NVIDIA Titan XP Reviews and Benchmarks

[Prysin]

10 hours ago, SteveGrabowski0 said:

LOL what? They caught up to last generation's Nvidia cards in performance per watt after a die shrink.

oh, hi steve. Once you are done stroking your GTX 970, can you come back to reality and face some performance facts?

 

May i remind you that DX11 is on the way out, and AMDs DX12 and Vulcan drivers present much lower driver overhead, so perf per watt improves even more.

[Prysin]

8 hours ago, patrickjp93 said:

Pascal Cards higher than the 1060 support up to 4-way SLI, but, Nvidia has admitted what many of us already knew: scaling beyond 2 cards sucks. So, it doubled the bandwidth of the bridge if you were willing to settle on 2-way.

 

You do if the clock speed suffers to keep those 6144 cores in a reasonable TDP.

 

No it hasn't. The Rx 480 uses way more power than the 1060 just to match its performance.

 

Then OEMs will go for the 1060 over the Rx 480. It's cool as a cucumber, and Nvidia offers good bulk deals.

 

It's not really a double node shrink. It's 20nm with FinFET transistors and a couple little trimmings. So is Samsung/GloFo 14nmFF.

 

Eh, I think the improvements will show up over time, even without Nvidia "gimping" Maxwell.

you forget that TSCM's 20nm was closer to what? 22-24nm in "intel measurements" ? That one flopped, so by all technicalities, it is a "double" jump.

[SteveGrabowski0]

2 hours ago, Prysin said:

oh, hi steve. Once you are done stroking your GTX 970, can you come back to reality and face some performance facts?

 

May i remind you that DX11 is on the way out, and AMDs DX12 and Vulcan drivers present much lower driver overhead, so perf per watt improves even more.

I'm not stroking the 970 so much as shitting on AMD for the 480. LOL them having the nerve to tout their power efficiency when they can't even power their card within spec. LOL putting a 6-pin on their reference just so they can look like the card draws less power. I hope Vulkan does become big though and that would be awesome to see AMD knock Nvidia down a peg or two with it, but it's way too early to claim victory just based off Doom.

[Exotath]

Well damn... now that actual reviews are popping up this thing is nuts on the performance side.

 

One of these more than demolishes my current 980 SLI setup which a 1080 did a good job of matching.

[Prysin]

3 hours ago, SteveGrabowski0 said:

I'm not stroking the 970 so much as shitting on AMD for the 480. LOL them having the nerve to tout their power efficiency when they can't even power their card within spec. LOL putting a 6-pin on their reference just so they can look like the card draws less power. I hope Vulkan does become big though and that would be awesome to see AMD knock Nvidia down a peg or two with it, but it's way too early to claim victory just based off Doom.

Shit on Nvidia for having the nerve to claim 4GB of VRAM o na product, despite 512MB of it running like shit.

Shit on Nvidia for STILL having issues with laptop display scalers breaking with their mobile GPUs, despite the bug being known since 2014.

Shit on Nvidia for trying to make a closed ecosystem out of everything.

Shit on AMD for being overly ambitious.

Shit on AMD for being innovative.

Shit on AMD for taking risks.

 

 

 

Also  i think claiming victory based on the fact that Vulkan is basically Mantle, which is a API AMD has been working on since the beginning of GCN.... yeah, Vulkan is probz going to be a "AMD" biased API for a reason. DX12 should be relatively "neutral".

[Hayabusa1989]

On 8/2/2016 at 2:33 PM, roylapoutre said:

They used vaseline in the past, it was only 1000$

[Hayabusa1989]

Also,

The whole Nvidia vs AMD thing is so boring,

• RX 480 presents an entry into VR gaming and looking at the doors of high end
• GTX 1060 is the Doorway to High End
• There is quite a big price difference between the 2.
• Neither is better than the other, for the sales its trying to get they both hit a different price point and consumer so stop comparing.
• AMD has got advantages like Vulkan
• Nvidia's has things like SMP and just better performance per watt
• You could nitpick all day and find reasons why this or that should be different but overall they do a good job
• We have not seen anything from AMD yet on the side of High-End 
• This is a critical time period for AMD so i really can't judge anything they do until the come to the end of this phase
• Nvidia this Gen has decided that they will follow an Apple like approach of "give us your f**king money you peasents"
• AMD really screwed up on the power draw thing, 
• But then again the 1080 and 1070 should have had an extra power thingy 
• Because of that the AMD guys do have a point.
• But most importantly READ BELOW
• This is a Thread about Fucking Titan X's (Extreme High-Tier) so why the hell are we discussing Mid-Tier to Very Low High End ?

