Nvidia Pascal GPU to Feature 17B Transistors and 32GB HBM2 VRAM

http://wccftech.com/nvidia-pascal-gpu-17-billion-transistors-32-gb-hbm2-vram-arrives-in-2016/

by cma • 11 years ago

134 90 11 years ago

Zekio• 11 years ago

Looks like the next few generations of GPUs are gonna be fast!

I think this may be specs for their Tesla cards. I doubt they would throw 32 HBM2 even on a Titan unless they can keep the price point the same (even for folks who can spend 1k USD have limits on their budget or at least their perception of a budget). But I wouldn't doubt by Q2/2016 Nvidia will be bringing more competition to the market with chips based on HBM usage.

IMO, AMD pulled the trigger a tad too soon on their HBM cards, but to be fair their last line of R9 and R7 cards weren't interesting at least to me (and I have a Radeon HD 7870). So, they had to go first to get their customer base excited for the future.

noir_lord2• 11 years ago

Surprises me that RAM hasn't increased more, the machine I'm on is 5 years old and it has 2GB of RAM on the GPU (HD6950) which I paid ~200 quid for 5 years ago.

hajile3• 11 years ago

As GPU history has shown, having an extra generation of experience with new things matters a lot (doesn't matter if it's GDDR3/4/5, fabrication nodes, specialized hardware).

blinkingled2• 11 years ago

AMD get it together, pretty please. Nobody wants to live in a world where only NVidia produces discrete GPUs!

That 32GB seems too high for cost constrained consumer market though - may be they will have a leaner variant for desktops/gaming.

bryanlarsen3• 11 years ago

It could be argued that AMD does have it together. If you look at the benchmarks for pretty much any particular price point, AMD & nVidia cards are roughly comparable. AMD is better at some games and resolutions, and nVidia is better at others. Rarely is the difference more than 10%. nVidia used to have a sizable power/noise advantage, but Fury appears to close that gap. AMD even seems to be closing the driver stability gap.

But the Internet effect on the video card market is huge. There are hundreds of online benchmarks, huge fanboy communities, et cetera. The benchmarks magnify small differences causing a winner-take-all effect. The communities create a bandwagon effect. It's a wonder that AMD maintains the share that it does.

Yes, AMD is obviously #2, but it should be a close #2. In most markets close #2 is not a bad position to be in.

raverbashing3• 11 years ago

Not sure about their graphics division, but AMD has stopped innovating in the CPU department...

noir_lord2• 11 years ago

I've used http://www.videocardbenchmark.net/high_end_gpus.html and https://www.cpubenchmark.net/high_end_cpus.html for years.

It's some what asinine to reduce the complexity down to a single number but I've found that they are usually pretty reflective of what you find in practice.

x3sphere• 11 years ago

As someone who has bought AMD cards for the past 5 years for gaming, I switched over to Nvidia this time.

You've overlooked that Maxwell / 980 Ti has a ton of overclocking headroom. OC to OC, you're looking at 20-30% performance difference at more commonly used resolutions like 1440p and 1080p. The gap narrows only at 4K.

Voltage is locked on Fury right now and AMD isn't saying why. The best OC I've seen on one is 10% higher clocks, with Maxwell 30% is common. Huge difference considering both cards are in the same price bracket.

Lastly pricing is also an issue with the Fury cards. Usually AMD undercut Nvidia, but this time the flagship matches the 980 Ti in cost. As the underdog, AMD is going to have a tougher time winning people over at the same price point. This happens in any market.

I like to support the underdog, but I couldn't this time given the large disparity in OC performance.

TrevorJ• 11 years ago

Gotta say though, in terms of driver stability and user experience in terms of related software, etc, NVIDIA has the edge.

dman2• 11 years ago

AMD already has HBM chips out (Fury X and Fury) and already has a shipping chip with 32 GB of memory (S9170). If you are willing to work with OpenCL the double precision rating of AMD chips is very compelling (look at the W9100 that has been out for a while now has a peak rating of 2.62 TFLOPS double precision!). I am looking forward to HBM equipped firepro cards where the double precision support isnt gimped.

rbanffy• 11 years ago

32GB seems too high for cost constrained consumer market

This thing is definitely not for the consumer market, much less the cost constrained one. This is a small dedicated number crunching machine that, for reasons unfathomable to me, can spit out rendered 3d environments with admirable speed.

