TechPowerUp's Titan X (Pascal) Review

[Sweepr]

31.5% faster than Geforce GTX 1080 at 4K
19.3% gain from overclocking with power limit and temperature maxed
Power consumption below GTX 980 Ti / Titan X (Maxwell) levels
Best performance per watt out of all Pascal cards so far (with 12GB GDDR5X)

www.techpowerup.com/reviews/NVIDIA/Titan_X_Pascal/

 

[antihelten]

19.3% gain from overclocking with power limit and temperature maxed

And fan at 100%, and this is a fan that is already noisy at stock settings.

 

[Arachnotronic]

A very nice card. Particularly impressed that perf/watt improved even with a bigger chip and wider memory interface, really speaks to the quality of the chip.

Kudos to NVIDIA, hope they can follow this up nicely with Volta. :thumbsup:

 

[Sweepr]

And fan at 100%, and this is a fan that is already noisy at stock settings.

Not that bad at stock, still below Titan X (Maxwell) / Fury / RX 480 according to TPU.

https://tpucdn.com/reviews/NVIDIA/Titan_X_Pascal/images/fannoise_load.png

A very nice card. Particularly impressed that perf/watt improved even with a bigger chip and wider memory interface, really speaks to the quality of the chip.

Kudos to NVIDIA, hope they can follow this up nicely with Volta. :thumbsup:

Mighty impressive considering it packs 12GB of GDDR5X. Basically double the perf/watt of the older Titan X (Maxwell).

 

[Headfoot]

Extremely impressive card. I'm really looking forward to the inevitable 1080 Ti as I can't justify buying any card for $1200. That, and I'd really like to buy a AIB version with a better fan and hopefully more free voltage control BIOS like you saw on some 980 Ti models

 

[antihelten]

Not that bad at stock, still below Titan X (Maxwell) / Fury / RX 480 according to TPU.

The fact that it's comparable to a bunch of other cards with poor noise levels isn't exactly a good thing. And again this is stock, the overclocking in question was done at 100% fan speed.

 

[MrTeal]

That's sad. At only 205W I'd really need four of them to give my loop a reasonable workout, and quad-sli isn't supported anymore. How's a guy supposed to keep his feet warm in the winter with these things?

The pricing on these is really tough to swallow, but I just can't see a 1080 Ti coming out any time soon given how constrained supply of the the 1080 still is and the lack of top end competition from AMD. No one looking at dropping $1200 on a GPU would want to see a $700 GTX in four months.

 

[Sweepr]

Looking at some of their game tests, GP102 delivers >45 FPS @ 4K in all games tested (most crossing 60 FPS).

 

[thilanliyan]

What a beast of a card!! Amd really needs to show up to the party, but I'm guessing the 14nm process isn't working out as well as the 16nm, plus design differences aren't helping the situation either in terms of performance and performance per watt.

 

[Newbian]

What a beast of a card!! Amd really needs to show up to the party, but I'm guessing the 14nm process isn't working out as well as the 16nm, plus design differences aren't helping the situation either in terms of performance and performance per watt.

One can only hope they do in 2017 when vega is released.

 

[JDG1980]

What a beast of a card!! Amd really needs to show up to the party, but I'm guessing the 14nm process isn't working out as well as the 16nm, plus design differences aren't helping the situation either in terms of performance and performance per watt.

I think the primary reason why AMD's high-end offerings aren't ready yet has little to do with process maturity and a lot to do with the fact that they bet on HBM2. In retrospect it seems clear that GDDR5X was the quickest path to market, but that was not obvious even 6 months ago.

 

[Arachnotronic]

I think the primary reason why AMD's high-end offerings aren't ready yet has little to do with process maturity and a lot to do with the fact that they bet on HBM2. In retrospect it seems clear that GDDR5X was the quickest path to market, but that was not obvious even 6 months ago.

Nope. GP100 is shipping to datacenters now using HBM2.

 

[Bacon1]

Nope. GP100 is shipping to datacenters now using HBM2.

Source?

 

[Keysplayr]

Source?

http://wccftech.com/nvidia-tesla-p100-gp100-june-2016/
https://www.semiwiki.com/forum/cont...erformance-lead-neural-network-computing.html

 

[jpiniero]

I think the primary reason why AMD's high-end offerings aren't ready yet has little to do with process maturity and a lot to do with the fact that they bet on HBM2. In retrospect it seems clear that GDDR5X was the quickest path to market, but that was not obvious even 6 months ago.

AMD made it pretty clear that Vega's release date had more to do with AMD simply not having the resources to finish it in a timely fashion. The availability of HBM2 in volume probably played into the decision to focus on Polaris.

 

[tviceman]

I think the primary reason why AMD's high-end offerings aren't ready yet has little to do with process maturity and a lot to do with the fact that they bet on HBM2. In retrospect it seems clear that GDDR5X was the quickest path to market, but that was not obvious even 6 months ago.

That's not it at all. Samsung has been mass producing HBM2 now for a few months while Skynix starts mass production in Q3. GP100 is in production now while Nvidia was on the market with G5X 4-6 weeks or so after micron announced mass production.

 

[Arachnotronic]

Source?

From NV's most recent investor call:

Yeah. Thanks a lot, Deepon. We're expecting a lot of Pascal. Pascal was just announced for 1080 and 1070, and both of those products are in full production. We're in production with Tesla P100, and so all of our Pascal products that we've announced are in full production, so we're expecting a lot.