 

 

[Misanthrope]

18 minutes ago, Hayabusa1989 said:

Also,

The whole Nvidia vs AMD thing is so boring,

• There is quite a big price difference between the 2.

Actually no: both meet on the 250ish price bracket and even if that isn't so because of availability and high demand right now, both are way over 300 bucks anyway at the time of writting.

 

AMD is just potentially cheaper but with a crap cooler and gimped amount of VRAM vs the 1060 that runs cooler and doesn't has to deal with multiple amounts of vram for different skus.

 

In practice the basic choices, AIB, are virtually the same price.

[-BirdiE-]

1 hour ago, Prysin said:

Shit on AMD for being overly ambitious.

Shit on AMD for being innovative.

Shit on AMD for taking risks.

Move along.. No bias to see here...

 

 

 

 

 

Man, if we were to go buy only your posts... AMD has never done any wrong, and everything Nvidia has done is complete shit.

At least try to hide your unwavering bias towards AMD a little bit...

[SteveGrabowski0]

2 hours ago, Prysin said:

Shit on Nvidia for having the nerve to claim 4GB of VRAM o na product, despite 512MB of it running like shit.

Shit on Nvidia for STILL having issues with laptop display scalers breaking with their mobile GPUs, despite the bug being known since 2014.

Shit on Nvidia for trying to make a closed ecosystem out of everything.

Shit on AMD for being overly ambitious.

Shit on AMD for being innovative.

Shit on AMD for taking risks.

 

 

 

Also  i think claiming victory based on the fact that Vulkan is basically Mantle, which is a API AMD has been working on since the beginning of GCN.... yeah, Vulkan is probz going to be a "AMD" biased API for a reason. DX12 should be relatively "neutral".

AMD was innovative a couple of years ago with Hawaii and Mantle. Since then they have been nothing but disappointing. I couldn't believe they were ceding the high end until 2017 to the 1070/1080/GP102 until I saw how bad Polaris 10's performance per watt was. Unbelievable to be below 28 nm GM204 from 2014 with a 14 nm die in 2016 that was marketed for its efficiency. I figured Polaris 10 would end up where the 1060 is, eg performance of full GM204 in DX11 plus better DX12 performance at 120W, not cut down GM204 in DX11 plus better DX12 at 165W. It doesn't bode well for them being able to scale up and have a worthy competitor at the high end.

[patrickjp93]

14 minutes ago, SteveGrabowski0 said:

AMD was innovative a couple of years ago with Hawaii and Mantle. Since then they have been nothing but disappointing. I couldn't believe they were ceding the high end until 2017 to the 1070/1080/GP102 until I saw how bad Polaris 10's performance per watt was. Unbelievable to be below 28 nm GM204 from 2014 with a 14 nm die in 2016 that was marketed for its efficiency. I figured Polaris 10 would end up where the 1060 is, eg performance of full GM204 in DX11 plus better DX12 performance at 120W, not cut down GM204 in DX11 plus better DX12 at 165W. It doesn't bode well for them being able to scale up and have a worthy competitor at the high end.

There is some legitimacy to Raja Koduri's explanation that Polaris 10 was designed after Polaris 11 and that neither was designed with high clock rates in mind.

 

As a parallel, Intel Xeon designs vs. IBM Power 8 really explains what happens when trying to push clocks for chips which just weren't engineered with that idea in mind. A 24-core Xeon at stock 2.6GHz speeds has a 165W TDP. Meanwhile, a 12-core Power 8 at 4.7GHz has a 250W TDP. If you were to push the voltage through that Xeon necessary to hit those clocks, you'd be looking at a nuclear reactor a la 295x2, but in one monolithic die.

 

Polaris was designed to be ultra dense and ultra efficient at a certain upper clock speed. Pushing beyond that exposes the lack of extensible design, but that's not a bad thing. It's a natural fault in any IC design. Vega is a completely different architecture design with all the same hardware features. That will demonstrate if AMD's/RTG's team is really good enough to do both efficient and high-performance designs.

 

Where Nvidia is without question miles ahead is in software and drivers. Don't make the hardware work any harder than it has to to get X done in Y time. That's why the Nvidia driver is now just as big as the Windows NT kernel in Windows 10. All those corner cases which would be tough for the hardware to do the way the game programmers developed it are handled by efficient workarounds/conversions in the driver. Thus, Nvidia maintains an efficiency edge over AMD alongside equal to slightly better performance despite having what many would claim to be an inferior architecture.

