Titan Results Thread (OC/Performance/Benching)

[boxleitnerb]

What I read in the review is that overvolting gives more turbo bins and is not really overvolting. Your card is hitting 265watts and thus downclocking/volting to protect the power circuitry.

It may seem crazy, but the fan could be drawing enough wattage at higher speeds and higher GPU loads to hit the 265 watt limit and cause the GPU to underclock also.

That still doesn't explain why the card thinks 11xx MHz@1.0x V is okay when at stock the maximum clocks for the 1.0x V voltage point are in the 9xx range. After reinstalling drivers I've seen 1150 MHz@1.15 and 1.1625V (what I would expect) respectively, but only for a short period. Then the strange voltage drop occurs.

If the card hits any limit, it should lower voltages and clocks accordingly.

 

[VulgarDisplay]

That still doesn't explain why the card thinks 11xx MHz@1.0x V is okay when at stock the maximum clocks for the 1.0x V voltage point are in the 9xx range. After reinstalling drivers I've seen 1150 MHz@1.15 and 1.1625V (what I would expect) respectively, but only for a short period. Then the strange voltage drop occurs.

If the card hits any limit, it should lower voltages and clocks accordingly.

I'm not really sure what you are saying is happening I guess.

 

[boxleitnerb]

Bofox got it right I think. It seems that the whole clock curve is shifted by the GPU offset. So all points on that curve are affected, not just the clocks above 1006 MHz. That would certainly explain what I'm seeing here.

I just remembered it's the same with CPU OC with voltage offset, too. Then, all clocks have different voltages, for example idle clocks as well.

 

[AdamK47]

There is something I'm curious about. I've been reading from people here that clocking the memory too high causes reduction in core clock speeds. Does that reduction also reduce GPU temps? I'm willing to take the tradeoff of reduced core speeds with higher memory speeds if that also translates into lower GPU temperatures.

 

[PowerK]

Another user benchmarks.
680 Lightning vs Titan
http://www.overclock.net/t/1363440/nvidia-geforce-gtx-titan-owners-club/1960_20#post_19429141

 

[Annisman*]

There is something I'm curious about. I've been reading from people here that clocking the memory too high causes reduction in core clock speeds. Does that reduction also reduce GPU temps? I'm willing to take the tradeoff of reduced core speeds with higher memory speeds if that also translates into lower GPU temperatures.

Something I never actually observed while benching, I had a pretty agressive fan profile setup anyways, kept the gpus below 75C at all times.

 

[blackened23]

Something I never actually observed while benching, I had a pretty agressive fan profile setup anyways, kept the gpus below 75C at all times.

But, aggressive fan settings cause throttling? Seems like everyone at OCN is using 60% fan, because 68% and over cause a lot of core speed throttling.

 

[Annisman*]

But, aggressive fan settings cause throttling? Seems like everyone at OCN is using 60% fan, because 68% and over cause a lot of core speed throttling.

I never saw that with my cards, maybe I kept them so cool (I've got TONS of air hitting the cards) that it never cranked that high.

 

[blackened23]

I never saw that with my cards, maybe I kept them so cool (I've got TONS of air hitting the cards) that it never cranked that high.

The thing is, temperatures aren't really as much a contributor towards throttling as it was with GK104, GPU boost 2.0 is being reported to have very different behaviour. Are you using the actual manual fan to cool them or another method? Water perhaps? Fanspeeds and Total power are the most important from what OCN'ers are stating - Are you using the MSI afterburner or EVGA precision overlay to view the clockspeeds while doing a benchmark? Everyone over at OCN has noted downclocking when using 68% or higher manual fan speed.

If you go in the EVGA precision onscreen display controls, you can view the gpu/memory clockspeed dynamically onscreen as you game or perform a benchmark. It is an onscreen overlay which is similar to the steam overlay, but it shows your GPU / memory / etc clockspeeds and voltages. This will allow you to view how the clockspeeds change dynamically when you run 3dmark, for instance.

 

[notty22]

It's clear he's not using water, all you have to do is read the posts. Just because people are mob mentality something does not make it true, a fan can use up to 10 watts of power, which probably comes from board power and the TDP limit function, but that would only draw back the turbo slightly. The other is temperature, which does not start to 79c and is adjustable, I'm almost positive without working with one myself.

edit: As with much of 28nm without drastic voltage controls on top of extreme cooling air or water, 1150-1200mhz is it. There are examples of more, but not much. Asking a 7 billion transistor chip to break clock records of chips 2/3 and 1/2 it's size , is near impossible, imo.

