Titan V announced £2700 15TF 12nm Volta

[Telecaster]

Hardly worth the price over 1080Ti

£700 vs £2700 come ON Now!! Performance for the price needs to be much better am talking Double and then some.

"Double" at what though? There are probably things it can do 10/20/30x faster than a 1080ti, never mind double, but they won't be gaming. It has compute features that make it very attractive to people who need that sort of thing, you're comparing it to the wrong card as the 1080ti doesn't have any.

 

[Strife212]

Thanks. So in theory this is the fastest Volta card?

It isn't, in fact it's literally a reject Tesla V100.

 

[CAT-THE-FIFTH]

I think some of these benchmarks are fake.

I have been looking at the GPUZ image over at Videocardz and things like the memory bandwidth and pixel fill rate are way out.

https://videocardz.com/74382/overclocked-nvidia-titan-v-benchmarks-emerge

It would be more useful if they compared it with an overclocked Titan Xp.

Edit!!

Not sure why people expect big things with this card??

If Nvidia released a consumer version which was more focussed on FP32 performance than FP16 and FP64 with something like GDDR6,it probably would be a smaller chip and probably faster and cheaper.

Nvidia is releasing this because they can,and I expect for certain non-gaming usage this card might actually be considered "cheap".

 

[Vega]

I hope the performance is more than ~15% greater which some of these benchmarks are showing.

 

[Kaapstad]

I hope the performance is more than ~15% greater which some of these benchmarks are showing.

We really need figures from the more respected review sites as some of the benches I have seen so far are very odd.

Firestrike for example goes from the Volta card being faster than a Xp at Standard setting to being slower at the Ultra setting. This does not make sense as the HBM2 memory on the Volta card should walk away with it at the higher Ultra settings.

 

[JediFragger]

I'm not sure we will see Volta consumer gaming cards or an architecture based on Volta. As I said before it was likely we might see a Volta Titan around the end of the year but the hints I've had are that Volta isn't really intended for GeForce.

Have to agree, Ampere smells like a tweaked Pascal on 12nm. Which would make sense really, as they would have gotten a sense of Vega performance around 12 months ago and easily had time for a contingency like this

 

[Kaapstad]

Have to agree, Ampere smells like a tweaked Pascal on 12nm. Which would make sense really, as they would have gotten a sense of Vega performance around 12 months ago and easily had time for a contingency like this

If NVidia have no intention of having a gaming range of Volta cards then this is a good time for them to start selling the Volta Titan as it gives them the opportunity of shifting their surplus/substandard GV100 chips. They can also carry on selling Pascal gaming cards until they want to launch Ampere in maybe 6 or 9 months time, this would explain the appearance of the 1070 Ti.

 

[JediFragger]

this would explain the appearance of the 1070 Ti.

Agreed, was most odd considering the age of the current cycle. We may get something in June I think.

 

[CAT-THE-FIFTH]

Have to agree, Ampere smells like a tweaked Pascal on 12nm. Which would make sense really, as they would have gotten a sense of Vega performance around 12 months ago and easily had time for a contingency like this

Well,it might be actually quite solid even if it is tweaked Pascal. The GP102 is 471MM2,so if Nvidia were to increase the size of its successor to 600MM2,use GDDR6,etc it could actually end up a decent amount faster.

I wonder if that means the GP104 successor will end up being closer to GTX1080TI level peformance,if Nvidia can use GDDR6 on it??

 

[Panos]

Have to agree, Ampere smells like a tweaked Pascal on 12nm. Which would make sense really, as they would have gotten a sense of Vega performance around 12 months ago and easily had time for a contingency like this

Yeah. Seems is the understanding of some media outlets. Ampere is Pascal on 12nm. Which is a shrank Maxwell.
Does this reminds you anything from Intel side? Stagnation of new tech and small incremented increases, and ludicrous high prices on the good stuff, against lack of competition.

And if someone says anything about is AMD fault, I will paraphrase John 8:7, “He that is without sin among you, let him first cast a stone at her.” to "He who has not advised an Nvidia card over an AMD card when the latter was better and cheaper".

 

[Kaapstad]

From the benchmarks we have seen it does seem to answer another question we see on the forums.

Is HBM2 better for high end gaming GPUs?

Judging by the fact that the Volta Titan is only a fraction faster than a Titan Xp despite having a lot more transistors the answer is probably no.

 

[DarrenM343]

From the benchmarks we have seen it does seem to answer another question we see on the forums.

Is HBM2 better for high end gaming GPUs?

Judging by the fact that the Volta Titan is only a fraction faster than a Titan Xp despite having a lot more transistors the answer is probably no.

Not looked around for benchmarks but is that true? I would thought given the specs it would be better than a fraction better even if HBM2 gimped it a bit. Heck, AMD used first gen HBM on the fury, must have gimped it massively.?

 

[Kaapstad]

There is another clue than the Volta Titan is a one off and not the top of a new range of Volta gaming cards.

The lack of SLI support, I know it is not popular at the moment but I think any new gaming cards will still support 2 way mGPU.

I think NVidia have done the minimum amount of work needed to get the Volta Titan on the shelves so that they can use their surplus/substandard GV100 chips. Adding SLI would be a waste of money and resources for a one off card if there are not going to be a range of Volta gaming cards.

Ampere could be the next proper gaming cards and I am sure there will be at least 2 way SLI support.