[SteveGrabowski0]

2 minutes ago, patrickjp93 said:

There is some legitimacy to Raja Koduri's explanation that Polaris 10 was designed after Polaris 11 and that neither was designed with high clock rates in mind.

 

As a parallel, Intel Xeon designs vs. IBM Power 8 really explains what happens when trying to push clocks for chips which just weren't engineered with that idea in mind. A 24-core Xeon at stock 2.6GHz speeds has a 165W TDP. Meanwhile, a 12-core Power 8 at 4.7GHz has a 250W TDP. If you were to push the voltage through that Xeon necessary to hit those clocks, you'd be looking at a nuclear reactor a la 295x2, but in one monolithic die.

 

Polaris was designed to be ultra dense and ultra efficient at a certain upper clock speed. Pushing beyond that exposes the lack of extensible design, but that's not a bad thing. It's a natural fault in any IC design. Vega is a completely different architecture design with all the same hardware features. That will demonstrate if AMD's/RTG's team is really good enough to do both efficient and high-performance designs.

 

Where Nvidia is without question miles ahead is in software and drivers. Don't make the hardware work any harder than it has to to get X done in Y time. That's why the Nvidia driver is now just as big as the Windows NT kernel in Windows 10. All those corner cases which would be tough for the hardware to do the way the game programmers developed it are handled by efficient workarounds/conversions in the driver. Thus, Nvidia maintains an efficiency edge over AMD alongside equal to slightly better performance despite having what many would claim to be an inferior architecture.

Thanks for the explanation Patrick. That's good to hear Vega isn't just going to be a scaled up Polaris, it gives me some hope for AMD. But I had such high hopes for Polaris 10 a couple of weeks before launch when I saw Sapphire Italy tweeting that WCCF article claiming Polaris 10 would perform at Fury level at 150W. I was very tempted to sell my 970 to buy an RX 480 8GB when I saw that.

[Citadelen]

I wonder what the Titan after this will cost, maybe $1500, I can't believe people pay that amount money for just a GPU, the most I'd ever consider for a GPU is £500.

[Prysin]

1 hour ago, SteveGrabowski0 said:

Thanks for the explanation Patrick. That's good to hear Vega isn't just going to be a scaled up Polaris, it gives me some hope for AMD. But I had such high hopes for Polaris 10 a couple of weeks before launch when I saw Sapphire Italy tweeting that WCCF article claiming Polaris 10 would perform at Fury level at 150W. I was very tempted to sell my 970 to buy an RX 480 8GB when I saw that.

thing is though, there is more to polaris then the tech press has revealed. The Die size is tiny, sub 200mm2, which means it was made for high volume production. Yields were much more important to AMD then performance. It was meant to sell, HUGE amounts with as few losses as possible. This is probably why it has taken a while before the 470 came out, which is basically broken RX 480s.

 

Through Anandtechs testing, we see that Tesselation performance of the RX 480, is on par with a GTX 970. A Fury X, yes a full blown Fiji XT core, is slower then a 780Ti Kepler GPU in tesselation and miles behind a 970.

Meaning "Crapworks" wont hurt the 400 series remotely as bad as before. That being said, Pascal is even faster then Maxwell at tesselation, but the gap between AMD and Nvidia in that regards has been shrunk A LOT.

 

Then there is the improvement in ROP throughput, something they havent improved since GCN first came into existence.

Memory compression was further improved by a little bit. Nvidia is still better at this, but AMD isnt half bad at it either.

 

The little bugger is impressive, for being a 200$ card. It has features and improvements you would expect from a much "bigger" architecture with more cut down chips.

[Hayabusa1989]

On 8/3/2016 at 5:18 PM, Misanthrope said:

Actually no: both meet on the 250ish price bracket and even if that isn't so because of availability and high demand right now, both are way over 300 bucks anyway at the time of writting.

 

AMD is just potentially cheaper but with a crap cooler and gimped amount of VRAM vs the 1060 that runs cooler and doesn't has to deal with multiple amounts of vram for different skus.

 

In practice the basic choices, AIB, are virtually the same price.

Here in the UK it's a bigger gulf (from at least what i have seen),

I agree that the AIB Partners have made it close that is very sad but true,

However i feel like nobody read this bit :
 

On 8/3/2016 at 4:58 PM, Hayabusa1989 said:

This is a Thread about Fucking Titan X's (Extreme High-Tier) so why the hell are we discussing Mid-Tier to Very Low High End ?

