Overclock.net › Forums › Graphics Cards › NVIDIA › [Official] NVIDIA Titan X Pascal Owners Thread
New Posts  All Forums:Forum Nav:

[Official] NVIDIA Titan X Pascal Owners Thread - Page 285

post #2841 of 7589
Quote:
Originally Posted by Z0eff View Post

What's so bad about linus?

It's over simplified content for the general masses. A lot of conclusions are made that aren't accurate or are based off of improper testing techniques and comparisons, which is fine for the layman. But I wouldn't put any stock into their analysis of the feasibility of using a xeon processor as a gaming cpu, for example.
post #2842 of 7589
Add me to the old enough to know better but still too young to care crowd....

2 x Titan X (P's) in SLI. EK waterblocks, factory backplates (with a little Rube Goldberg know how and thermal paste). Asus ROG 3 way SLI bridge (due to slot spacing as well as I wanted both SLI tabs on each card connected since the Nvidia HB bridges wont fit EK blocks) +150 on the clocks, no change to memory times.

13598 Fire Strike Ultra score:
http://www.3dmark.com/3dm/14308478

post #2843 of 7589
Quote:
Originally Posted by skypine27 View Post

Add me to the old enough to know better but still too young to care crowd....

2 x Titan X (P's) in SLI. EK waterblocks, factory backplates (with a little Rube Goldberg know how and thermal paste). Asus ROG 3 way SLI bridge (due to slot spacing as well as I wanted both SLI tabs on each card connected since the Nvidia HB bridges wont fit EK blocks) +150 on the clocks, no change to memory times.

13598 Fire Strike Ultra score:
http://www.3dmark.com/3dm/14308478


You can cut HB bridges easily. No damage. Then those fit too.
post #2844 of 7589
Quote:
Originally Posted by toncij View Post

You can cut HB bridges easily. No damage. Then those fit too.

Thx for the info (also found a PC world article where the completely removed the plastic shroud from the bridge to make it fit). For me it doesn't matter since I ordered the wrong length HB bridge anyway. I thought the "3 slot" model actually meant Tri SLI. The reason I wanted a Tri SLI length bridge is because my slot spacing for 2x SLI is equal to that of 3 x SLI setups (i.e. 4 slots).

I think its officially called "spaced SLI". However, when I ordered the HB bridge opening day, it wasn't in the forefront of my mind that since nvidia had officially given up 3 and 4 way sli, that 3 "slot" bridge actually means 3 slots and not 3 x cards smile.gif So its too short.

I talked to EK tech support and they confirmed they are working on their own custom HB bridges and also mentioned that the EVGA PRO HB bridges will fit, I would need this one:
http://www.evga.com/Products/Product.aspx?pn=100-2W-0028-LR

But Ill wait a while and see if any better looking ones come out.

Thx for the info.
post #2845 of 7589
Quote:
Originally Posted by HyperMatrix View Post

It's just unfortunate because I find 2 cards to be a little underpowered right now for maxed out 165Hz. I even set my monitor back to 144Hz to help lower gpu usage. But even then there are times where FPS will drop to 110~ FPS in GTA V. Or even dip as low as 80fps~ in Witcher 3. With blocks and a modified bios, best I can expect is another 10% performance, which is ok for 120Hz gaming at 1440p, but ends up being shy of 144-165Hz. If only the Titan XP was a full 610mm2 die like GP100. That extra 30% would have been amazing. I do like the simplicity of 2-card SLI. And much happier with heat management as a result.

Ah well. Here's hoping for 600mm2 Volta with HBM2 next year.

30%? 3840 cores vs 3584 is only 7% more, hardly worth waiting if you think that would be the Volta. Volta needs to be in the range of 5k cores to be a meaningful upgrade. 5k can be placed on a chip the size of GP100 since you don't need FP64 cores etc.

