• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Nvidia Volta is 16nm, expected in May 2017

Exentryk said:
WCCFTech posted a rumor that 1080 Ti is coming in Jan 2017, and gave the following specs:
Click to expand...

If these specs are true, Titan X owners got screwed. Since 1080Ti's will come with superior aftermarket coolers, I can easily see them offering better performance through overclocking to make up for the marginally less CUDA cores.
 
Evo X said:
If these specs are true, Titan X owners got screwed. Since 1080Ti's will come with superior aftermarket coolers, I can easily see them offering better performance through overclocking to make up for the marginally less CUDA cores.
Click to expand...

Yeah, but Titan owners always get screwed eventually. It happened with the original and the X. That's the price to pay (literally and figuratively) for getting that level of power at the first opportunity.

I'm quite comfortable with my 980ti and Gsync, so I'll be waiting it out and seeing what hits next year.
 
Evo X said:
If these specs are true, Titan X owners got screwed. Since 1080Ti's will come with superior aftermarket coolers, I can easily see them offering better performance through overclocking to make up for the marginally less CUDA cores.
Click to expand...

This happens every time, surely Titan owners expect it by now?

I feel like they got less screwed this time though if the 1080 Ti isn't coming out until January. That'll be 5 months since the Titan release, a much bigger gap than last time. 980 Ti came out less than three months after the TX.
 
Aranath said:
Yeah, but Titan owners always get screwed eventually. It happened with the original and the X. That's the price to pay (literally and figuratively) for getting that level of power at the first opportunity.

I'm quite comfortable with my 980ti and Gsync, so I'll be waiting it out and seeing what hits next year.
Click to expand...


I am doing the same, going to save and move to the next nvidia card and also possibly upgrade from 6700k to a 6800k or a 6900k.
 
Evo X said:
If these specs are true, Titan X owners got screwed. Since 1080Ti's will come with superior aftermarket coolers, I can easily see them offering better performance through overclocking to make up for the marginally less CUDA cores.
Click to expand...

Part of being a Titan owner (two times here...) is knowing you'll be screwed over one way or another.
 
Evo X said:
If these specs are true, Titan X owners got screwed. Since 1080Ti's will come with superior aftermarket coolers, I can easily see them offering better performance through overclocking to make up for the marginally less CUDA cores.
Click to expand...

So just like with Maxwell Titan X and the 980 Ti then? The cycle repeats...

At this stage, Titan X buyers should know that they're paying a premium for the fastest card on the market, which will more than likely be equalled by a lower-priced card within 3-6 months of buying it.
 
Compsiox said:
Do they want to lose the trust of consumers? Like damn. Don't get me wrong I'd be glad but I'd feel bad for someone who picked up a 1080
Click to expand...

I picked up 1080 a couple of months ago and I'm completely fine with 1080Ti releasing tomorrow. I don't understand why they would loose my trust because of this.
 
People forget that Titan's are used by Professional's aswell as gamers. The titan line allows you to get Quadro like features for a fraction of the cost. Namely high amounts of VRAM (titan's usually have more vram - not the case with )and TCC/Quadro drivers. I also believe that the Titans receive much better support than geforce for professionals (dont quote me on that as we dont use GPU rendering here yet).

Compared to Quadro cards, the Titan is a steal, compared to Geforce Cards then Titan doesnt make sense.
 
Will probably wait till the new line to upgrade from my 970.

I did consider getting a 1070 before it's launch but they priced me out. If I'm going to pay Nvidia's higher "were popular and in demand" prices, then I at least want a significant upgrade. 970 is mostly doing me fine, outside of outliers like Forza Horizon 3 which runs rather poorly all things considered,
 
Evo X said:
If these specs are true, Titan X owners got screwed. Since 1080Ti's will come with superior aftermarket coolers, I can easily see them offering better performance through overclocking to make up for the marginally less CUDA cores.
Click to expand...

Lol no, I refused to believe anyone is so stupid as to feel "screwed" because they bought a super high-end product and didnt expect something else to come along afterwards.
 
