***Official Reviews Thread*** Nvidia Geforce GTX Titan - Launched Feb. 21, 2013

[tviceman]

So its efficiency only counts at resolutions that support your argument, rather than everything? It must be amazing to not have to think when you post, keep it up "genius."

Wow. LOL. You're right. LOLOLOLOLOLOLOLOLOLOLOL. From now on, lets run video cards at 50% load and test their efficiency. Make sure it's entirely CPU bound. That makes perfect sense.

I was right. You are a genius.

 

[BallaTheFeared]

They've power gated the DP units so while interesting I don't think it's worth spamming the thread with a large power efficiency graphic.

It still has 1/24 DP (I believe that's correct) when not using the enabled 1/3 DP performance setting.

 

[MrK6]

Yeah, so really, we dont want anyone buying them right now...do you really think they would shift the price?, even with no competition?

I don't think you understand competition that well, if at all. Competition doesn't have that great of an effect in a limited market, such as the upper echelon of enthusiast hardware. The price will stay the same and the people that spend $1000 on video cards will continue to do so, just the same as the GTX 690, 7990, and any other premium part. Nvidia didn't make this part to move volume, only to make profit off an already solid design in another market. The only reason I could see them moving the price is if there's a ton of backlash and they need a PR stunt.

 

[AdamK47]

I'm posting way too much, lol!

Anyways, did you see the voltage? I wonder if it's dynamic and went way up, or if he's actually using shipped voltage levels and riding high on the LN2? He's coming about 250Mhz short of his 680 score (at least the last one I saw).

Certainly did something with the voltage considering the modifications done on the back of the card hidden by a half inch of frost.

 

[BallaTheFeared]

Certainly did something with the voltage considering the modifications done on the back of the card hidden by a half inch of frost.

Missed the wires, good catch Adam!

 

[SolMiester]

Your own link, genius.

<snip>

Titan is 550mm^2 and is more efficient in perf/watt than GK104 and Tahiti when all are under full load. The "all resolutions" graph which you probably looked at and decided to respond with, is not indicative with either of those chips hitting full loads in the games tested.

Sweet sweet post bro.....

 

[Vesku]

It still has 1/24 DP (I believe that's correct) when not using the enabled 1/3 DP performance setting.

My point is once you factor in that some of those 7.1 Billion transistors are powered off for gaming, the efficiency is interesting from a design perspective but not amazing.

 

[MrK6]

Wow. LOL. You're right. LOLOLOLOLOLOLOLOLOLOLOL. From now on, lets run video cards at 50% load and test their efficiency. Make sure it's entirely CPU bound. That makes perfect sense.

I was right. You are a genius.

No, you just don't understand the hardware unfortunately, but I'm glad everyone can see your ignorance as well as your childish behavior, good to reference in the future. The same bottleneck happens at higher resolutions for other cards for the same reasons (limited by ROP's or or any number of other components) and they again they lose efficiency.

 

[MrK6]

Wonder how long it will live with such a high OC?

Probably years to come since it never sees 50C.

 

[SolMiester]

From Yahoo's dictionary:

Effiency:

1. The ratio of the effective or useful output to the total input in any system.
2. The ratio of the energy delivered by a machine to the energy supplied for its operation.

I don't want to discuss the technical aspects of this card but rather its general nature. Efficiency regarding purchases is relative so this is only my opinion.

Isn't that ROI rather than efficiency?

 

[SolMiester]

I don't think you understand competition that well, if at all. Competition doesn't have that great of an effect in a limited market, such as the upper echelon of enthusiast hardware. The price will stay the same and the people that spend $1000 on video cards will continue to do so, just the same as the GTX 690, 7990, and any other premium part. Nvidia didn't make this part to move volume, only to make profit off an already solid design in another market. The only reason I could see them moving the price is if there's a ton of backlash and they need a PR stunt.

Well actually, i do understand competition, thats why I bought it up, when has NV ever reduce price except for when there was competition. I bought up the idea of no purchasing to make a statement: ie backlash!

 

[RussianSensation]

I brought it up, but the predominant AMD crowd that is this thread no longer cares about perf/w, even though they don't have performance either. Now it's perf/$ that matters and nothing else (not even smoothness).

