Ars Technica's First look at NVIDIA Pascal P100 (Tesla)

[AngryPandaPC]

Article Link: http://arstechnica.co.uk/gadgets/2016/04/nvidia-tesla-p100-pascal-details/

 

Article Highlights:

Quote

The first full-fat GPU based on Nvidia's all-new Pascal architecture is here. And while the Tesla P100 is aimed at professionals and deep learning systems rather than consumers

 Full fat baby!

 

Quote

if consumer Pascal GPUs are anything like it—and there's a very good chance they will be—gamers and enthusiasts alike are going to see a monumental boost in performance.

Argh! I want to see it coming out now. Damn tease.

 

Quote

HBM2 is abig boost in performance over Nvidia's current performance champs like the Maxwell-based Tesla M40 and the Titan X/Quadro M6000.

HBM 2 at least on professional cards!

 

Quote

Nvidia says the P100 reaches 21.2 teraflops of half-precision (FP16) floating point performance, 10.6 teraflops of single precision (FP32), and 5.3 teraflops (1/2 rate) of double precision. By comparison, the Titan X and Tesla M40 offer just 7 teraflops of single precision floating point performance.

7->10 teraflops of single precision improvement.

 

Quote

Memory bandwidth more than doubles over the Titan X to 720GB/s thanks to the wider 4096-bit memory bus, while capacity goes up to 16GB.

4096 Bit Memory Bus. Nice.

 

Quote

The Tesla P100 isn't even a fully-enabled version of Pascal; it's based on the company's new GP100 GPU, with 56 of its 60 streaming multiprocessors (SM) enabled. The GP100 die, with a surface area of 610 square millimetres, is roughly the same size as the GM200 Titan X. Rather than shrink down the die thanks to the smaller 16nm process, Nvidia has instead chosen to simply fill the same space up with a lot more transistors—15.3 billion of them to be precise—almost doubling that of the top-end GM200 Maxwell chip.

That's an interesting surprise.  There is still top end to unlock.  Hmm, is this the first whisps of a future Pascal Ti ?!

 

Quote

There's a core clock of 1328MHz and a boost clock of 1480MHz—both much higher than Maxwell-based GPUs

Looks like Pascal will have a high clock increase like Maxwell did over the previous gen.

 

Quote

Launching alongside the Tesla P100 is Nvidia's DGX-1 Deep Learning System. The DGX-1 features eight Tesla GP100 GPU cards providing 170 teraflops of half-precision performance from its 28,672 CUDA cores. The DGX-100 also features two 16-core Intel Xeon E5-2698 v3 2.3GHz CPUs, 512GB of DDR4 RAM, 4x 1.92TB SSD RAID, dual 10GbE, 4x InfiniBand EDR network ports, and requires a maximum of 3200W of power. It'll cost a mere $129,000 (~£103,000) when it launches in June. Those after a P100 in other types of server will have to wait a little longer, with cards expected to reach systems in 2017, likely due to binning and HBM2 manufacturing constraints.

Nvidia displaying their AI and VR first approach.  Sounds like a beast system for developers.

 

All in all very exciting! Can't wait to see new systems with new architecture! Lets just hope we get to see it all in 2016.

[DocSwag]

Isn't this a repost....

[zMeul]

26 minutes ago, AngryPandaPC said:

 Full fat baby!

actually it isn't

the GP100 is the Tesla P100 is a cut down version because of the manufacturing issues - someday we'll get to see the full GP100, but this isn't it

full GP100 will have 60 shader modules, while the Tesla's P100 GPU only has 56; also will have more CUDA cores

[Spectrez]

Amazing how much people are reading into this card... 

 

 

[NumLock21]

1 hour ago, DocSwag said:

Isn't this a repost....

Yes it is, a lot of us saw it yesterday, when Jen hsun himself revealed it during the live stream. 

[Prysin]

and this card will be an overpriced belly-flop that wont give us what we need.

 

The more i hear about pascal, the sadder i get. I was hoping to do Multi-Vendor GPU setup with the next gen cards. Looks like ill be sticking with only one or the other, probably the other.

[DocSwag]

21 minutes ago, NumLock21 said:

Yes it is, a lot of us saw it yesterday, when Jen hsun himself revealed it during the live stream. 

?

This right here is the same topic but posted a little before.

[VagabondWraith]

Pascal = what Maxwell was supposedly supposed to be. I don't think we'll see the 'big jump' until Volta. Nonetheless, Pascal will still see nice gains.

[Unhelpful]

22 minutes ago, Prysin said:

and this card will be an overpriced belly-flop that wont give us what we need.

 

The more i hear about pascal, the sadder i get. I was hoping to do Multi-Vendor GPU setup with the next gen cards. Looks like ill be sticking with only one or the other, probably the other.

What? Did you not read anything? I'm convinced you had that response in a notepad somewhere ready to paste it without reading.

 

"if consumer Pascal GPUs are anything like it—and there's a very good chance they will be—gamers and enthusiasts alike are going to see a monumental boost in performance."

 

"HBM2 is abig boost in performance over Nvidia's current performance champs like the Maxwell-based Tesla M40 and the Titan X/Quadro M6000." (HBM2 is at least on pro-grade cards)

 

"There's a core clock of 1328MHz and a boost clock of 1480MHz—both much higher than Maxwell-based GPUs—along with a 300W TDP"

[NumLock21]

21 minutes ago, DocSwag said:

?