Is there any leaps forecasted (or 'just' happened) in the GPU industry? The only two I remember are HBM memory and stacked RAM, both of which haven't taken off yet. Are things looking like the same-ish gradual improvements or is there something 'big' going on?

The CPU development 'stagnated' in my view- relatively speaking since multi-cres appeared then stalled again. Nvidia seems to put major emphasis on Tegra. I think they did great with the latest GTX cards (10xx), but nothing major industry tech-wise.
 
Zafir said:
Will probably wait till the new line to upgrade from my 970.

I did consider getting a 1070 before it's launch but they priced me out. If I'm going to pay Nvidia's higher "were popular and in demand" prices, then I at least want a significant upgrade. 970 is mostly doing me fine, outside of outliers like Forza Horizon 3 which runs rather poorly all things considered,
Click to expand...

I completely agree with you. I thought about jumping in for a 1070 considering how FH3 is running but fuck spending that kind of money because devs can't optimize and release their games in an acceptable state. Considering what the XB1 is doing with FH3 the 970 should handle it no problem. We all know how much of a problem it actually is though.
 
Going to ride my 980Ti through at least 2017.

At that point I'll either jump on a dropped in price 2017 GPU or start scoping what 2018 has to offer. I also plan on seeing what exactly Scorpio has to offer and its price point.
 
Heaven Shard said:
Is there any leaps forecasted (or 'just' happened) in the GPU industry? The only two I remember are HBM memory and stacked RAM, both of which haven't taken off yet. Are things looking like the same-ish gradual improvements or is there something 'big' going on?

The CPU development 'stagnated' in my view- relatively speaking since multi-cres appeared then stalled again. Nvidia seems to put major emphasis on Tegra. I think they did great with the latest GTX cards (10xx), but nothing major industry tech-wise.
Click to expand...

HBM is stacked RAM, and we've just had a 28->16nm production switch. I don't know how much of a "leap" HBM is though as providing loads of memory bandwidth on its own will not lead to any "leaps" as can be seen from AMD's Fiji performance.
 
shandy706 said:
Going to ride my 980Ti through at least 2017.

At that point I'll either jump on a dropped in price 2017 GPU or start scoping what 2018 has to offer. I also plan on seeing what exactly Scorpio has to offer and its price point.
Click to expand...

Same here, unless the 1080Ti comes out this holiday at a reasonable price
 
Renekton said:
The smallish gap between Titan and 1080 makes slotting in 1080Ti tricky. Either 1080 buyer or Titan buyer may feel gypped.
Click to expand...

Hasn't stopped Nvidia in the past has it? Especially if enthusiasts keep buying.

its why I'm just sticking with mid tier cards since I don't really need high end.

AmyS said:
That makes no sense. GTX 1080 already has GDDR5X, why would 1080Ti go back to GDDR5 ?

Anyway, the rumors are saying 1080Ti will have GDDR5X.
Click to expand...

Makes no sense cause it's probably bullshit. It will have GDDR5X
 
Sword Of Doom said:
Makes no sense cause it's probably bullshit. It will have GDDR5X
Click to expand...

Exactly.

Anyway, at GTC 2017 (in April) I expect Nvidia to update their roadmap by showing the name and time frame of their next GPU architecture after Volta. This year was the first time Nvidia didn't update their roadmap, given the shifting of their roadmap in previous years with Volta, then Pascal, then Pascal and Volta.

Einstein
 
AmyS said:
Exactly.

Anyway, at GTC 2017 (in April) I expect Nvidia to update their roadmap by showing the name and time frame of their next GPU architecture after Volta. This year was the first time Nvidia didn't update their roadmap, given the shifting of their roadmap in previous years with Volta, then Pascal, then Pascal and Volta.

Einstein
Click to expand...

Tbh it's not a stretch to just assume that after 16nm Volta there will be a Volta evolution architecture on 10 or possibly 7nm. Doesn't really matter how they'll call it.
 
