New Titan X announced, 12GB GDDR5X, $1200

Status
Not open for further replies.
Jen-Hsun has been quoted as being a huge fan of Apple. The writing has been there since the GT200 days
Going after quality is now a bad thing? Also comparing a full PC/Phone?tablet? consumer goods manufacture to a gpu manufacture is a considerable stretch just because both sell halo products. I guarantee if AMD was capable of engineering such a GPU they would do so.
 
Can someone please explain why this isn't the "full" chip? People seem to be upset that nVidia is leaving something on the table. Is there a chart or something that shows what the untapped potential is?

The true full fat GP100 chip is capable of packing 60 SM units for a total of 3840 cores:

Like previous Tesla GPUs, GP100 is composed of an array of Graphics Processing Clusters (GPCs), Streaming Multiprocessors (SMs), and memory controllers. GP100 achieves its colossal throughput by providing six GPCs, up to 60 SMs, and eight 512-bit memory controllers (4096 bits total). The Pascal architecture’s computational prowess is more than just brute force: it increases performance not only by adding more SMs than previous GPUs, but by making each SM more efficient. Each SM has 64 CUDA cores and four texture units, for a total of 3840 CUDA cores and 240 texture units.

Inside Pascal: NVIDIA’s Newest Computing Platform

Probably because of yields such a chip doesn't exist yet (much like the original Titan and Tesla K20X which also had one SMX disabled), but instead what's available at the moment is a slightly cut down version with 4 disabled SMs for a total of 3584 cores.
 
Starting to think Titan just stands for the size of the price tag.

A lot of these end up being bought by the HPC market as they have similar FP16 (half precision) and FP32 (single precision) marks compared to the quadro counterparts at like 50% or more discount....
 
Coming back to this thread talking about GDDR vs HBM2.

I tested Rise of the Tomb Raider on full GP104, no power limit, 1983 core clock with no throttling:

VRAM speed
-10%: 62.92 FPS
stock: 73.99 FPS
+9.5%: 77.15 FPS

3440x1440, all settings maxed but HBAO+ rather than VXAO, DX11. Pretty impressed really. VRAM sat at 6-7.8GB usage.

So it does look like we're getting into diminishing returns area (on this one game). I'll still personally wait for HBM2. Maybe there's 5-10% to gain? Not like I am exactly hurting for performance right now. Plus I kinda want to see what AMD comes out with.

TLDR: GDDR is fine on the Pascal Titan X IMO, especially since as leldra said it has more bandwidth/core performance than the 1080.
 
Last edited:
Coming back to this thread talking about GDDR vs HBM2.

I tested Rise of the Tomb Raider on full GP104, no power limit, 1983 core clock with no throttling:

VRAM speed
8Ghz: 62.92 FPS
10 Ghz: 73.99 FPS (stock settings)
11.9 Ghz: 77.15 FPS (only +4% performance for +19% clocks)

3440x1440, all settings maxed but HBAO+ rather than VXAO, DX11. Pretty impressed really.

So it does look like we're getting into diminishing returns area (on this one game). I'll still personally wait for HBM2. Maybe there's 5-10% to gain? Not like I am exactly hurting for performance right now. Plus I kinda want to see what AMD comes out with.

TLDR: GDDR is fine on the Pascal Titan X IMO, especially since as leldra said it has more bandwidth/core performance than the 1080.

Nice results, I haven't don't thus un a while, I'll test in tomb raider as well, what are your settings.
Ah nvm you're using uw :(
 
Can we expect this beast to be bottlenecked by the CPU at high refresh resolutions lower than 4k? Specifically 1440p at 144+Hz.
 