I liked to joke that no serious computer has keyboard/mouse/video ports because no serious computer would be used like that. That assumption held well until the late 80's. ;-)

But no. For gaming, this is the superlative of overkill.

rasz_pl• 11 years ago

Funny you say 'AMD get it together', because its AMD who designed HBM in the first place. Just like it was AMD who pioneered GDDR5 in graphic cards.

This article you just read is typical Nvidia "we are the best .. in 2 years, you just wait!!1". They produced same charts, slides and test results before releasing Tegra, Tegra2 and Tegra3. Every time it was supposed to revolutionize industry and beat competition, every time it released late and benched slower than products on the market.

blinkingled2• 11 years ago

I meant AMD get it together financially, of course AMD is to be thanked for lot of things including AMD64!

hyperpallium• 11 years ago

4096bit memory interface is surely supercomputer territory.

My ZX81 only had 8192 bits of memory (1KB).

onion2k• 11 years ago

And it could do 3D graphics. Sort of. 3D Monster Maze: https://www.youtube.com/watch?v=nKvd0zPfBE4

foobarge• 11 years ago

Nice - who needs 8K. 8. K. I haven't even updated to 4K. I don't think I have anything that runs 1080p.

rwmj• 11 years ago

What are the parts marked "R125" around the edge?

bearbin• 11 years ago

I believe they are inductors, part of a switch-mode power supply to reduce the voltage of incoming power to the level required by the cores (AFAIK ~1V).

zokier2• 11 years ago

The definitely are inductors (being labeled "Lxx" on PCB is a sure tell-tale sign), and the small 48ish pin packages near them are most likely SMPS controllers. Seems bit crazy that there are total of 18 of them, but I suppose that's what you gotta do when you are pushing possibly hundreds of amps of current.

jaytaylor• 11 years ago

Those look like voltage regulators [0].

[0] https://en.wikipedia.org/wiki/Voltage_regulator_module

robocat• 11 years ago

R125 is often used to refer to resistance of 0.125 Ohms - They may be high wattage resistors (I didn't load article because I am using roaming data while travelling!).

whoisthemachine2• 11 years ago

The 17 Billion transistors on the Pascal GPU are twice the transistors found on the GM200 Maxwell and the Fiji XT GPU core which is literally insane

Seems like Moore's law is alive and well in the graphics/attached processor space.

icegreentea3• 11 years ago

That's because GPUs were frozen at the 28nm node since like 2011. It'll be ~4 years of no die shrinks at the top end of GPUs when they finally transition to 16nm. If anything, Moore's law is behind schedule in the GPU space.

Note, that both NVidia and AMD rely on TSMC to manufacture their chips, so they're completely constrained by TSMC's ability to implement new process nodes.

varelse3• 11 years ago

2011: GTX 580, 1.5 TFLOPS

2012: GTX 680, 3.0 TFLOPS (~2.0 attainable)

2013: GTX Titan, 4.4 TFLOPS (~3.2 attainable)

2014: GTX 980, 4.6 TFLOPS

2015: GTX Titan X, 6.7 TFLOPS

Looks to me like they're doubling perf roughly every 2 years.

Meanwhle, my Core i7-5930k's SOL is <1/2 of 2011's GTX 580 at 672 GFLOPS and it still doesn't have fast approximate transcendentals. Skylake begins to fix this, but c'mon, GPUs have had these for almost a decade now...

whoisthemachine2• 11 years ago

My mistake, I don't follow it very closely anymore, so I made my statement with only 2 datapoints :). Have an upvote.

minthd• 11 years ago

Moore's law is also about cost per transistor.It has risen, so expect to see a very expensive card. Most of the cost/benefit ratio of this card would be probably due to the huge fast memeory, and not the 17 billion transistors.

raverbashing3• 11 years ago

I'm waiting for the comparisons between NVidia's offerings and Xeon Phi in real benchmarks

semi-extrinsic• 11 years ago

I believe the Xeon Phi is doing quite bad in that comparison, so you don't see much benchmark trumpeting from Intel. Here's one from Nvidia though (so add a pinch of salt or two), showing a 2x - 5x advantage of Tesla K80 over Knights Corner:

http://www.nvidia.com/object/justthefacts.html

mtarnovan• 11 years ago

"The 17 Billion transistors on the Pascal GPU are twice the transistors found on the GM200 Maxwell and the Fiji XT GPU core which is literally insane. "

https://xkcd.com/725/

chillingeffect2• 11 years ago

https://search.yahoo.com/yhs/search?p=literally+now+means+fi...