 

[Kaapstad]

Not looked around for benchmarks but is that true? I would thought given the specs it would be better than a fraction better even if HBM2 gimped it a bit. Heck, AMD used first gen HBM on the fury, must have gimped it massively.?

The benches I have seen so far are a bit suspect and we need more reliable ones from the better tech sites.

Having said that the Volta Titan is about 10% or 15% faster than a Titan Xp, slower in one and a lot faster in Unigine Superposition.

The weird thing is the benches it does worse in are the ones that HBM2 should be helping with.

https://videocardz.com/74382/overclocked-nvidia-titan-v-benchmarks-emerge

 

[Spyhop]

I think NVidia have done the minimum amount of work needed to get the Volta Titan on the shelves so that they can use their surplus/substandard GV100 chips. Adding SLI would be a waste of money and resources for a one off card if there are not going to be a range of Volta gaming cards.

Ampere could be the next proper gaming cards

You're bang on IMO.

 

[KompuKare]

Well,it might be actually quite solid even if it is tweaked Pascal. The GP102 is 471MM2,so if Nvidia were to increase the size of its successor to 600MM2,use GDDR6,etc it could actually end up a decent amount faster.

Thing of their poor margins!

When GV100 was first announced, I though the only way they're going to increase performance for Volta (the same applies to Ampere) is by increasing the die size.
There are basically three ways to increase performance of GPUs:

1. Clock it faster
2. Enhance the architecture to get more done per clock tic.
   
3. Add more shaders

Well, for 1. they probably won't get that much more out of it since TSMC don't seem to boast of clock enhancements with their '12nm' node.
For 2., I'm sure they'll try some ideas but the big efficiency gain they managed from Kepler to Maxwell is not likely to be repeated anytime soon.
So that only leaves them with 3. adding more shaders. Unless they can make those smaller, that means larger dies.
GP104 and even GP102 have plenty of space to grow, but their margins will suffer. I'd imagine they've targetted a certain extra performance for any new cards but I suspect it will in the low range of what people would be expecting. Still, +20% to +30% should allow the 1080 replacement to reach 1080Ti speeds.

 

[JediFragger]

“He that is without sin among you, let him first cast a stone at her.” to "He who has not advised an Nvidia card over an AMD card when the latter was better and cheaper".

Pass me that ****ing stone

 

[Quartz]

I'm a little sceptical about Ampere. A brief search indicates that the only source is one report by Heise.de. Everything else refers back to that.

 

[CAT-THE-FIFTH]

Thing of their poor margins!

When GV100 was first announced, I though the only way they're going to increase performance for Volta (the same applies to Ampere) is by increasing the die size.
There are basically three ways to increase performance of GPUs:

1. Clock it faster
2. Enhance the architecture to get more done per clock tic.
   
3. Add more shaders

Well, for 1. they probably won't get that much more out of it since TSMC don't seem to boast of clock enhancements with their '12nm' node.
For 2., I'm sure they'll try some ideas but the big efficiency gain they managed from Kepler to Maxwell is not likely to be repeated anytime soon.
So that only leaves them with 3. adding more shaders. Unless they can make those smaller, that means larger dies.
GP104 and even GP102 have plenty of space to grow, but their margins will suffer. I'd imagine they've targetted a certain extra performance for any new cards but I suspect it will in the low range of what people would be expecting. Still, +20% to +30% should allow the 1080 replacement to reach 1080Ti speeds.

Thats what I think too - the 12NM node is meant to be lower leakage so they can plonk more transistors in and with GDDR6 being higher bandwidth,they should be able to stick with a smaller memory controller over the GTX1080. With the GP102 sucessor its most likely they will move closer to 600MM2 and have more shaders but lower clocked,but with more memory bandwidth and some uarch improvements.

 

[Rroff]

I'm a little sceptical about Ampere. A brief search indicates that the only source is one report by Heise.de. Everything else refers back to that.

Yeah there is no official source for it.

They have another architecture additional to Pascal and Volta that was only known as Einstein - not a proper name just how it was referred to which we might see brought in after Pascal rather than Volta on GeForce. Many years back there was some switching around of the architectures and some elements of Einstein ended up on what is now Volta but I believe it still exists.

Thing of their poor margins!

When GV100 was first announced, I though the only way they're going to increase performance for Volta (the same applies to Ampere) is by increasing the die size.
There are basically three ways to increase performance of GPUs:

1. Clock it faster
2. Enhance the architecture to get more done per clock tic.
   
3. Add more shaders

Well, for 1. they probably won't get that much more out of it since TSMC don't seem to boast of clock enhancements with their '12nm' node.
For 2., I'm sure they'll try some ideas but the big efficiency gain they managed from Kepler to Maxwell is not likely to be repeated anytime soon.
So that only leaves them with 3. adding more shaders. Unless they can make those smaller, that means larger dies.
GP104 and even GP102 have plenty of space to grow, but their margins will suffer. I'd imagine they've targetted a certain extra performance for any new cards but I suspect it will in the low range of what people would be expecting. Still, +20% to +30% should allow the 1080 replacement to reach 1080Ti speeds.

"12NM" was largely brought in because nVidia needed something smaller than 16nm FF+ to hit the targets for the commercial obligations (amongst other contracts they had to hit specific perf/watt targets for Summit) when TSMC 10nm didn't turn up the goods in time. 10 and even 7nm will be available for their next gen GPUs though the design cost difference versus 12/16FF might have some implications depending on how their margins shape up.