 

[blackened23]

It's clear he's not using water, all you have to do is read the posts. Just because people are mob mentality something does not make it true, a fan can use up to 10 watts of power, which probably comes from board power and the TDP limit function, but that would only draw back the turbo slightly. The other is temperature, which does not start to 79c and is adjustable, I'm almost positive without working with one myself.

edit: As with much of 28nm without drastic voltage controls on top of extreme cooling air or water, 1150-1200mhz is it. There are examples of more, but not much. Asking a 7 billion transistor chip to break clock records of chips 2/3 and 1/2 it's size , is near impossible, imo.

Manual fan speed 70% or higher is one variable that has been confirmed to cause throttling with EVGA titans and the 314.09 drivers. Period. Fact. If that bothers you, i'm sorry. Currently, they are trying to find a solution to this "problem", and I have been looking at the thread myself as I am interested in getting a Titan when the custom versions (such as the lightning) are released. This pertains to me because I obviously like the lightning line of GPUs, and will certainly buy a lightning Titan. This has also been noted at XS. Perhaps you should argue with them and tell them how crazy they are and don't know what they're talking about, I mean after all your superior operating knowledge over the guys that actually use and benchmark it every day could help.

 

[notty22]

Blah blah blah,I'm not going to argue over a possible bug. So keep the fan at 69%, that's fast for a blower if you need someone to tell you that.
By the way you have been obviously , publicly irked about the Titan launch in so many ways it's beyond strange. I can't grasp your motives, to tell you the truth. So now you want one?


edit: By the way,blackened23 , aren't you xoleras over at H ?

 

[Annisman*]

The thing is, temperatures aren't really as much a contributor towards throttling as it was with GK104, GPU boost 2.0 is being reported to have very different behaviour. Are you using the actual manual fan to cool them or another method? Water perhaps? Fanspeeds and Total power are the most important from what OCN'ers are stating - Are you using the MSI afterburner or EVGA precision overlay to view the clockspeeds while doing a benchmark? Everyone over at OCN has noted downclocking when using 68% or higher manual fan speed.

If you go in the EVGA precision onscreen display controls, you can view the gpu/memory clockspeed dynamically onscreen as you game or perform a benchmark. It is an onscreen overlay which is similar to the steam overlay, but it shows your GPU / memory / etc clockspeeds and voltages. This will allow you to view how the clockspeeds change dynamically when you run 3dmark, for instance.

I don't use manual, but I set an aggressive automatic profile.

I use EVGA precision, and I am very familiar with keeping track of all my numbers in real time, using my Logitech G15 LCD display, something I have been doing for many years now.

If what you are saying is true, I'll def. stay away from a manual profile in the future.

 

[AdamK47]

Ran some quick tests with 3DMark11 last night. Would have run 3DMark (2013), but there are still problems with 3-Way and 4-Way SLI scaling with it.

P27540
http://www.3dmark.com/3dm11/6135522

X14286
http://www.3dmark.com/3dm11/6135547

 

[AdamK47]

Performed some actual stress testing on the 3-Way Titans. Settled on some stable settings after a day of testing.

Found settings that passed. 1.5 hours of Unigine Valley @ 2560x1600 maxed with FXAA enabled in drivers followed by 1.5 hours of Unigine Tropics @ 2560x1600 maxed also with FXAA enabled in drivers. This produced a great amount of heat. My computer room is nice an toasty right now. The settings I used are as follows:

MSI Afterburner is set at +20mV core, 106% power limit, 91 degrees C max temp, +125 core, +100 memory.

This is what MSI Afterburner showed after exiting towards the end of Unigine testing (Tropics).

Tried higher memory clocks (+500, +250, and +200), but found some GPUs would spontaneously downclock to 837MHz and would remain there. At that time the GPU usage would be around only 50%. FPS would be limited as well. Only a reboot would fix it. Seems like it performs a safeguard rather than locking up completely. Found that aspect of Titan overclocking interesting. +100 on the memory did not produce this problem.

 

[Annisman*]

The Titan is an interesting specimen, the memory will overclock like a banshee, but going too far will ultimately degrade performance.

 

[boxleitnerb]

I did the unthinkable...I flashed the bios on my Titan. 150% power target, no throttling anymore

Just did a run of the AvP Benchmark with 1163/3506. With the original bios, the card would throttle to as low as 1050 MHz even without reaching 106% PT. That's a thing of the past now. I'll have to see what clocks are doable without increasing power consumption too much and more importantly, what makes sense considering that Titan is similarly bandwidth bottlenecked when overclocked.