Another small bit of info: How Nvidia’s Own Saturn V DGX-1 Cluster Stacks Up

Here’s the fun bit to contemplate. Nvidia has demonstrated it can deliver 42.5 teraflops of peak performance and about 28.7 teraflops of sustained performance in a 4U server node. That is about 10.6 teraflops per 1U of rack form factor space. IBM, Nvidia, and Mellanox are working to get more than 40 teraflops of capacity into a 2U “Witherspoon” system next year with the “Summit” supercomputing they are building for the US Department of Energy facility at Oak Ridge National Laboratory. To our way of thinking, that means the future “Volta” GV100 accelerator cards should have about twice the performance of the Pascal GP100 cards currently shipping, if IBM sticks with the 2U form factor and only puts four GPU accelerators into the Witherspoon box alongside a pair of 24-core Power9 chips as we expect.
Click to expand...
 
hmmm grab a 1080ti or wait for Volta... ugh. I keep postponing my build but I think I just need to put my foot down. First I was waiting for 7700k's to launch (but may grab a 6700k when they drop to $299 in 2 weeks), and waiting for the new Asus z270 itx boards (becasue the 170's had a lot of missing features like a m.2 slot, but now u.2 to m.2 adapters are coming out in good quanities thats a non-issue) and now Volta may launch next year. Ugh. I think I'll be fine for 3 years with a 1080ti, 6700k and 32gb of ram.
 
I can't wait have the disposable income to chase the latest greatest GPU... but for now I'll rejoice that we'll (hopefully) get some more price cuts on the 10 series GPUs :D
 
AmyS said:
That makes no sense. GTX 1080 already has GDDR5X, why would 1080Ti go back to GDDR5 ?

Anyway, the rumors are saying 1080Ti will have GDDR5X.
Click to expand...

The current rumor is at 10GB. GDDR5X or non X doesn't seem to be confirmed yet. However, a 12GB 1080Ti with GDDR5 (NON X) memory would still have more memory bandwidth than the 8GB 1080 with GDDR5X. I haven't calculated 10GB yet...too lazy.
 
Belmire said:
The current rumor is at 10GB. GDDR5X or non X doesn't seem to be confirmed yet. However, a 12GB 1080Ti with GDDR5 (NON X) memory would still have more memory bandwidth than the 8GB 1080 with GDDR5X. I haven't calculated 10GB yet...too lazy.
Click to expand...

Okay gotcha'

Also, some interesting guesses about Volta.

https://www.nextplatform.com/2016/11/20/details-emerge-summit-power-tesla-ai-supercomputer/

The Summit configuration also tells us, perhaps, something about the Volta GPUs, or at least the ones being used inside of Summit.

Way back when, in early 2015, Nvidia said that it would be able to deliver Pascal GPUs with 32 GB of HBM memory on the package that delivered 1 TB/sec of bandwidth into and out of that GPU memory. What really happened was that Nvidia was only able to get 16 GB of memory on the package and only delivered 720 GB/sec of bandwidth with that on-package HBM with the Tesla P100 card. No one is making promises about the amount of GPU memory or bandwidth coming with Volta, as you can see above. What Nvidia has said, way back in 2015, is that the Volta GP100 GPUs would deliver a relative single precision floating general matrix multiply (SGEMM) of 72 gigaflops per watt, compared to 40 SGEMM gigaflops per watt for the Pascal GP100.
Click to expand...

If you use that ratio, and then cut it in half for double precision, then a Volta GPU held at a constant 300 watts (the same as the Pascal package) would have a little over 9.5 teraflops of double precision performance, and four of them would deliver 38.2 teraflops of oomph, the vast majority of the more than 40 teraflops of performance expected in the Summit node. Six GPUs at this performance level would deliver a total of 57.2 teraflops just from the GPUs alone, which no one has promised and that is why we think Nvidia is gearing these Volta GPUs down to maybe 200 watts. If you cut the clocks and therefore the thermals down by 100 watts on each card, you can stay in the same 1,200 watt GPU envelope as four Pascal P100 cards but maybe only cut performance by 20 percent to 25 percent against that 33 percent wattage drop. By moving from four Voltas to six Voltas per node, the HBM memory per node could increase by a lot (100 percent in capacity per card and another 50 percent from having more cards) and the performance per watt and the aggregate performance could be pushed a little further, too. With a geared down Volta card running at 200 watts, you could have a V100 card that deliver 7.6 teraflops at double precision and 38.2 gigaflops per watt compared to something like 38.2 gigaflops per watt for the faster V100 card we theorized above.