;)

gwern• 11 years ago

With 8Gb per DRAM die and 2 Gbps speed per pin, we get approximately 256 GB/s bandwidth per HBM2 stack. With four stacks in total, we will get 1 TB/s bandwidth on NVIDIA’s GP100 flagship Pascal which is twice compared to the 512 GB/s on AMD’s Fiji cards and three times that of the 980 Ti’s 334GB/s.
The Pascal GPU would also introduce NVLINK which is the next generation Unified Virtual Memory link with Gen 2.0 Cache coherency features and 5 – 12 times the bandwidth of a regular PCIe connection. This will solve many of the bandwidth issues that high performance GPUs currently face.

To point out the obvious, this sounds like it could be fantastic for deep learning. Not just is the RAM big enough to hold a lot of current datasets, it'll help alleviate the latency bottlenecks in updating stuff.

bd• 11 years ago

Yup, this is exactly the marketing angle Nvidia CEO was using in his GPU Technology Conference presentation (10x speed-up on deep-learning vs their current Maxwell architecture):

http://blogs.nvidia.com/blog/2015/03/17/pascal/

ak217• 11 years ago

This also seems to be the motivation behind the already available Titan X with its 12GB of RAM. Some of the gaming hardware reviewers are scratching their heads as to why anyone would need 12GB attached to one die. I chuckled when I saw those reviews that were totally oblivious to the deep learning applications.

itsnotlupus2• 11 years ago

Yes.. with the current deep dreaming fad, this could attract quite a few hobbyists eager to get near-real time video processing through large nn.

sidrajaram• 11 years ago

The current deep dreaming fad is going to last about one more week before everyone gets sick of looking at those stupid pictures. They really serve no purpose other than "look how crazy this stuff is" - deep learning is cray-cray.

m_mueller• 11 years ago

One note on Nvlink: To get that you will AFAIK have to buy a Power8 system. They would need a cooperation with Intel to get support for this on their architecture - and Intel so far likes to keep them on a slow bus.

curiousjorge• 11 years ago

I've been look at gtx 750 ti which has maxwell processor in it. I don't care too much about playing games on ultra but the sheer quiet, cool temperature on load, and low power consumption is what makes me attracted to it. I wonder if the new Pascal GPU will make room for more such low power consumption and quiet cards that pack plenty of punch?

samch• 11 years ago

Now, if only I could get a MackBook Pro with 32GB of system memory... Does anybody happen to know why that is such a difficult thing to achieve? It's amazing that, starting next year, there will be mainstream graphics cards with more memory than top-of-the-line laptops.

M4v3R• 11 years ago

Do you need 32GB of RAM in a laptop? 16GB ought to be enough for anybody ;).

Sarcasm aside, this is probably purely business driven decision. The tech is there and it's nearly zero difference for the manufacturer to put either 8GB or 16GB dies on the board. It's the same as with SSD's - companies have to milk existing capacity tier to offer users next ones, otherwise they will have lower profits.

kozukumi2• 11 years ago

I don't think you could get 16GB SO-DIMMs until earlier this year could you? I know you can get 32GB in a W550s which is probably the thinnest/lightest laptop capable of 32GB.

rincebrain• 11 years ago

Can you? I have a W550s and didn't see anyone hocking 16GB SODIMMs even earlier this year.

CyberDildonics2• 11 years ago

This is not a real site, they have a review from 4 months ago benchmarking a 990 Ti 24GB card with 'minesweeper' and 'solitaire'.

largote• 11 years ago

Was that review written on April 1st?

CyberDildonics2• 11 years ago

It must have been. They mixed in it with all their other headlines and the date was '4 months ago' so it wasn't labeled April 1st. All their other stories are shaky rumors so I thought the whole site was joke articles predicting extrapolating the next press release.

Lejendary• 11 years ago

Why so much VRAM, is that really necessary?

Nvidia Pascal GPU to Feature 17B Transistors and 32GB HBM2 VRAM

Related Stories