NVIDIA Presents the TITAN RTX 24GB Graphics Card at $2,499

[EarthDog]

So Titan V is a snowflake since it doesn't have the 1:32 ratio, yes?

If the chip would have been identical except it had less DP cores then yes, it would have considered that to be gimped.

I wish my knife was a sharp as yours. I can't split hairs like that my man!

 

[Ubersonic]

Display Port? I don’t understand this. Does it have some kind of handicap with outputs?

Double Precision.

Refers to the FP64 mathematical performance, it's historically what separates Nvidia's consumer cards (GTX/RTX/Titan) from the industrial ones (Quadro/Tesla/etc). Though they did improve performance on the Titan XP via a driver update in response to Vega 64 teabagging it for half the price due to their standard gimping of DP performance.

 

[jabbadap]

So Titan V is a snowflake since it doesn't have the 1:32 ratio, yes?

I wish my knife was a sharp as yours. I can't split hairs like that my man!

GV100 is HPC chip as is GP100, a full fp64 compute capability chip. Nvidia does not make Geforces from hpc chips anymore, so there's no software fp64 crippled cards anymore(like gtx780/780ti were). So yes Titan V has all fp64 compute capabilities intact(Titan is not geforce anymore). Only thing what is crippled in geforce Turings is to half rate fp16w/fp32 tensor accumulate.

Spoiler: List of Turing tensors operations by anandtech

 

[EarthDog]

industrial ones (Quadro/Tesla/etc).

except these rtx quadros it seems according to vya (not that I dont believe him...just saying)?

 

[jabbadap]

Well about how Nvidia are targeting their cards, Anandtech's Ryan Smith said it well on Beyond3d forums:

You get full FP32 accumulate on the tensor cores for a fraction of the price of a Quadro RTX 6000. That alone will sell it to droves of data scientists.

Tesla: Servers
Quadro: Content creation & CAD
Titan: Developers & data scientists; develop on Titan now, then sell these users some Teslas when they need to scale out
GeForce: Consumer gaming & everyone who can't afford a better card.

 

[EarthDog]

I think you said it best. It isnt a GeForce card... yet most people are up in arms over its $/perf for gaming (as that is all they do and know). Others have a right to be disappointed due to lack of fp64/dp performance for their work,which was tempered earlier in this thread, though still significant in their specific use case.

In the end, to me, the only worthwhile gripe is that. Otherwise, it's cheaper and expected to perform 15% better tha. Titan V. If your focus on this being a gaming card, I think it is off the mark. Nowhere in the PR does this say gaming..it is not a part of the GF line... it just happens to not follow the Quadro driver stack and can game (very well is the thought).

 

[Deleted member 177333]

I remember the Nvidia RTX conference being open in one tab, and the RTX 2080ti in my product cart in the other tab. I was so close to impulse buying the thing as it was being announced. Bad habit of buying a new high end gpu or cpu each year. RTX made me break the cycle.

Yep I'm with ya - I had been planning to upgrade my 1080 Ti cards to 2080 Tis but as soon as I saw the price when considering the pathetic performance improvement versus the cards being ~$450 overpriced per unit, I said nope - you can keep 'em NVidia. I'll wait and buy 'em used or even for the next cycle or two...NVidia's stock is still in the toilet too. Hope it stays there as long as they're trying to price gouge customers.

 

[Vya Domus]

Nvidia does not make Geforces from hpc chips anymore

Too soon to say that, it's just one generation. And it was also too soon to replace those giant V100s anyway.

 

[GoldenX]

People defending a Titan at $2500. Will they do the same when an x060 costs $800? Even Apple had to roll back prices, the only one enduring is Intel, maybe holding up/waiting for Zen 2.

 

[jabbadap]

Too soon to say that, it's just one generation. And it was also too soon to replace those giant V100s anyway.