 

[Grooveriding]

I did the unthinkable...I flashed the bios on my Titan. 150% power target, no throttling anymore

Just did a run of the AvP Benchmark with 1163/3506. With the original bios, the card would throttle to as low as 1050 MHz even without reaching 106% PT. That's a thing of the past now. I'll have to see what clocks are doable without increasing power consumption too much and more importantly, what makes sense considering that Titan is similarly bandwidth bottlenecked when overclocked.

Can you provide a link to the BIOS you used, alternately PM a like to it, please ?

You can use http://www.datafilehost.com/ to host it.

Thanks.

 

[boxleitnerb]

I'm not allowed to share it at this time as the bios was not made by me. Sorry

 

[tviceman]

I did the unthinkable...I flashed the bios on my Titan. 150% power target, no throttling anymore

Just did a run of the AvP Benchmark with 1163/3506. With the original bios, the card would throttle to as low as 1050 MHz even without reaching 106% PT. That's a thing of the past now. I'll have to see what clocks are doable without increasing power consumption too much and more importantly, what makes sense considering that Titan is similarly bandwidth bottlenecked when overclocked.

Holy crap! What scaling are you getting with that big of an overclock on both the core and vram? Can you post a before and after benchmark?

 

[blackened23]

Can you provide a link to the BIOS you used, alternately PM a like to it, please ?

You can use http://www.datafilehost.com/ to host it.

Thanks.

It is linked in the OCN Titan owner's thread as the 150% BIOS. Others aren't having luck with it at all and many are still having throttle. That's practically the only thing being discussed for about 100 pages now. Anyway, I don't have a link to the direct post, but I saw it in that thread - if I manage to find it I will post it. You will find it there still, somewhere.

http://www.overclock.net/t/1363440/nvidia-geforce-gtx-titan-owners-club/3450

It's a pretty good resource (way better than here) for pertinent Titan information. Unfortunately no definitive throttle fix has been found, and I think it may be a driver issue - compounding this is the fact that GPU boost 2.0 is a complete mystery in terms of throttle prioritization (what it's supposed to do DOESNT happen) and that some people apparently don't get throttle while many others do. I don't know. Driver issue perhaps - It happens now and then, I even find anomalies with 314.09 and 314.14 on my 680 which causes them to downclock at high manual fan speeds. That is perhaps the stupidest thing i've managed to find (it is also reported in a multi page thread on the nvidia driver forum), it's almost like nvidia is telling us to make our cards get as hot as possible for best performance? That doesn't make sense.

Either way, the card is still a great card at stock speeds. The main issue is GPU boost 2.0 not working as intended - ManuelG says that the nvidia driver team is aware of the throttling and manual fan issues with 314.x driver issues and they have reproduced it. So the ball is in their court now for those affected. I'm sure it will all be sorted out.

 

[boxleitnerb]

Holy crap! What scaling are you getting with that big of an overclock on both the core and vram? Can you post a before and after benchmark?

I surely will, right now I'm drunk and won't mess with the cards in that state, lol
Before I ran 1202/3506 on the worse one of my two cards

 

[AdamK47]

Did some more stress testing today. Ran tests until the heat in my little computer room became bit too unbearable. That's just how I like it though when I do this sort of testing. I want worst case scenarios. It must be part my nature since I do software load testing for a living and want to get things to their failure point.

These are the settings I'm going to use for my 3-Way Titan setup:
+20mV core, 106% power limit, 90 degrees C max temp, +100 core, +0 memory.

Ran Unigine Heaven for an hour with +100 memory again today and saw the problem I noticed yesterday. All of a sudden the GPU usage would be cut down to between 50% and 60% on affected GPUs. It would tank the framerates. Like yesterday, the only way to recover from this is to reboot the PC. This has to be some sort of fail safe implemented by Nvidia for the Titan. Keeping the memory at default speeds (+0) kept the GPUs functioning fine. There were no sudden drops in usage and the tests ran for several hours.

Another interesting fact is that adding voltage increases clocks. With a +0 voltage setting and a +100 core setting the GPUs were as follows:
GPU1 @ 1.37V / 1045MHz
GPU2 @ 1.37V / 1058MHz
GPU3 @ 1.50V / 1071MHz

With +100 core and an offset of +20mV the GPUs were as follows:
GPU1 @ 1.50V / 1058MHz
GPU2 @ 1.50V / 1071MHz
GPU3 @ 1.62V / 1097MHz

 

[Grooveriding]

Nvidia is going to sell the SLI bridge used for the Titan marketing shots.

http://www.neogaf.com/forum/showpost.php?p=48030106&postcount=1736

Thoughts? Since it lights up I expect them to charge $1000 for it.

 

[blackened23]

Mid range SLI bridge