Anyway, dang fine execution from NV -- GP100, GP102, GP104, and GP106 all out in a matter of a few months.

 

[dark zero]

NVIDIA game literally all their money in Pascal, but considering that is a Maxwell shrink with some changes, I am not impressed. NVIDIA could do it really fast.

But seems that in the lower tier can't catch up against the competition and to make it worse is not scaling perfectly.

Even more, they have no plans to release the GTX1050 or the GP 107 based cards. Seems that they are focusing on high tier markets before going on the lower tier ones.

 

[Keysplayr]

NVIDIA game literally all their money in Pascal, but considering that is a Maxwell shrink with some changes, I am not impressed. NVIDIA could do it really fast.

But seems that in the lower tier can't catch up against the competition and to make it worse is not scaling perfectly.

Even more, they have no plans to release the GTX1050 or the GP 107 based cards. Seems that they are focusing on high tier markets before going on the lower tier ones.

Wut?

 

[tviceman]

NVIDIA game literally all their money in Pascal, but considering that is a Maxwell shrink with some changes, I am not impressed. NVIDIA could do it really fast.

But seems that in the lower tier can't catch up against the competition and to make it worse is not scaling perfectly.

Even more, they have no plans to release the GTX1050 or the GP 107 based cards. Seems that they are focusing on high tier markets before going on the lower tier ones.

I'd say, unequivocally, Polaris is more of a die shrunk iteration of GCN than Pascal is of Maxwell. Perf/mm2, perf/w and hardware feature sets prove that. Anyways,
weren't people saying Nvidia wouldn't have any mid range to counter AMD in the months leading up to Polaris? Weren't people saying AMD would beat Nvidia to the new nodes? Weren't people saying AMD would have a manufacturing advantage and a die shrunk Fiji would have Fiji like performance and be the size Polaris? All of which were proven false.

The sky is always going to be falling on Nvidia for the hardcore AMD optimists. GP106 is better than Polaris by most measurements and GP107 is coming, it will be out months ahead of Vega. AMD can't execute and once again we will have another generation with Nvidia dominating sales 75-25% and AMD not making enough money to fully fund R&D.

 

[nerp]

NVIDIA game literally all their money in Pascal, but considering that is a Maxwell shrink with some changes, I am not impressed. NVIDIA could do it really fast.

But seems that in the lower tier can't catch up against the competition and to make it worse is not scaling perfectly.

Even more, they have no plans to release the GTX1050 or the GP 107 based cards. Seems that they are focusing on high tier markets before going on the lower tier ones.

Yes, a die shrink and some changes. Super easy. Just right click on the die in the "die builder" app, choose "resize" then click "send to fab" so lazy, right??

 

[guskline]

Yes, a die shrink and some changes. Super easy. Just right click on the die in the "die builder" app, choose "resize" then click "send to fab" so lazy, right??

Funny quote nerp. Appreciate the cynicism.

I recently read an post about understanding the big change coming from 28nm gpus to 16/14 (Nvidia/AMD).

The poster indicated that 28 nm has been around for awhile so both companies had plenty of time to refine it.

In this first generation of 16/14 you don't have the depth of refinement yet. You have higher prices to recoup the investment costs etc.

I have been very pleased with my GTX 1080 vs my GTX 980TI SC below. Both very good cards.

I'm anxious to see how Nvidia and AMD fill the $150 to $400 price range.

Finally, complaining about the price of a Titan X to me is silly. Just the name alone tells me it is going to be salty.

 

[raghu78]

I'd say, unequivocally, Polaris is more of a die shrunk iteration of GCN than Pascal is of Maxwell. Perf/mm2, perf/w and hardware feature sets prove that. Anyways, weren't people saying Nvidia wouldn't have any mid range to counter AMD in the months leading up to Polaris? Weren't people saying AMD would beat Nvidia to the new nodes? Weren't people saying AMD would have a manufacturing advantage and a die shrunk Fiji would have Fiji like performance and be the size Polaris? All of which were proven false.

The sky is always going to be falling on Nvidia for the hardcore AMD optimists. GP106 is better than Polaris by most measurements and GP107 is coming, it will be out months ahead of Vega. AMD can't execute and once again we will have another generation with Nvidia dominating sales 75-25% and AMD not making enough money to fully fund R&D.

Given how poorly AMD has executed with Polaris it would be easy to say that Vega and Zen will disappoint. But for anybody who wants competition in the CPU and GPU market we need more competitive products from AMD. Its too early to write off this generation. AMD seems to be limited by the state of GF 14LPP and there is scope for improvements even with Polaris once the process issues are sorted out. Now we have to wait and see if AMD can make Polaris competitive with a second revision. But I would not hold my breath on that happening. As for Vega its prudent to not expect too much as AMD has constantly overpromised and underdelivered. So we should assume that GP104 will win against small Vega and GP102 will win against big Vega. This way we will atleast avoid being disappointed. I hope Vega is built at TSMC 16FF+ so that they don't have to compete with process disadvantage but until the product launches we won't know.

 

[rancherlee]

Makes me wonder with the die sizes increasing so quickly if we are not in for a LONG stagnation period in graphics improvements if we are stuck on 16/14 as long as we were on 28nm. The 1070/1080/titan x are a HUGE jump in performance, did they use up most of the die shrink advantage already?

 

[Genx87]

Yes, a die shrink and some changes. Super easy. Just right click on the die in the "die builder" app, choose "resize" then click "send to fab" so lazy, right??

lol