Can we expect this beast to be bottlenecked by the CPU at high refresh resolutions lower than 4k? Specifically 1440p at 144+Hz.

in some games that don't support proper threading yes. It's always been this way though. Btw faster system memory shows pretty good gains for these crowds. Here are some ram benches in games that are known to be cpu limited.
Fallout.png

ARMA3.png


I'm actually excited for this Titan X. I may buy one. Nvidia has been on a roll lately getting all these cards out. 1060/1070/1080/ and now titan X. All my sources told me that the 1060 wasn't going to be out until end of August so looks like Nvidia pushed the card out a little faster than they had originally planned. Meanwhile AMD has only released the RX-480 officially, haven't even released the 470 and 460's yet. I mean your competitors will have basically their full lineup out, and AMD is stuck with one card. Sad really. Hopefully with this announcement of the Titan X, AMD will give details of Vega.
 
in some games that don't support proper threading yes. It's always been this way though. Btw faster system memory shows pretty good gains for these crowds. Here are some ram benches in games that are known to be cpu limited.
Fallout.png

ARMA3.png


I'm actually excited for this Titan X. I may buy one. Nvidia has been on a roll lately getting all these cards out. 1060/1070/1080/ and now titan X. All my sources told me that the 1060 wasn't going to be out until end of August so looks like Nvidia pushed the card out a little faster than they had originally planned. Meanwhile AMD has only released the RX-480 officially, haven't even released the 470 and 460's yet. I mean your competitors will have basically their full lineup out, and AMD is stuck with one card. Sad really. Hopefully with this announcement of the Titan X, AMD will give details of Vega.

And here's the issue yet again. Those complaining of nVidia's pricing sure they have a point. But AMD has launch on low end gaming GPU that's certainly not a bad part but nothing really like some of the hype was portraying. Now nVidia has a FULL line up starting from the low end and well past anything AMD has right now and will for months, maybe even the rest of the year. If this is how AMD is going to do business then yes nVidia is going to milk it for what it's worth because there are no other options at the higher end right now.

So if one is supposed to vote with their wallet, what exactly would those who are complaining of nVidia's pricing supposed to do? I'd like the prices to be lower. But I also wanted a very fast GPU for gaming. What are the options for people like me? Just wait on AMD to deliver something?
 
And here's the issue yet again. Those complaining of nVidia's pricing sure they have a point. But AMD has launch on low end gaming GPU that's certainly not a bad part but nothing really like some of the hype was portraying. Now nVidia has a FULL line up starting from the low end and well past anything AMD has right now and will for months, maybe even the rest of the year. If this is how AMD is going to do business then yes nVidia is going to milk it for what it's worth because there are no other options at the higher end right now.

So if one is supposed to vote with their wallet, what exactly would those who are complaining of nVidia's pricing supposed to do? I'd like the prices to be lower. But I also wanted a very fast GPU for gaming. What are the options for people like me? Just wait on AMD to deliver something?

Not much a consumer can do in this situation. AMD needs a gofundme at this point. Pretty sure some of us who have disposable income who buys Nvidia would throw some money AMDs way to help them push their technology forward to get back in the game.
 
People crack me up sometimes on here. You buy the fastest card and pay the huge premium for it, then dump it as soon as a card comes out thats a little faster, yet when you bought the card you defended it by saying I will have this card for a long time and it's worth that insane amount of cash. This is why Nvidia sells these cards, for the few of you willing to just throw cash at them, card has a cost around 250 bucks maybe even 300 and they are selling it to you for 1200. Only to complain about how even at 4k it still bogs down with all the eye candy on. I remember when I bought my 290x for 1080 gaming, people said it was overkill and they were right. However as new games came out and demanded more power I needed all the power of that 290x to keep good frame-rates. Still using that 290x and will see what Vega offers, otherwise I will wait this generation out as cards have become way overpriced for what they are.
 
Not much a consumer can do in this situation. AMD needs a gofundme at this point. Pretty sure some of us who have disposable income who buys Nvidia would throw some money AMDs way to help them push their technology forward to get back in the game.

Never thought I would see the day that AMD literally becomes a charity case.
 