For fun, let’s call it 50 teraflops per node in Summit. That is a total of 512 GB of main memory (with 120 GB/sec of bandwidth, according to specs provided by IBM earlier this year), and if Nvidia can reach its original goal of 32 GB of HBM memory per GPU accelerator it hoped to hit with Pascal on the Voltas, that works out to 192 GB of HBM memory with 6 TB/sec of bandwidth. That is a lot more than the 64 GB of HBM memory and aggregate 2.8 TB/sec of GPU memory bandwidth in the current “Minsky” Power Systems LC precursor to the Summit’s Witherspoon node. There is another 800 GB of non-volatile memory in the Summit node, and we are pretty sure it is not Intel’s 3D XPoint memory and we would guess it is flash capacity (probably NVM-Express drives) from Seagate Technologies but Oak Ridge has not said. The math works with this scenario, with 512 GB of DDR4 main memory, a total of 192 GB of HBM memory on the GPUs and 800 GB of flash, across 4,600 nodes that is a total of 6.9 TB of aggregate memory. (By the way, that chart has an error. The “Titan” supercomputer has 32 GB of DDR3 memory plus 6 GB of GDDR5 memory per node to reach a total of 693 TB of aggregate memory.)

At that 50 teraflops of performance per node, which we think is doable if the feeds and speeds for Volta work out, that is a 230 petaflops cluster peak, and if the performance of the Volta GPUs can be pushed to an aggregate of 54.5 teraflops, then we are talking about crossing through 250 petaflops – a quarter of the way to exascale. And this is also a massive machine that could, in theory, run 4,600 neural network training runs side-by-side for machine learning workloads (we are not saying it will), but at the half precision math used in machine learning, that is above an exaflops of aggregate compute capacity.

Boom.
Click to expand...

Help me with math, what would a single GV100 do in single precision?

Then we might know roughly how much a paired-back GV102 (Titan X Volta) would do, assuming Nvidia goes with a GV102 rather than GV100 for GTX Titan and GTX 1180 Ti ;)
 
AmyS said:
Help me with math, what would a single GV100 do in single precision?

Then we might know roughly how much a paired-back GV102 (Titan X Volta) would do, assuming Nvidia goes with a GV102 rather than GV100 for GTX Titan and GTX 1180 Ti ;)
Click to expand...

If GV100 is anything like GP100 then it would be really hard to make anything out of GV100 specs for the gaming GV10x lineup. There's also a distinct possibility that GV102 may in fact be as large as GV100 and contain more FP32 SPs while omitting FP64 and HPC features as a chip targeted primarily at gaming (and deep learning?) which makes the situation even more complex and leads to such GV102 being actually faster than GV100 in FP32(/FP16?) workloads.
 
MadJackChurchill said:
It took them a year and a half to renew the 980. There's no way this will be released in May 2017 unless it's for the supercomputing sector.
Click to expand...

Maxwell update was limited by the availability of 16nm production process. If Volta will use the same 16nm process there won't be any reason to wait for anything but the readiness of the architecture itself.
 
Depends a lot on what AMD puts out with Vega. Then Nvidia will probably put out a new lineup, otherwise they might wait if AMD does not compete with their 1070/1080 cards.

Meanwhile displays are not keeping up with GPUs. We still only have 4K @ 60 Hz displays. I was hoping for 120+ Hz models already at the end of this year but it looks like they will be available earliest next summer if even then.
 
dr_rus said:
Maxwell update was limited by the availability of 16nm production process. If Volta will use the same 16nm process there won't be any reason to wait for anything but the readiness of the architecture itself.
Click to expand...

They can't release it too fast after 1080ti since people would be really pissed of without at least 1 year of owning top end card.
 
laxu said:
Depends a lot on what AMD puts out with Vega. Then Nvidia will probably put out a new lineup, otherwise they might wait if AMD does not compete with their 1070/1080 cards.