However, I think we'll first see a 14nm (Samsung fabs) refresh of the Pascal like 3840 cores, 1.8/2GHz official base/boost and keeping the same power envelope. That'd probably happen early 2017 as 1080Ti and possible "Titan X Black (Pascal)".





Here is my latest testing: http://imgur.com/a/XPoK3

Titan X (Pascal) is projected, calculated result as average of 60% faster than Titan X (Maxwell) and 33% faster than 1080 (max theoretical advantage), but 1080, 1080 SLI na old Titan X are measured in exactly the same environment and setup.

Now, except for that, theoretically, Titan X (Pascal) beats 1080 SLI in most situations, I've ran across a very strange thing.. 3 different 1080 cards (EVGA, EVGA, FE) in a single and SLI config exhibit a strange behavior in Rise of The Tomb Raider benchmark - Syria (2nd test) has a hiccup at the beginning and is abysmally slow in 3rd test too dropping minimum framerate to 3-4 in the 2nd and 15-16 in the 3rd test. Old Titan X (Maxwell) does not have that problem, minimum is very high for it.

Can anyone test on their own Titan X (Pascal)?

Thx.
post #2846 of 7589
Quote:
Originally Posted by toncij View Post

30%? 3840 cores vs 3584 is only 7% more, hardly worth waiting if you think that would be the Volta. .

Die size of GP100. Just as Nvidia did with the Titan X over the OG Titan by dropping fp64.
post #2847 of 7589
hi everyone,

I'm not particularly tech savvy but my Titan XP seems to be generating massive heat. I know it's a good thing the heat is being exhausted but I would like to know if it's the same experience for the users who are just using the stock cooler. Please if anyone could chime in I'd appreciate it thank you.
post #2848 of 7589
Quote:
Originally Posted by HyperMatrix View Post

Die size of GP100. Just as Nvidia did with the Titan X over the OG Titan by dropping fp64.

Current Titan X (Pascal) already has 3840 cores, just not all are active. 4 SMs have been disabled due to yield problems. As I've said in the same post, with a die size of GP100 we'd get something like 5760 cores, which is a full 60 SM chip, a fully enabled GP100 with FP64 replaced by FP32.

Unless they make a better yielded refresh, we won't see anything new soon.

That kind of chip would be 60% faster than Titan X (Pascal) and 125% faster than 1080. I presume they'll again have yield problems and will not make current cards obsolete that fast - probably not before summer 2017 or later.
post #2849 of 7589
Quote:
Originally Posted by toncij View Post

Current Titan X (Pascal) already has 3840 cores, just not all are active. 4 SMs have been disabled due to yield problems. As I've said in the same post, with a die size of GP100 we'd get something like 5760 cores, which is a full 60 SM chip, a fully enabled GP100 with FP64 replaced by FP32.

Unless they make a better yielded refresh, we won't see anything new soon.

That kind of chip would be 60% faster than Titan X (Pascal) and 125% faster than 1080. I presume they'll again have yield problems and will not make current cards obsolete that fast - probably not before summer 2017 or later.

All I ever said was that I wish the Pascal Titan X was a full sized 610mm2 die without fp64, because 2 of the current card just aren't enough for maxed out gaming.
post #2850 of 7589
Quote:
Originally Posted by atreides View Post

hi everyone,

I'm not particularly tech savvy but my Titan XP seems to be generating massive heat. I know it's a good thing the heat is being exhausted but I would like to know if it's the same experience for the users who are just using the stock cooler. Please if anyone could chime in I'd appreciate it thank you.

Skip to about 10 mins of this guys review:
http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-Titan-X-Pascal-12GB-Graphics-Card-Review

He's using a single card and the stock cooler of course. Yes it runs hot under load. 85 degrees C on his.

Get some temp monitoring software like AIDA 64 or just use EVGA precision X (it's free) and enable the OSD and select temp monitoring in the options (be warned the precision X interface is not very easy to figure it the first time you use it).
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › [Official] NVIDIA Titan X Pascal Owners Thread