People crack me up sometimes on here. You buy the fastest card and pay the huge premium for it, then dump it as soon as a card comes out thats a little faster, yet when you bought the card you defended it by saying I will have this card for a long time and it's worth that insane amount of cash. This is why Nvidia sells these cards, for the few of you willing to just throw cash at them, card has a cost around 250 bucks maybe even 300 and they are selling it to you for 1200. Only to complain about how even at 4k it still bogs down with all the eye candy on. I remember when I bought my 290x for 1080 gaming, people said it was overkill and they were right. However as new games came out and demanded more power I needed all the power of that 290x to keep good frame-rates. Still using that 290x and will see what Vega offers, otherwise I will wait this generation out as cards have become way overpriced for what they are.

If it can't do 4K with all the settings then people will buy another and OC them. Boom that's the [H] way.
 
  • Like
Reactions: RMXO
like this
Prices don't exist in a vacuum. If nobody buys the merchandise at a particular price, then the price will have to come down eventually. Titan Z is a perfect example of that.

The Titan Z didn't sell because the 295X2 offered superior performance for half the price...;)
 
The Titan Z didn't sell because the 295X2 offered superior performance for half the price...;)

And how well did the 295X2 sell at its initial launch price of $1500? IIRC the 295X2's price dropped by almost 50% within 6 months of release.

We could go even further back and look at the 8800 Ultra. There was nothing from ATi that remotely resembled competition for the 8800 GTX, let alone the 8800 Ultra. But the 8800 Ultra didn't sell like hotcakes either despite it having the undisputed performance crown.

Not much a consumer can do in this situation. AMD needs a gofundme at this point. Pretty sure some of us who have disposable income who buys Nvidia would throw some money AMDs way to help them push their technology forward to get back in the game.

I think we should start a kickstarter for AMD lol.

Lisa Su's base salary in 2014 when she was appointed CEO was $850K and went up to $950K this year.

Unless the kickstarter manages to raise several millions, all you're doing is funding Lisa Su's bank account LOL
 
Last edited:
I think we should start a kickstarter for AMD lol.


You have to collect 2-3B$ and wait 3-4 years for the result ;)

Lisa Su's base salary in 2014 when she was appointed CEO was $850K and went up to $950K this year.

Unless the kickstarter manages to raise several millions, all you're doing is funding Lisa Su's bank account LOL

You forgot bonuses. Lisa Su is one of the best paid CEOs, a traditional thing for AMD.

Base wage means absolutely nothing for these people. I would be surprised if she ranks in less than 10M$/year. She got 11.75M$ in 2014 as compensation alone for example.
 
Last edited:
Again Nvidia being Nvidia. Next VGA will be 1080ti which is already READY.
 
You forgot bonuses. Lisa Su is one of the best paid CEOs, a traditional thing for AMD.

Base wage means absolutely nothing for these people. I would be surprised if she ranks in less than 10M$/year. She got 11.75M$ in 2014 as compensation alone for example.

Yes I know she gets crazy bonuses and stock options (as does our own CEO), but I simply wanted to highlight the futility of kickstarter (yes I know it's a joke) if we just looked at base salary alone and nothing else.
 
I was planning on doing 1070 SLI, but factoring in waterblocks the price difference between SLI and a single Titan is ~$150. Now I've had good experiences with SLI, but that was back in the 4xx/2xx eras. Everything I hear these days is negative, so is it worth the $150 to go for a beefy single card? I must say I don't like how a single card looks in an ATX case though.
 
I was planning on doing 1070 SLI, but factoring in waterblocks the price difference between SLI and a single Titan is ~$150. Now I've had good experiences with SLI, but that was back in the 4xx/2xx eras. Everything I hear these days is negative, so is it worth the $150 to go for a beefy single card? I must say I don't like how a single card looks in an ATX case though.

I actually mounted my 120mm rad below my card in a mATX. Fit perfect.

I would avoid SLI if at all possible. The only way I recommend it is if you already have the top card and need more horsepower.