Not sure if serious. Performance/$ has always been either the most important or one of the most important metrics for GPU buyers. This isn't the first generation it's being talked about. If it wasn't, we'd all be buying $1-3K Titan setups. If performance/$ didn't matter, what's even the point of researching GPUs? We would all be rolling to the next fastest card out.

Regarding performance/watt, no one seriously cares when a GTX680 system is using 40-50W less on a system that's already drawing nearly 400W but GTX680 costs $450 and 1Ghz 7970s are going for $380 with free AAA games. Let me know what the break-even point is on the electricity costs with that 680 card after paying an $80 premium and getting a slower card.

And honestly, if you care that much about heat/costs as you keep talking about it so often referring to HD7970 as a "Fermi" (even though an 1180mhz HD7970 uses just 10-20W more power than a stock 580 while being 60% faster), why are you running GTX470 @ 960mhz in Tri-SLI? It's interesting how people buying $500 GPUs suddenly started to care about 40-50W of extra power and yet bought Fermis over HD5000/6000 series. Performance/watt is nice but I'd take a card with 600W of power if I could if it was 2x faster than the Titan for $500. Performance/watt is a nice metric but under no circumstances can it somehow justify GTX680's price premium or the Titan's unless you are running hundreds . HD7950 OC for $280-290 will match a GTX680 or even beat it and still use just about 225W of power doing it. Try doing break-even analysis on power consumption costs on that!

I'm surprised not much has been said about Titan's perf/watt efficiency in this thread. It's pretty damn incredible, putting gk104 and Tahiti to shame.

From an engineering point of view indeed it is impressive. Most enthusiasts who have overclocked i5/i7s and 185W GPUs (GTX680 and higher) don't really care about GPU performance/watt if the differences are so small as they are between 7970 and 680. It matters more if you are going Tri-SLI/CF and you are stuck with hot and loud reference blowers. For after-market cards, you fix the noise levels and temperatures. 40-50W of extra power is nothing in costs or heat on an annual basis unless you are gaming/using the GPU 24/7. In that case as I said before, those people better be running 92% platinum PSUs.

The difference in idle/load power consumption between motherboards on Z77 chipset with IVB @ 4.6ghz is greater than the difference between GTX680 and HD7970. I don't see people caring about 40W of extra power their high-end Z77 board uses vs. a lower end one.

http://www.xbitlabs.com/images/mainboards/asrock-z77-extreme3/power-5.png
http://www.xbitlabs.com/images/mainboards/asrock-z77-extreme3/power-7.png

Why don't people compare motherboards on a performance/watt basis too? Going from a 1st generation Core i7 @ 3.6ghz to 4.2ghz increased power consumption 100W. I never saw anyone here caring. I agree though that from an engineering perspective it's very impressive that a 7.1B 551mm2 chip is using barely more power than an HD7970GE. The Titan's near 1.0V GPU voltage vs. 1.256V on the HD7970GE is a big factor here.

 

[SolMiester]

Probably years to come since it never sees 50C.

Titan runs cool as well, and perhaps cooler under water, however have put ceiling on the OV for the reason that high voltage reduces the life of the product....I take it you are OV?

 

[SolMiester]

Not sure if serious. Performance/$ has always been either the most important or one of the most important metrics for GPU buyers. This isn't the first generation it's being talked about. If it wasn't, we'd all be buying $1-3K Titan setups. If performance/$ didn't matter, what's even the point of researching GPUs? We would all be rolling to the next fastest card out.

Regarding performance/watt, no one seriously cares when a GTX680 system is using 40W less on a system that's already drawing nearly 400W but GTX680 costs $450 and 1Ghz 7970s are going for $380 with free AAA games. Let me know what the break-even point is on the electricity costs with that 680 card after paying an $80 premium.

And honestly, if you care that much about heat/costs as you keep talking about it so often referring to HD7970 as a "Fermi" (even though an 1180mhz HD7970 uses just 10-20W more power than a stock 580 while being 60% faster), why are you running GTX470 @ 960mhz in Tri-SLI? It's interesting how people buying $500 GPUs suddenly started to care about 40-50W of extra power and yet bought Fermis over HD5000/6000 series. Performance/watt is nice but I'd take a card with 500W of power if I could if it was 2x faster than the Titan for $500. Performance/watt is a nice metric but under no circumstances can it somehow justify GTX680's price premium. HD7950 OC for $280-290 will match a GTX680 or even beat it and still use just about 225W of power doing it. Try doing break-even analysis on power consumption costs on that!

