Nvidia's New Titan Xp Graphics Card Unlocks The GP102's Full Potential

[redgarl]

AMD is having a 12 TF card also... however chance are that it will cost 600$ instead. Once again, Nvidia will need to cut their prices.

 

[renz496]

AMD is having a 12 TF card also... however chance are that it will cost 600$ instead. Once again, Nvidia will need to cut their prices.

only in raw number only....Fury X at 8.6Tflops was considered as monster at the time of it's released. but then 5.6tflops 980ti just able to keep up just fine

 

[bit_user]

So there would need to be a GTX card based on the GP104 processor then as that has 8 memory channels.

Well, why would consumers need more than 12 GB? For professional/enterprise users, Nvidia offers the P6000 with 24 GB. So, you can get more memory if you're willing to pay for it. If Titan Xp had 24 GB, it would only eat away at the market share of the P6000.

Though I thought it interesting that didn't seem to be the case when both the Maxwell Titan X and M6000 had 12 GB.

I don't follow your point. Please explain.

Last year Nvida doubled the memory of the Quadro M6000 to 24 GB while keeping it at the same 5,000$ price,

Why are you comparing this to $5000 professional cards? Very different markets. And you're calling the Titan Xp overpriced?

...just saying that it seems the pro sector has been getting a break with prices remaining the same for the last 5 years while us enthusiasts dealt with price increases for improved performance.

I think it's funny that you call this "a break". The pro cards are basically just massively-overpriced consumer cards with ECC memory and some OpenGL-optimized drivers. If they didn't do anything else to differentiate them, fewer people would buy them. If professional graphics applications switch to Vulkan, then you might really see prices of the pro cards start to plummet.

...and yes, compared to the 1080 Ti, the Titan Xp is overpriced given the slight degree of advantage it offers.

You think I disagree with you? Do you think anyone here is saying otherwise?

 

[bit_user]

I dont like this .. Nvidia are pissing off their Titan X early buyers like this .. not cool.

I'm struggling to see the logic in this point.

They already pissed off their early Pascal Titan X buyers with the $700 GTX 1080 Ti. Now, I think those people should've known it was coming, and have basically no right to be upset. It's not like that make their cards any slower.

The way to look at this is as a refresh to the Pascal Titan X, to give basically anyone a reason still to buy one. Otherwise, everyone would just buy a factory-overclocked 1080 Ti. Again, if you bought the original, this doesn't make your card any less fast. Sure, it might ding the resale value, but most of that damage was already done by the 1080 Ti, as could be expected.

 

[Th3pwn3r]

Man, I swear the must be some troll in Nvidia's marketing dept, gleefully watching all the reviewers and fanatics trip over themselves having to explain exactly which Pascal-based Titan they're talking about.

Or maybe they just know that practically no one buys these $1200 cards anyway, so it doesn't really matter what they call them. : P

Or maybe you just don't know what you're talking about. Titan Xs were very hard to get your hands on because they were either sold out or being sold for over $1500 USD.

 

[cryoburner]

Or maybe you just don't know what you're talking about. Titan Xs were very hard to get your hands on because they were either sold out or being sold for over $1500 USD.

That's because they only bother to make a small number of them, not because they're flying off the shelves.

 

[bloodroses]

Or maybe you just don't know what you're talking about. Titan Xs were very hard to get your hands on because they were either sold out or being sold for over $1500 USD.

That's because they only bother to make a small number of them, not because they're flying off the shelves.

At $1200+, I can understand why they don't make many of them. Even half that price is near ridiculous for a single component strictly used for gaming. There will always those that are willing to pay that much though, and why video cards have been becoming more and more insanely priced.

 

[bit_user]

Or maybe you just don't know what you're talking about. Titan Xs were very hard to get your hands on because they were either sold out or being sold for over $1500 USD.

That's because they only bother to make a small number of them, not because they're flying off the shelves.

With such a margin, why on earth wouldn't they be making as many as they could sell?

It's probably safe to assume any early supply shortages were due to poor silicon yield or fab bottlenecks.

 

[renz496]

Or maybe you just don't know what you're talking about. Titan Xs were very hard to get your hands on because they were either sold out or being sold for over $1500 USD.

That's because they only bother to make a small number of them, not because they're flying off the shelves.

not many gamer buy them but i think most of the titan are end up in the hands of prosumer that use the titan for compute purpose. even some company out there offering their professional solution using titan as a cheaper alternative together with more expensive tesla solution:

http://www.thinkmate.com/systems/servers/gpx/titan-x

 

[kyotokid]

So there would need to be a GTX card based on the GP104 processor then as that has 8 memory channels.

Well, why would consumers need more than 12 GB? For professional/enterprise users, Nvidia offers the P6000 with 24 GB. So, you can get more memory if you're willing to pay for it. If Titan Xp had 24 GB, it would only eat away at the market share of the P6000.