This right here is the same topic but posted a little before.

 

[zMeul]

28 minutes ago, JoeyDM said:

What? Did you not read anything?

he's just pissed nVidia took a big dump on AMD with being 1st to publicly reveal their next gen GPUs and being the 1st to show HBM2 implemented

it's not a desktop implementation, but still ..

[Unhelpful]

6 minutes ago, zMeul said:

 

Shhh. You're a fanboy too. Fanboys on every side are obnoxious. He may be pissed about it, but that doesn't mean that you should bask in it.

[AlexGoesHigh]

1 hour ago, Spectrez said:

Amazing how much people are reading into this card... 

 

 

Because it has been a while since the release of the 900 series, and it also has a bunch of new tech that's coming to market like HBM2 and its finally a die shrink since 2012 28nm that we have been stuck with.

[zMeul]

2 minutes ago, JoeyDM said:

Shhh. You're a fanboy too.

I am!? really?! 

ever since last year I said that nVidia will 1st reveal Tesla accelerators before the desktop variants - there's huge amounts of money to be made in the HPC market

Quote

doesn't mean that you should bask in it

it absolutely means that, AMD talked big and AMD's minions need a lesson

[Unhelpful]

1 minute ago, zMeul said:

 

You are! Congratulations! Your below post confirms it!  or if you aren't an nvidia fanboy, then you're an AMD hater. Which is just as annoying as AMD fanboys!

 

Objectivity, it's something that people here need to work on.

[zMeul]

Just now, JoeyDM said:

You are! Congratulations! Your below post confirms it!  or if you aren't an nvidia fanboy, then you're an AMD hater. Which is just as annoying as AMD fanboys!

I'm not an nVidia fanboy, I just hate AMD with every fiber in my body - I explained the why a long time ago

[Unhelpful]

Just now, zMeul said:

I'm not an nVidia fanboy, I just hate AMD with every fiber in my body - I explained the why a long time ago

 

2 minutes ago, JoeyDM said:

then you're an AMD hater. Which is just as annoying as AMD fanboys!

So I'm correct?

 

If you have a clear and obvious irrational bias, then you really have no way you can contribute. 

[zMeul]

4 minutes ago, JoeyDM said:

obvious irrational bias

no, mate .. nothing irrational here

I'm paying back AMD, in my own way, for the shit they've done to me - I'm pointing out every shit and lie they've said and done

as for AMD's fanboys .. they can pop a vein, don't care - I have them all on my ignore list

[Unhelpful]

Just now, zMeul said:

 

So you got burned as a consumer and are now whiny? Good to know.

 

Asus has burned me several times. So has intel. That isn't enough to make me a crybaby.

[zMeul]

2 minutes ago, JoeyDM said:

-

so you're just as ignorant as the rest of them, OK .. good to know

you can sleep tight, won't bother you ever again 

[Unhelpful]

Just now, zMeul said:

 

How am I ignorant? I'm objective about products. By closing your eyes and whining because you got closed as a consumer, you are knowingly and willingly choosing to be ignorant.

[Tomsen]

27 minutes ago, zMeul said:

he's just pissed nVidia took a big dump on AMD with being 1st to publicly reveal their next gen GPUs and being the 1st to show HBM2 implemented

it's not a desktop implementation, but still ..

Wait wait wait.

AMD revealed their next gen GPU already back in December..

Did he show a chip with HBM2 memory implemented? If so I most have missed it.

 

It was a few fancy rendered images with some fancy promising words on the side.

[Kinda Bottlenecked]

7 minutes ago, zMeul said:

no, mate .. nothing irrational here

I'm paying back AMD, in my own way, for the shit they've done to me - I'm pointing out every shit and lie they've said and done

as for AMD's fanboys .. they can pop a vein, don't care - I have them all on my ignore list

 

You forgot the people who point out your bullshit as well  

 

On Topic: 

Anw. 300TDP. Not surprising since it has so many more cores and even higher clock speeds. Looks like Nvidia might choose to shed the perf/watt thing. Doesn't really matter since There will probably be a significant performance increase.

[Nisco3000]

People accusing other to be fan boys are the worst. They are just trowing more jet fuel on the fire!

So what you do then!? ...IGNORE there post...

It's simple as that.

[Notional]

1 hour ago, zMeul said:

I'm not an nVidia fanboy, I just hate AMD with every fiber in my body - I explained the why a long time ago

I don't believe you ever did. But I would love to see a link where you explain why.

1 hour ago, zMeul said:

no, mate .. nothing irrational here

I'm paying back AMD, in my own way, for the shit they've done to me - I'm pointing out every shit and lie they've said and done

as for AMD's fanboys .. they can pop a vein, don't care - I have them all on my ignore list

So you're on a vendetta for some unknown reason, so now you spend all your time spamming this forum with FUD about AMD? Yeah completely rational.

I know you won't see this as I'm on your ignore list, for pointing out all your BS. Ignore lists are pathetic and only used by people too weak minded to defend their points and unable to understand facts.

2 hours ago, zMeul said:

he's just pissed nVidia took a big dump on AMD with being 1st to publicly reveal their next gen GPUs and being the 1st to show HBM2 implemented

it's not a desktop implementation, but still ..

Polaris was revealed a long time ago, and to a much more detailed extent than Pascal. Both will release HBM2 cards about the same time, as the delay is due to lack of production rather than lack of cards.