So onto the Titan XP

 

Nvidia have ruined the point of the Titan, Yes it is the ULTIMATE CARD, however it was supposed to bridge a gap between Geforce and Quadro, There was a niché for this type of card for people who needed the extra features but wanted to also have fun and not be a boring Apple-Hippie "Creator".

Without having any extra features (like the original Titan did): 
This Card makes no sense for anybody unless you are very rich and or Very Impatient.

 

1080ti will be 95% of this card in performance and just over half the price.

[HKZeroFive]

12 minutes ago, Hayabusa1989 said:

1080ti will be 95% of this card in performance and just over half the price.

That's a bold assertion. Given that we don't even know if a GTX 1080Ti will even be released.

 

The Titan XP is an already cut-down GP102 chip, so a further cut-down chip is not going to yield much more performance than the GTX 1080. That and it'll have even less video memory (if they do cut it) than the GTX 1080 which isn't going to be appealing for consumers at all. Since the performance gap between the GTX 1080 and Titan XP is roughly 20-30%, that makes it very difficult to comfortably slot in further cut down card in between.

 

Also, 'just over half the price'? That's GTX 1080 pricing, I doubt they want to cannibalise their own cards.

 

This reminds me of the Kepler days, where the OG Titan was also a cut-down Titan. If AMD respond strongly with Vega, then that's the most likely time NVIDIA will release a full GP102 3840 core GTX 1080Ti or a Titan XP Black.

[PRTI]

On 8/2/2016 at 3:27 PM, Deli said:

Nvidia should sell the card at $2,000 to make it even more exclusive, like a Mclaren P12, a status symbol. To satisfy the hardcore Nvidia fans.

 

$1,200 a piece is just too cheap.

Nah they should create a dual GPU card which will sell for $5000 just because why not (GTX Titan Z master race)

[Hayabusa1989]

8 minutes ago, HKZeroFive said:

That's a bold assertion. Given that we don't even know if a GTX 1080Ti will even be released.

 

The Titan XP is an already cut-down GP102 chip, so a further cut-down chip is not going to yield much more performance than the GTX 1080. That and it'll have even less video memory (if they do cut it) than the GTX 1080 which isn't going to be appealing for consumers at all. Since the performance gap between the GTX 1080 and Titan XP is roughly 20-30%, that makes it very difficult to comfortably slot in further cut down card in between.

 

Also, 'just over half the price'? That's GTX 1080 pricing, I doubt they want to cannibalise their own cards.

 

This reminds me of the Kepler days, where the OG Titan was also a cut-down Titan. If AMD respond strongly with Vega, then that's the most likely time NVIDIA will release a full GP102 3840 core GTX 1080Ti or a Titan XP Black.

GP100 is the full 3840 core's ? At least thats what i thought,

Regardless there are and will ve Equal Performance Options at lower Prices. Regardless of Red or Green.

[HKZeroFive]

Just now, Hayabusa1989 said:

GP100 is the full 3840 core's ? At least thats what i thought,

Regardless there are and will ve Equal Performance Options at lower Prices. Regardless of Red or Green.

No, a full GP102 chip has 3840 cores, the Quadro P6000 proved that. The GP100 is basically GP102 with double precision. Great for computing but you can forget gaming.

 

I can agree with that statement, the price gap is pretty indicative of that.

[alphaproject]

Notice how the boards aren't nearly as popular as they used to be when talking about Titans....  I recall when the Titan X (OG) came out people were going nuts, talking about them left and right... now there's 1 topic here... this one... and most of the talking is not even about the Tian XP. ROTFL. People's attention spans are getting wiser.

 

Maybe some people are finally getting the hint. Maybe...???

[mrchow19910319]

On 8/2/2016 at 9:29 PM, roylapoutre said:

20% faster than a 1080 for a 80% increase in price, still not fully capable of 4K ultra @60fps... Still a scam from Nvidia, I love those Titan cards more and more every year  

 

[mrchow19910319]

On 8/2/2016 at 9:39 PM, Arokhantos said:

 

1400$ for custom advised price, while 1500$ for founders edition.

1000$ MSRP. 

[mrchow19910319]

On 3 August 2016 at 5:19 AM, patrickjp93 said:

Nope, right after GTC as the rumors go. I suspect Nvidia planned for a 10nm delay and prepared a 16nm variant of Volta way ahead of time. The great thing (for Nvidia) is that its R&D costs get cut down to release multiple generations.

What do you mean ?? Could you elaborate a lil bit??

[PRTI]

[mrchow19910319]

On 3 August 2016 at 5:46 AM, Prysin said:

The real interesting improvements with 16nm wont show before Volta/Vega.

Totally agree with you.