Perf/$ wasnt the be all when Fermi came out, Im sure you remember it was perf/w, even though Fermi was the perf/$ winner.

 

[raildogg]

Isn't that ROI rather than efficiency?

ROI is a broad term and I don't feel comfortable using it. It can mean many things really. But maybe I misspoke about the efficiency part but I felt I had to respond since I don't consider the Titan efficient (in a non-technical way). Then again, efficiency is also a broad term that can be used in many different ways. I do appreciate your response however.

 

[MrK6]

Well actually, i do understand competition, thats why I bought it up, when has NV ever reduce price except for when there was competition. I bought up the idea of no purchasing to make a statement: ie backlash!

Then I misunderstood your point and I agree. It will be interesting to see how it develops. :thumbsup:

Titan runs cool as well, and perhaps cooler under water, however have put ceiling on the OV for the reason that high voltage reduces the life of the product....I take it you are OV?

Exactly, and unfortunately nvidia has locked down their specs in reference to air cooling and not water cooling. The extreme results seen under LN2 have shown that this architecture has incredibly potential, especially if the rumors of nvidia offloading leaky chips to the consumer market are true. Unfortunately, unless there's an easy trimmer mod, that potential will never be realized with more conventional cooling methods.

 

[Imouto]

Perf/$ wasnt the be all when Fermi came out, Im sure you remember it was perf/w, even though Fermi was the perf/$ winner.

http://www.techpowerup.com/reviews/ASUS/GeForce_GTX_470/32.html

I, I don't even...

 

[SolMiester]

Then I misunderstood your point and I agree. It will be interesting to see how it develops. :thumbsup:

Exactly, and unfortunately nvidia has locked down their specs in reference to air cooling and not water cooling. The extreme results seen under LN2 have shown that this architecture has incredibly potential, especially if the rumors of nvidia offloading leaky chips to the consumer market are true. Unfortunately, unless there's an easy trimmer mod, that potential will never be realized with more conventional cooling methods.

Oh, so the locked down is on air only?...or are you referencing AIB partner not having the lock down when under water?

 

[SolMiester]

http://www.techpowerup.com/reviews/ASUS/GeForce_GTX_470/32.html

I, I don't even...

I guess Im confused too when it comes to GPU load, and what resolutions to reference!.
As stated before, hardly much point measuring efficiency or perf/$ when the GPU is half load...so I was looking at 16x10 & 19x10, where it does/did have the lead.<shrug>
So, as I said, fans appear to move the mark....1 min its perf/w then perf/$

 

[RussianSensation]

Well actually, i do understand competition, thats why I bought it up, when has NV ever reduce price except for when there was competition. I bought up the idea of no purchasing to make a statement: ie backlash!

NV's pricing premium during GTX280/480/580 was actually "reasonable" if you could call it that (HD6950 Unlocked in CF cost about the same as a GTX580). Now NV doubled what you pay for 1% increase in performance. Therefore, NV raised the price regardless of where HD7970Ghz stands. If AMD lowers the price of HD7970 to $299, the Titan still won't drop from $1,000 because NV knows its fans will pay that much for it. How do they know? Because they sold GTX690 for $1K without problems for almost a year. NV is slowly creating a new enthusiast price level.

Wonder how long it will live with such a high OC?

A long time. Asus warranties the Asus Matrix 7970 up to 1.38V I believe which suggests Tahiti silicon is pretty robust. Even if his card dies, it fully paid for itself many times over with bitcoin mining. A $900 Titan is a depreciating asset and chances are it'll lose $500 when $500 20nm GPUs launch in 2014. HD7970 that paid for itself was akin to getting a $550 card for free and saving $900 out of your pocket for your next upgrade, for instance when 20nm Maxwell/Volcanic Islands launch. MrK6 got his as soon as possible being aware of this awesome perk.

Perf/$ wasnt the be all when Fermi came out, Im sure you remember it was perf/w, even though Fermi was the perf/$ winner.

