscrews before me. I both included that and brought up the perfectly valid reasons to use Maxwell which you took exception to. I was in a way defending Nvidia and you as per usual decided I was making a 'bad' post because that is yours and Kaaps default stance.

But that is also a fairly default reply, jump to a conclusion without reading properly, when it's pointed out how wrong you are instantly jump to "just chill" or "don't get so upset", when people are wrong on the internet it's hilarious how often people try to play it down as the other person getting upset to deflect.

 

[D.P.]

I would suggest you stop focusing on the 8TF and look at the other things this can do compared to a TitanX.

Screen grab of the presentation.

the real take home message, ignoring all synthetic numbers but concentrating on real-world applications, is the new Drive PX system is over 6x faster than a TitanX based on a standard image classification tasks with deep learning, and uses much less power. When it comes to gaming many other factors will be involved (e.g. if the gpu is pixel limited then its compute capabilities might be constrained) from that Pascal is very much inline with Nvidia's prior announcements.

The amount of theoretical terraflops vs actual real-world performance can be vastly different.

 

[drunkenmaster]

Could be worse

NVidia could have said their cards were "an overclockers dream".

If any manufacturer had made that claim it would be silly now wouldn't it.

Didn't read the post also, that post and the previous one is basically stating for everyone else that dev kits often/usually use previous hardware to emulate not yet available hardware. So you're piling on about me explaining why it could be Maxwell gpus.

Can I ask what reason you felt the need to 'pile on' when that was the basis of both posts?

 

[FoxEye]

dm you seem mighty ****ed off today, for whatever reason.

 

[drunkenmaster]

the real take home message, ignoring all synthetic numbers but concentrating on real-world applications, is the new Drive PX system is over 6x faster than a TitanX based on a standard image classification tasks with deep learning, and uses much less power. When it comes to gaming many other factors will be involved (e.g. if the gpu is pixel limited then its compute capabilities might be constrained) from that Pascal is very much inline with Nvidia's prior announcements.

The amount of theoretical terraflops vs actual real-world performance can be vastly different.

It indicates the likely hood of one very specific upgrade to one very specific type of algorithm. IE look at a CPU that got a say 2000% increase in performance for say a specific cryptography function(as both Intel and AMD have had in recent years) while the actual general performance increase in basically every other activity is 5-10% as usual.

It's also quite possible the reason for two smaller gpus providing more bandwidth and more front end hardware scheduling hardware. It's been rumoured Nvidia would need to move hardware scheduling back on die rather than offloaded to the CPU as they did for Maxwell, which would likely help out with compute significantly though also likely reduce power efficiency(by itself, there will be other architectural power efficiency savings and finfet alone will improve it drastically).

 

[Kaapstad]

Didn't read the post also, that post and the previous one is basically stating for everyone else that dev kits often/usually use previous hardware to emulate not yet available hardware. So you're piling on about me explaining why it could be Maxwell gpus.

Can I ask what reason you felt the need to 'pile on' when that was the basis of both posts?

The point I was making is both AMD and NVidia have been know to stretch the truth a bit.

Having said that a Fury X does overclock better than a GPU using woodscrews (but only just).

 

[Drollic]

Said it a number of times, but I'm really hoping the 970 replacement will be enough for 1440p along with my 3570k. I don't want to shell out for the 980 replacement in this rig as I'd imagine I'd be CPU bottlenecked. If it's not enough for 60fps+ then it's looking like Broadwell-e and 980 equivalent.

 

[Westie]

A 980 equivilent wouldnt bottleneck a 3570 if its oced past 4.4+

 

[Mtom]

I don't know if you noticed, but that image doesn't show anything it can do that TitanX can't, however the discussion was if they were real Pascal gpus or not, nothing more less and 8TF is still a low number for 250W if that can be achieved on 28nm.



Just chill, really, who was it that got all upset at nothing. The discussion moved to if they were real Pascal GPUs, I neither brought this up, others mentioned the fake 480gtx and woodscrews before me. I both included that and brought up the perfectly valid reasons to use Maxwell which you took exception to. I was in a way defending Nvidia and you as per usual decided I was making a 'bad' post because that is yours and Kaaps default stance.

But that is also a fairly default reply, jump to a conclusion without reading properly, when it's pointed out how wrong you are instantly jump to "just chill" or "don't get so upset", when people are wrong on the internet it's hilarious how often people try to play it down as the other person getting upset to deflect.

Yep this was the main question.

 

[Orangey]

Good news for GPU makers:

http://www.kitguru.net/apple/matthew-wilson/apple-has-apparently-cut-iphone-production-by-30/

 

[Boomstick777]

Good news for GPU makers:

http://www.kitguru.net/apple/matthew-wilson/apple-has-apparently-cut-iphone-production-by-30/

Good news for me as well. I hate Apple with a passion

 

[D.P.]

Had a good look at the drive PX2 card, no wooden screws visible. Chips don't say much, will upload a photo later. NVIDIA are all over CES automotive, lots of partnrrs, e.g. HERE maps.

 

[drunkenmaster]

Lots of partners, but sounds like Audi spent a year with Nvidia and apparently are dumping Nvidia across the board for Qualcomm from 2017(not sure if they make the switch some point this year). With Tegra based products the question will be how long they stick around. They had loads of partners in phones and tablets till people actually used Tegra in 1-2 generations of products before they all stopped wanting to work with Nvidia any more.

 

[D.P.]

Under NDA so can't say much but there are some big players who are getting plenty of press that are working with nvidia. Spoke to one engineer from an unamed company working of ADAS and he was extremely excited about drive PX2 and Nvidia deep learning software.

Auto OEMS like to have multiple providers, they rarely go exclusive. For map data one model year might have one provder, the next model year has a different, the next year yet another, heck different models within the sqme yesr couks use duffetebt suppliets. When looking at CPU solutions the OEMS are mostly looking at costs, if Qualcomm is $5 cheaper and meets minimum spec then they will get the deal. Althoigh it shpuld ne noted NV ate doing much more thsn hardware, their software solutions is part of their value add which others ignore.

I can say is last yearn nVida were no where to be seen at CES automotive, now they have a giant booth slap bang next to Ford. Their booth was packed with senior Toyota honda, chevy GM, autoliv, samsung, etc. execs. They are providing several OEMS with hardware I can't discuss.

 

[P.B]

Good news for me as well. I hate Apple with a passion

I echo that, can't stand them as a company and would rather see gpu tech pushed than phones where there really doesn't need any more advancement in this area. Need to focus on batteries more.