GTX_1080 benchmark

[Timmaigh!]

Repost from anandtech forums, test performed/screen posted by user named Dufus...for anyone wondering how far u can get with 1080 performance wise:

[anklove]

Anklove and Uncia, could you let me know the temperatures at the end of the benchmarks for a single 1070 and a single 1080 at stock clocks?

Thanks in advance!

About 67/65 C

[arashiii]

Here are the results for the 1070 stock:

1 x 1070 > 117.61
4 x 1070 > 482.43

[mbutler2]

So if it is the case that two GTX 1070s bench at 241, then one would bench at 120 (my personal tests show linear, so I'm assuming.)

And all of the GTX 980 TI's I have bench at around 125 each.

Then there's absolutely no reason to buy any more expensive 980 TI's.

Not to mention the power and heat improvements over the hot and hungry 980 TI's.

And improvement from 1070 to 1080 has a bigger price increase than performance increase.

So if you have room for a bunch of cards, it seems like 1070s is the way to go. Because cost per speed is more expensive in a 1080.

Any opinions, guys?

[cayn99]

So if it is the case that two GTX 1070s bench at 241, then one would bench at 120 (my personal tests show linear, so I'm assuming.)

And all of the GTX 980 TI's I have bench at around 125 each.

Then there's absolutely no reason to buy any more expensive 980 TI's.

Not to mention the power and heat improvements over the hot and hungry 980 TI's.

And improvement from 1070 to 1080 has a bigger price increase than performance increase.

So if you have room for a bunch of cards, it seems like 1070s is the way to go. Because cost per speed is more expensive in a 1080.

Any opinions, guys?

so far i think you are right, but maybe we should wait for proper optimization of the 1xxx series; i think those numbers are not final.

Cheers~!

[arashiii]

@mbutler2

My idea was to have a quad gpu system which doesnt need liquid cooling. 4 of 980 ti cards squeezed together become too hot when just air cooled and in consequence start underclocking. so the full potential can only be used when there is a proper cooling installed.

my tests showed that after 5 min of excessive rendering the 1070s in the middle reach the temperature limit of about 83 degree C and start underclocking. with 1080s this effect should appear even quicker. so i needed to change the factory fan speed settings to higher rpms in hot conditions. the noise is still ok and they all run with 100% speed.

lower power consumption and the better price/performance tag (compared to 1080 and above) and more ram (compared to 980 ti) were the main reasons for me to go with the 1070s.

[darkline]

Not so sure. I recently bought a 980ti and paid less for it than a gtx 1070. It benchmarks at 141 (evga FTW edition).

Have they released the optimisation for Pascal yet? That could push 1070 beyond a 980ti?

[coilbook]

140 for 1080 and only 160 for titan x pascal. Why is titan so slow?

[Timmaigh!]

140 for 1080 and only 160 for titan x pascal. Why is titan so slow?

Partially cause of its lower clocks compared to 1080 and then, the performance increase does not correlate with the number of CUDA cores. GTX 1060 at half the cores of 1080 gets the score of 90, so 1080 should be 180 instead of 140 too.
Hopefully this will be resolved with the CUDA 8 Pascal optimized versions.

[coilbook]

140 for 1080 and only 160 for titan x pascal. Why is titan so slow?

Partially cause of its lower clocks compared to 1080 and then, the performance increase does not correlate with the number of CUDA cores. GTX 1060 at half the cores of 1080 gets the score of 90, so 1080 should be 180 instead of 140 too.
Hopefully this will be resolved with the CUDA 8 Pascal optimized versions.

i see Thank you