On the whole, HD5000/6000 beat Fermi in both performance/watt and performance/$. Not sure how you derived those conclusions. There wasn't a single card in NV's line that could touch a $230 HD6950 2GB unlocked into a 6970 in performance/$ on the high-end, especially not at 1600P. There isn't even a point of discussing HD5000 vs. GTX470 as a $259 HD5850 OC came within 15% of a GTX480 in performance and got there 6 months earlier! Sure they were specific cases like the awesome GTX460 OC, but overall both GTX400/500 lost on price/performance if you look across many price levels and prices of that generation at least in North America. This is why Kepler is so great because it put NV into the driver seat by reclaiming the performance/watt and temporarily price/performance when GTX670/680 launched. That was a slam dunk launch. I have no idea what the price premiums for NV were in Oceania region but from what I heard NV charges even higher premiums there. I can't see how GTX400/500 series were winning price/performance metric in Europe or Oceania where in most countries NV cards cost more.

 

[Vesku]

$500 20nm GPUs launch in 2014.

Probably $550-600 on launch, imo, maybe more depending upon how much more skewed the wafer price over time curve will be with TSMC supplying Apple with 20nm chips.

Unless Nvidia provides Titan owners with at least some of the Quadro series driver features they are still pretty much gouging ~$200-300.

 

[BallaTheFeared]

HD5000/6000 beat Fermi in both performance/watt and performance/$. Not sure how you derived those conclusions. There wasn't a single card in NV's line that could touch a $230 HD6950 2GB unlocked into a 6970 in performance/$ on the high-end, especially not at 1600P. There isn't even a point of discussing HD5000 vs. GTX470 as a $259 HD5850 OC came within 15% of a GTX480 in performance and got there 6 months earlier! Sure they were specific cases like the awesome GTX460 OC, but overall both GTX400/500 lost on price/performance if you look across many price levels and prices of that generation at least in North America. I have no idea what the price premiums for NV were in Oceania region but from what I heard NV charges even higher premiums there.

So we're going to use the least demanding dx9 games, and ignore the demanding DX11 titles where often Fermi would destroy 5xxx series? Well that's a polar opposite from you!

Also unlocking being considered in favor of AMD's perf/$ as if they intended it, lol that's funny!

And you left out overclocking, which is odd since once again a polar opposite from this generation.. Need I remind you where the 6970 stood after overclocking too place?










In other news, anyone try overclocking their monitor with Nvidia's info? I was able to get almost 20% more performance out of mine, 71Hz @ 1080p, pretty crappy but hey it's 20% more frames than I could see before!

 

[RussianSensation]

So we're going to use the least demanding dx9 games, and ignore the demanding DX11 titles where Fermi destroyed 5xxx series? Well that's a polar opposite from you!

Talking about average performance:
http://alienbabeltech.com/abt/viewtopic.php?p=41174

Add overclocking to that, it makes no difference.

Even if GTX580 was OC to 950mhz, it wouldnt' be more than 30% faster than an HD6950 unlocked. HD6950 unlocked was going on avg for $250 in retail when GTX580 was going for $450-500. GTX570 could never beat HD6970 by 30% on average, not a chance. At 1600P, NV's cards got worse too.

Also unlocking being considered in favor of AMD's perf/$ as if they intended it, lol that's funny!

Why wouldn't it be considered? It was 15 min of work max. Your specific case of buying $50 x 3 GTX470s and waterblocks isn't in-line with market prices. So if you got a special discount on those cards, then price/performance could have been better for you.

And you left out overclocking, which is odd since once again a polar opposite from this generation.. Need I remind you where the 6970 stood after overclocking too place?

Where? It takes a 770mhz GTX470 just to match a stock 6970. Without waterblocks, most 470s crapped out at that point. Take a 470/570 and add a waterblock + cooling system and it your price/performance was still worse compared to an HD5850 OC / HD6950 OC+Unlocked.

P.S. HD5000/6000 also paid for themselves with bitcoins. You know what it means? Most of us have been upgrading for free starting with HD4000 series. OOps, that's thousands of dollars saved. Let me know how Fermi won on price/performance. And this wasn't a special case either since most could unlock 6950s for at least 6 months and the success rate was > 90%. The interesting part is those HD6950 2GB unlocked / CF owners can still use those cards while GTX470/570 / SLI are running into VRAM bottlenecks.

Probably $550-600 on launch, imo, maybe more depending upon how much more skewed the wafer price over time curve will be with TSMC supplying Apple with 20nm chips.

I figure since NV/AMD added 35%+ performance moving from 40nm to 28nm, $550-600 GPUs should come very close to the Titan.

