Titan V announced £2700 15TF 12nm Volta

Silent_Scone · 8 Dec 2017 at 08:54

The only Titan I've never purchased was the Xp, I even had Titan Blacks. I think we're moving into the principle of things, but it's a big die so was never going to look cheap. It is for all intents and purposes an affordable workstation card. If you're not looking at it that way, then you're absolutely insane to part with £2,700.

Kaapstad · 8 Dec 2017 at 08:56

Dicehunter said:
£2700 for a single GPU ? Sorry but whoever gets this for "gaming" needs to go see a psychologist.

It's not even the full fat GV100, The Titan Vx will be out in 6 months for the low low price of £3000 and some sucker will lap it up.

It has all 5120 SP cores enabled so it is fully enabled for gaming and not a cut down chip like the 2016 Pascal Titan.

ChrisD. · 8 Dec 2017 at 08:57

But does it run Crysis?

Gregster · 8 Dec 2017 at 08:59

Kaapstad said:
It uses less power than Vega.

If I use the Volta Titan for the next 5000 years I will save some money on electricity !!!

lol

Sargatanas2511 · 8 Dec 2017 at 09:00

Going by the clocks, unless it boosts way beyond them I don't see it being a huge improvement over the current top end Pascals as a gaming card.

Silent_Scone · 8 Dec 2017 at 09:02

Kaapstad said:
It has all 5120 SP cores enabled so it is fully enabled for gaming and not a cut down chip like the 2016 Pascal Titan.

It will be a fair cut above the Titan Xp that there is no doubt, but you are paying more than ever now for the privilege and wasting the immense compute power you're paying for.

Sargatanas2511 said:
Going by the clocks, unless it boosts way beyond them I don't see it being a huge improvement over the current top end Pascals as a gaming card.

I'd hold that thought lol

. Besides which, you can't compare clocks across GPU architecture

Dicehunter · 8 Dec 2017 at 09:02

Kaapstad said:
It has all 5120 SP cores enabled so it is fully enabled for gaming and not a cut down chip like the 2016 Pascal Titan.

The full GV100 has a 4096 bit bus compared to the 3072 on the Tian V, The full GV100 has 900GB/s bandwidth compared to the 600GB/s on the Titan V and the full GV100 has 16GB of HBM2 compared to the 12GB on the Titan V so Nvidia releasing a full fleshed out Titan V is pretty much a guarantee.

Sorry dude, Don't get me wrong I think your bench threads are awesome but if you buy this for gaming I think it's safe to say you need to start reevaluating some things in your life and the constant immense throwing away of money.

I could quite happily buy one of these, But I won't, As I don't do machine learning

Gregster · 8 Dec 2017 at 09:09

I wouldn't mind seeing performance but not sure I could justify the cost.

imginy · 8 Dec 2017 at 09:11

Dicehunter said:
The full GV100 has a 4096 bit bus compared to the 3072 on the Tian V, The full GV100 has 900GB/s bandwidth compared to the 600GB/s on the Titan V and the full GV100 has 16GB of HBM2 compared to the 12GB on the Titan V so Nvidia releasing a full fleshed out Titan V is pretty much a guarantee.

Sorry dude, Don't get me wrong I think your bench threads are awesome but if you buy this for gaming I think it's safe to say you need to start reevaluating some things in your life and the constant immense throwing away of money.

I could quite happily buy one of these, But I won't, As I don't do machine learning

He obviously going to buy one and he obviously has enough money to do it.

If I won the lottery or something and my hobby was benching pc's for sure I would do the same.

Kaapstad · 8 Dec 2017 at 09:14

Dicehunter said:
The full GV100 has a 4096 bit bus compared to the 3072 on the Tian V, The full GV100 has 900GB/s bandwidth compared to the 600GB/s on the Titan V and the full GV100 has 16GB of HBM2 compared to the 12GB on the Titan V so Nvidia releasing a full fleshed out Titan V is pretty much a guarantee.

Sorry dude, Don't get me wrong I think your bench threads are awesome but if you buy this for gaming I think it's safe to say you need to start reevaluating some things in your life and the constant immense throwing away of money.

I could quite happily buy one of these, But I won't, As I don't do machine learning

None of the above will make any difference to the gaming capabilities of the GPU and the core is fully enabled, what you are talking about is the memory in the same way as if a 290X has 4gb or 8gb. Reducing the memory from 16gb to 12gb could even make the card faster in the same way as the 780 Ti was slightly faster than the Titan Black due to having less memory to drive and also the DP cores disabled.

Quartz · 8 Dec 2017 at 09:15

I'm going to be very interested in the benchmarks.

Silent_Scone · 8 Dec 2017 at 09:18

Kaapstad said:
None of the above will make any difference to the gaming capabilities of the GPU and the core is fully enabled, what you are talking about is the memory in the same way as if a 290X has 4gb or 8gb. Reducing the memory from 16gb to 12gb could even make the card faster in the same way as the 780 Ti was slightly faster than the Titan Black due to having less memory to drive and also the DP cores disabled.