Meanwhile displays are not keeping up with GPUs. We still only have 4K @ 60 Hz displays. I was hoping for 120+ Hz models already at the end of this year but it looks like they will be available earliest next summer if even then.
Click to expand...

Right now even a Titan X Pascal cannot do 4k/60fps in many games on it's own, you need a pair and even then not every game you can hit 4k/60fps and you want a 4k at 120mhz? What the hell you going to use to drive that??? I will be happy to have a single GPU that can do 4k/60fps at full settings. Give me a GPU than can run Witcher 3 & Division at 4k/60fps with max settings.
 
laxu said:
Probably not. You could grab a cheap 980 Ti and use that. When heavily overclocked it is almost the same as a stock 1080.
Click to expand...

You'll have to forgive me. I'm not too tech savvy / an enthusiast regarding understanding computer parts (at least, not yet), but I don't really understand how overclocking works other than it makes the thing run faster than regular settings.

Would getting a stock 1080 be better in the long run as it seems new 1080 asus is cheaper than a new 980 ti
 
DataGhost said:
Shit and I was planning on building my first gaming pc this cyber monday. Should I wait for the volta to release?
Click to expand...

You can say that every year, you can wait or you can go now, but the GPU industry releases stuff pretty fast, so you can always play the waiting game or make a decision and jump in.
 
Hawk269 said:
You can say that every year, you can wait or you can go now, but the GPU industry releases stuff pretty fast, so you can always play the waiting game or make a decision and jump in.
Click to expand...

Unfortunately, being the idiot I am, that's probably what's been holding me back every year. That I'm losing out. I get anxious about losing out to something that could make this years cards look like nothing. If not for a friend's convincing me to finally build a gaming rig, I probably would have waited 80 years as he likes to joke about. Volta sounds like it would be cheaper and of course more efficient, similar to how a Pascal refresh (whatever that means) could make it that the 11xx or 20xx cards could make the 10xx that I plan to buy look like I lost out / made a bad call. It's not the best when I also know that the 10xx is the first batch of cards using Pascal so they haven't even perfected their technique yet of architecturing the card for more efficiency.

EDIT: also considering amd's new thing might give nvidia some competition, that might get me some better "deals"
 
Evo X said:
If these specs are true, Titan X owners got screwed. Since 1080Ti's will come with superior aftermarket coolers, I can easily see them offering better performance through overclocking to make up for the marginally less CUDA cores.
Click to expand...

Hasn't that been always the case? My Gigabyte G1 980ti OC'd is slightly faster in games than a Titan
 
michaelius said:
They can't release it too fast after 1080ti since people would be really pissed of without at least 1 year of owning top end card.
Click to expand...

I still can't understand why anyone would be "pissed" by this as this is pretty much how it works and always have been working in the PC space. A top end card rarely remains top end for more than half a year, anyone buying one should be totally aware of this.

Then again a Volta substitution for 1080Ti (if there will even be such a card this time) probably won't come until 2018, even though GV104 cards will probably end up being faster than said "1080Ti" if only by some 20% again.
 
if you bought a 1080 it's not going to instantly turn into crap when a new card launches
people need to chill about these things
 
aDnc.png

https://twitter.com/AIDA64_Official/status/833757338196643840

Getting closer.
 
You must log in or register to reply here.

Similar threads

Nvidia RTX 6000 Series Expected to Launch in Early 2027
Hardware  2 3
107
11K
CuNi replied
Super Metroid is expected to get a remake sometime in 2026
Rumor 
47
630
schaft0620 replied
  • Locked
Nvidia is building its second biggest Campus, in Israel
2
83
1K
DonkeyPunchJr replied
Videocardz - Board Channels: AMD and NVIDIA GPU price hikes may reach AIB cards in Jan-Feb 2026
2
69
2K
Bojji replied
NVIDIA adds Auto Shader Compilation beta to NVIDIA App
Hardware Platform 
10
414
Bojji replied
Top Bottom