While you are right: it's too soon to say that, I just don't see the point of fp64 on gaming gpu. Not to mention now when they are pushing not just RT cores but Tensor cores too to take precious die area on gaming gpus. BTW. first HPC only chip from nvidia was old kepler gk210, while it has same amount of shaders as gk110 it has some differencies too and never used on consumer products(not even titans, which were geforces then).

 

[EarthDog]

People defending a Titan at $2500.

Why wouldn't you outside of the FP64 performance? Again, its a cheaper card and faster than last generation with one (notable for some) caveat. Arent they going in the right direction? Cheaper and Faster than the last Titan?

 

[Vya Domus]

I just don't see the point of fp64 on gaming gpu.

That was never the case anyway, the important bit is that it was the same chip used, with the same limitations in terms of number of shaders, clocks, etc.

 

[jabbadap]

Why wouldn't you outside of the FP64 performance? Again, its a cheaper card and faster than last generation with one (notable for some) caveat. Arent they going in the right direction? Cheaper and Faster than the last Titan?

Heh well yeah it's better cooled OCed Quadro RTX 6000 for less. Of course it lacks ECC memory and quadro 24/7 driver support. But hey if you don't need either, now you don't have to pay for them.

 

[efikkan]

People defending a Titan at $2500. Will they do the same when an x060 costs $800? Even Apple had to roll back prices, the only one enduring is Intel, maybe holding up/waiting for Zen 2.

AMD also have their Vega Frontier Edition at double price.

And I'll say it again; It's not for you!
These cards are very useful for developers and researchers.
Thankfully, this time Nvidia have doubled the memory again like the first Titan, making it a very good offering.

 

[skates]

Well, if they are going to call it T-Rex, I hope someone will throw an asteroid at it.

You sir, get a gold star.

 

[EarthDog]

Heh well yeah it's better cooled OCed Quadro RTX 6000 for less. Of course it lacks ECC memory and quadro 24/7 driver support. But hey if you don't need either, now you don't have to pay for them.

I recall another forum member here, Mr. Genius perhaps, that mentioned something about GDDR5/5X/GDDR6 have ECC? Is it a different implementation or enabled there versus not here or on the GeForce cards?

 

[bibob94]

well at least its not priced around 3 grand, in the US right? right?

 

[Fluffmeister]

People defending a Titan at $2500. Will they do the same when an x060 costs $800? Even Apple had to roll back prices, the only one enduring is Intel, maybe holding up/waiting for Zen 2.

It's a valid question, but I'm sure Navi won't be that awful.

 

[GoldenX]

It's a valid question, but I'm sure Navi won't be that awful.

It can be.

 

[xorbe]

Only thing what is crippled in geforce Turings is to half rate fp16w/fp32 tensor accumulate.

Oh that's probably partly why nVidia has to do the DLSS thing in-house per game.

 

[Power Slave]

RIP OFF!

/thread

 

[Ubersonic]

People defending a Titan at $2500.

Why wouldn't you

Because you're sane? lol.

 

[jabbadap]

I recall another forum member here, Mr. Genius perhaps, that mentioned something about GDDR5/5X/GDDR6 have ECC? Is it a different implementation or enabled there versus not here or on the GeForce cards?

Sure, all those memory types have ECC readiness by their Jedec specifications. Though nvidia's software implementation does not necessary take full advantage of it. Not sure why but I remember ECC being enabled only when graphics card is turned on compute mode, so it's mainly feature for servers.

Oh that's probably partly why nVidia has to do the DLSS thing in-house per game.

Nvidia is making those DLSS ML profiles by running games on their Saturn V supercomputer, I don't think game developers have those kind of machines on their own.

 

[EarthDog]

Because you're sane? lol.

Maybe....

I'd call it proper perspective on the product.

 

[Aquinus]

Maybe....

I'd call it proper perspective on the product.

I'm not convinced that such a thing exists on a product costing $2,500 for the consumer where you're literally paying twice as much for probably 10% bump in performance over what I already consider an egregiously priced product (the 2080 Ti.) How sad is it that you probably could get two 2080 Tis with it's crippled SLI performance which will probably still be better than getting this new GPU.