That was more to do with the memory IC being faster but the rest stands

triss · 8 Dec 2017 at 09:21

Dicehunter said:
The full GV100 has a 4096 bit bus compared to the 3072 on the Tian V, The full GV100 has 900GB/s bandwidth compared to the 600GB/s on the Titan V and the full GV100 has 16GB of HBM2 compared to the 12GB on the Titan V so Nvidia releasing a full fleshed out Titan V is pretty much a guarantee.

Sorry dude, Don't get me wrong I think your bench threads are awesome but if you buy this for gaming I think it's safe to say you need to start reevaluating some things in your life and the constant immense throwing away of money.

I could quite happily buy one of these, But I won't, As I don't do machine learning

Kaap seems to buy most high end cards and seems rather happy. I suspect he is happy to spend on his hobby as he please and doesnt need to start reevaluating his life spending. Some spend on cars some house's some GPU's if you can afford it and it makes you happy good for you..

PS Hope who ever buys one posts in the owners thread and benches some games

shamus21 · 8 Dec 2017 at 09:23

They always release a new GPU just before year end but this time it is not much use for the gamer even with big pockets. Looking forward to March/April time frame when we will see the consumers versions.

Silent_Scone · 8 Dec 2017 at 09:23

Ugh, it's still Digital River. Wish they'd stop cutting etailers out, too.

drunkenmaster · 8 Dec 2017 at 09:23

Kaapstad said:
Looks like DM got it wrong about NVidia using NV Link for a gaming card.

It's funny, you're the main reason I rarely bother visiting around here but low and behold I check something about a new card and find you talking rubbish yet again.

A I said NVlink isn't coming to non server motherboards. NVlink is just higher power PCI-E and it's the second iteration of something IBM already did, there isn't anything fancy about it. Remove all power limitations on PCI-E and increase lanes and boom, NVlink which is exactly what IBM did on an earlier version of PCI-e standard previously. This isn't a gaming card and NVlink level bandwidth would provide exactly zero benefit for gaming, literally 0%, it would be useful SOLELY for applications that run purely on the gpu and might talk to each other with high bandwidth which is not at all what gaming does.

One of the main reasons and the most important use of NVLink is using super high bandwidth system memory and allowing more bandwidth access from memory to gpu, which requires the CPU to have NVlink controller instead of PCI-E so NVlink's best and only real implementation that can have any benefit to gaming would be via a IBM cpu which could theoretically increase bandwidth between system memory and gpu to far beyond that of any current gaming PC. For Gaming the bottleneck is between the CPU and GPU and between system memory and GPU, not between GPUs, massively increasing bandwidth between GPUs when the limitation is elsewhere does nothing.

Silent_Scone · 8 Dec 2017 at 09:28

Yeah, very little information passes over the SLI interface. The HB bridge offers more than enough bandwidth for that.

psychas · 8 Dec 2017 at 09:35

hopefully it will be like titan z and nobody will buy it till it drops price , but doubtful

edit: nvm its not for gaming

Dicehunter · 8 Dec 2017 at 09:40

shamus21 said:
They always release a new GPU just before year end but this time it is not much use for the gamer even with big pockets. Looking forward to March/April time frame when we will see the consumers versions.

I'd actually like to see this used in a video from an AI centric company about machine learning rather than games, Now that would be interesting.

Kaapstad · 8 Dec 2017 at 09:49

drunkenmaster said:
It's funny, you're the main reason I rarely bother visiting around here but low and behold I check something about a new card and find you talking rubbish yet again.

A I said NVlink isn't coming to non server motherboards. NVlink is just higher power PCI-E and it's the second iteration of something IBM already did, there isn't anything fancy about it. Remove all power limitations on PCI-E and increase lanes and boom, NVlink which is exactly what IBM did on an earlier version of PCI-e standard previously. This isn't a gaming card and NVlink level bandwidth would provide exactly zero benefit for gaming, literally 0%, it would be useful SOLELY for applications that run purely on the gpu and might talk to each other with high bandwidth which is not at all what gaming does.

One of the main reasons and the most important use of NVLink is using super high bandwidth system memory and allowing more bandwidth access from memory to gpu, which requires the CPU to have NVlink controller instead of PCI-E so NVlink's best and only real implementation that can have any benefit to gaming would be via a IBM cpu which could theoretically increase bandwidth between system memory and gpu to far beyond that of any current gaming PC. For Gaming the bottleneck is between the CPU and GPU and between system memory and GPU, not between GPUs, massively increasing bandwidth between GPUs when the limitation is elsewhere does nothing.

As it turns out you were right.

https://www.techpowerup.com/239519/nvidia-titan-v-lacks-sli-or-nvlink-support

If I am the main reason you rarely bother visiting around here, this does concern me to the point where this will be the last time I will comment on any of your posts or mention you in any way. Free speech is important.