In reality PL mod + AIO or water block and the new Titan should be pretty awesome. And you get that awesome all the time vs some of the time.
 
Apple? Quality? Don't confuse high price and 50-60% margins with quality.

20+ years ago you might have been right.
They still make pretty high-quality products with some of the highest customer satisfaction in the industry. Just because you don't like them or their products or some other weird anti-Apple bias does not change that.
 
The true full fat GP100 chip is capable of packing 60 SM units for a total of 3840 cores: Probably because of yields such a chip doesn't exist yet (much like the original Titan and Tesla K20X which also had one SMX disabled), but instead what's available at the moment is a slightly cut down version with 4 disabled SMs for a total of 3584 cores.

I don't think it is not yields but business calculation. There should be coming yet an 'ultimate' Titan XXL with HBM2 memory and chip having slightlly more SMs than the Titan X which will help in product differentiation and justify terrifying price.
 
Pascal Titan X black should be a beast of a card.

Theres only 2SMs cut from Pitan. A full die will be 10% faster at most, probably with double the vram.

Titan is not intended to be a gaming card, it targets low budget hpc market, and doubles as a 'halo' gaming card
 
I know that but it's still way too gimped for the big data crowd. If it were say a third as powerful in compute as say gp100 that'd make a lot more sense.
 
I don't think it is not yields but business calculation. There should be coming yet an 'ultimate' Titan XXL with HBM2 memory and chip having slightlly more SMs than the Titan X which will help in product differentiation and justify terrifying price.

Maybe, maybe not. I'm more inclined to say it's yield because even GP100 used in Tesla P100 only has 56 functional SMs with 4 disabled.

Theres only 2SMs cut from Pitan. A full die will be 10% faster at most, probably with double the vram.

Titan is not intended to be a gaming card, it targets low budget hpc market, and doubles as a 'halo' gaming card

It's actually 4 SMs since each Pascal SM is half the size of a Maxwell SMM. Also if the new Titan X doesn't come with improved DP capability, then I doubt it'll gain any traction in the HPC market, and will end up being another pure gaming card just like the old Titan X.
 
Maybe, maybe not. I'm more inclined to say it's yield because even GP100 used in Tesla P100 only has 56 functional SMs with 4 disabled.



It's actually 4 SMs since each Pascal SM is half the size of a Maxwell SMM. Also if the new Titan X doesn't come with improved DP capability, then I doubt it'll gain any traction in the HPC market, and will end up being another pure gaming card just like the old Titan X.

No, it's 2 SMs. The gaming oriented cards have 128 shaders per SM, GP100 has 64 shaders per SM.
 
Yes both GP104 and GP106 are 128 shaders per SM while GP00 is 64 shaders per SM.

I guess we'll just have to wait and see whether GP102 is closer to GP100 or GP104 in core arrangement.
 
Yes both GP104 and GP106 are 128 shaders per SM while GP00 is 64 shaders per SM.

I guess we'll just have to wait and see whether GP102 is closer to GP100 or GP104 in core arrangement.

GP102 is likely based on the same arrangement as GP104/106.
 
I actually mounted my 120mm rad below my card in a mATX. Fit perfect.

I would avoid SLI if at all possible. The only way I recommend it is if you already have the top card and need more horsepower.

In reality PL mod + AIO or water block and the new Titan should be pretty awesome. And you get that awesome all the time vs some of the time.
Although any game that doesn't have an SLI profile probably wouldn't need more than a 1070 anyways...

I guess I'll wait and see how the Titan performs at 4k. I think it's about time I ditch triple-wide for gaming.
 
Is this actually known or just speculation?

It is known, because GP100 doesn't have the capability to do 4xINT8, GP100 uses SM6.0, GP102/104/106 use SM6.1.

SM6.0 can do 2xFP16, SM6.1 4xINT8.
 
I honestly have no idea what you just wrote lol. I would love to read up on some documentation/whitepaper if you could link me to any thanks.
 
Status
Not open for further replies.
Back
Top