 

[Cookie Monster]

No, you just don't understand the hardware unfortunately, but I'm glad everyone can see your ignorance as well as your childish behavior, good to reference in the future. The same bottleneck happens at higher resolutions for other cards for the same reasons (limited by ROP's or or any number of other components) and they again they lose efficiency.

Well for someone who supposedly understands hardware, comments like "Limited by ROPs" makes no sense...

But bickering aside, as with all launches drivers for Titan may not be optimal at this point in time which was highlighted by its OpenCL bugs and rather low GPGPU performance. For instance from the AT review:

For their part, NVIDIA of course already has OpenCL working on GK110 with Tesla. The issue is that somewhere between that and bringing up GK110 for Titan by integrating it into NVIDIA’s mainline GeForce drivers – specifically the new R314 branch – OpenCL support was broken. As a result we expect this will be fixed in short order, but it’s not something NVIDIA checked for ahead of the press launch of Titan, and it’s not something they could fix in time for today’s article.

So I think the potential is there to get even better seeing as the GK110 was initially optimized as Tesla products. Just a matter of time like how the 7970 only performed ~20% faster than the GTX580 initially til the gap widened with constant driver updates.

One thing that caught my eye is that from age old past where GPUs were compared via absolute avg FPS, is this figure really telling the truth? Id love if PcPer got their setup all working nicely. Its similiar to arguing about pure performance numbers in GPGPU applications but reality is that the numbers alone don't dictate which solution is more robust and overall better. But going back to the original point, its something like this that caught my eye e.g.

And i still don't get this performance/price "fairness" debate raging in this thread. Im surprised people aren't up in arms on Intel and their "EE" line or motherboard makers that charge more for having that extra sata port and a more exotic VRM heatsinks. It would be such a nice world if prices were dictated by performance..

I think this negative interest expressed from alot of users actually lead me to believe that they would love to have this card except that they are somehow upset because its way outside their price range.

Its a really power efficient video card given its power envelope. I wonder if they will ever release a 13SMX (cheaper) or 15SMX versions ($1499, yep I said it ), e.g. Titan II or something.

 

[SolMiester]

NV's pricing premium during GTX280/480/580 was actually "reasonable" if you could call it that (HD6950 Unlocked in CF cost about the same as a GTX580). Now NV doubled what you pay for 1% increase in performance. Therefore, NV raised the price regardless of where HD7970Ghz stands. If AMD lowers the price of HD7970 to $299, the Titan still won't drop from $1,000 because NV knows its fans will pay that much for it. How do they know? Because they sold GTX690 for $1K without problems for almost a year. NV is slowly creating a new enthusiast price level.



A long time. Asus warranties the Asus Matrix 7970 up to 1.38V I believe which suggests Tahiti silicon is pretty robust. Even if his card dies, it fully paid for itself many times over with bitcoin mining. A $900 Titan is a depreciating asset and chances are it'll lose $500 when $500 20nm GPUs launch in 2014. HD7970 that paid for itself was akin to getting a $550 card for free and saving $900 out of your pocket for your next upgrade, for instance when 20nm Maxwell/Volcanic Islands launch. MrK6 got his as soon as possible being aware of this awesome perk.



On the whole, HD5000/6000 beat Fermi in both performance/watt and performance/$. Not sure how you derived those conclusions. There wasn't a single card in NV's line that could touch a $230 HD6950 2GB unlocked into a 6970 in performance/$ on the high-end, especially not at 1600P. There isn't even a point of discussing HD5000 vs. GTX470 as a $259 HD5850 OC came within 15% of a GTX480 in performance and got there 6 months earlier! Sure they were specific cases like the awesome GTX460 OC, but overall both GTX400/500 lost on price/performance if you look across many price levels and prices of that generation at least in North America. This is why Kepler is so great because it put NV into the driver seat by reclaiming the performance/watt and temporarily price/performance when GTX670/680 launched. That was a slam dunk launch. I have no idea what the price premiums for NV were in Oceania region but from what I heard NV charges even higher premiums there. I can't see how GTX400/500 series were winning price/performance metric in Europe or Oceania where in most countries NV cards cost more.

You know RS, I'm not going to argue with you, my memory probably isnt what it used to be, Im just going by what I recall at that time.

What is however very scary, is you either have one of the best memories I have ever come across, or you have some sort of database of facts and figures along with pre-defined Q&As,.....or type very fast...LOL. There is probably some sort of job out there just for you mate!