...many of us are rendering with Iray which in one case has been integrated in an enthusiast's level 3D programme (Daz3D's Pro Studio). Big scenes with lots of polys, textures, and effects eat up VRAM like it was sack of Lay's Potato Crisps. If a scene file exceeds a card's VRAM it dumps to much slower CPU mode and that 700$ - 1,200$ GPU card becomes a boat anchor. Most people in this sector do not have the funds to purchase a Quadro P5000 much less a Quadro P6000.

Though I thought it interesting that didn't seem to be the case when both the Maxwell Titan X and M6000 had 12 GB.

I don't follow your point. Please explain.

...they doubled the VRAM of the M6000 without upping the price one cent. However when the GTX 980 Ti came out with an extra 2 GB VRAM the price increased by almost 200$. Same in the case of the 1080 Ti which boosted VRAM by 3 GB over the standard 1080. Pro studios usually have a lot more in the way of financial resources than us enthusiasts do yet they have been allowed to continue to pay the same price in spite of all the performance increases over the last five years (and can write the cost off as a business expense which we cannot do, basically we are paying more while they don't have to).

 

[renz496]

So there would need to be a GTX card based on the GP104 processor then as that has 8 memory channels.

Well, why would consumers need more than 12 GB? For professional/enterprise users, Nvidia offers the P6000 with 24 GB. So, you can get more memory if you're willing to pay for it. If Titan Xp had 24 GB, it would only eat away at the market share of the P6000.

...many of us are rendering with Iray which in one case has been integrated in an enthusiast's level 3D programme (Daz3D's Pro Studio). Big scenes with lots of polys, textures, and effects eat up VRAM like it was sack of Lay's Potato Crisps. If a scene file exceeds a card's VRAM it dumps to much slower CPU mode and that 700$ - 1,200$ GPU card becomes a boat anchor. Most people in this sector do not have the funds to purchase a Quadro P5000 much less a Quadro P6000.

Though I thought it interesting that didn't seem to be the case when both the Maxwell Titan X and M6000 had 12 GB.

I don't follow your point. Please explain.

...they doubled the VRAM of the M6000 without upping the price one cent. However when the GTX 980 Ti came out with an extra 2 GB VRAM the price increased by almost 200$. Same in the case of the 1080 Ti which boosted VRAM by 3 GB over the standard 1080. Pro studios usually have a lot more in the way of financial resources than us enthusiasts do yet they have been allowed to continue to pay the same price in spite of all the performance increases over the last five years (and can write the cost off as a business expense which we cannot do, basically we are paying more while they don't have to).

but Quadro M6000 is already super expensive to begin with. so it might not make sense to increase the price more just because they double the amount of VRAM.. remember those Quadro can easily 5 times more expensive than titan (even more so compared to regular geforce). it seems you just don't like the fact you have to pay a bit more (for more VRAM ) even if you're going to get the cards that cost significantly more cheaper than true Quadro card. and to be honest pro or not nvidia will want you to buy their Quadro/Tesla card for your work not their gaming card. so i don't think you can really complain when nvidia charging more for more VRAM with their "cheaper" gaming card

 

[bit_user]

Though I thought it interesting that didn't seem to be the case when both the Maxwell Titan X and M6000 had 12 GB.

I don't follow your point. Please explain.

...they doubled the VRAM of the M6000 without upping the price one cent. However when the GTX 980 Ti came out with an extra 2 GB VRAM the price increased by almost 200$. Same in the case of the 1080 Ti which boosted VRAM by 3 GB over the standard 1080.

The thing you need to understand about these GPUs is that they're all designed way ahead of time. There's probably at least 8 months between when the design is sent for fabrication of the first silicon and the time that graphics cards can actually reach store/warehouse shelves. On top of that, design and verification of the GPUs is a multi-year effort. So, all of the GPUs in a generation were designed way ahead-of-time.

What I'm saying is it's not like Nvidia took the GTX 980 and just bolted on an extra 50% more memory controllers to give the GTX 980 Ti its 6 GB. What they did was design the GM204 GPU with a 256-bit memory datapath and concurrently design the GM200 to have a 384-bit datapath.

During the design phase, they also looked at various factors and decided how wide to make the address bus. So, it might be that due to GPU design or GDDR5 memory supply, they could never deliver a GM204 with 8 GB. Or, it might be that the underlying support is there and it's simply a matter of finding the right price point for it to be profitable.

To get a better understanding, study this and you'll see how they can take only 3 chip designs, in the Maxwell 2xx series, and turn that into 21 different products addressing numerous markets at various price points:

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_900_series
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_900M_.289xxM.29_series
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#Quadro_Mxxx_series
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#Quadro_.28Mx000M.29_series
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#Tesla