GTX670 Upgrade and Overclocking Review (vs. 5850 crossfire)

[Termie]

Do me a favor when you retest. Just play Caspian multi on a 64 player map, test for 300 seconds and just play normal. I want to compare against my 1300 core 7850

Testing for just 60 seconds just doesn't give an accurate picture of gameplay.

Here ya go:

What do ya got for me?

By the way, this is not my preferred way to benchmark - dropping into the spawn screen messes things up and gives me very low minimums.

 

[Ieat]

Was that the max stable overclock you could get? 1058+125= 1183? Seems a bit low.

 

[Gepidae]

Great post, makes me think twice about buying my gtx670

 

[BallaTheFeared]

Don't really surprise me, though it does surprise me a bit that OC vs OC 5850s are that far behind 470 SLI.

 

[Termie]

Was that the max stable overclock you could get? 1058+125= 1183? Seems a bit low.

That's all I could get. Ran 3dMark11 at +150, but crashed in Heaven. Wasn't temperature-related, because it would crash at 64C.

 

[BallaTheFeared]

What did you OC the memory to?

 

[ZimZum]

Nice work

 

[BD231]

I get over 70 FPS average in that same BF3 run with an oc'd gtx 480 at 1080p and my 3D mark score is only 6700 ... what gives?

 

[Jaydip]

OP from all the benchmarks that i have seen on the web it seems that Kepler is greatly affected my memory overclocking.Increasing the clocks seems to have less of a impact on Kepler.

 

[Termie]

I get over 70 FPS average in that same BF3 run with an oc'd gtx 480 at 1080p and my 3D mark score is only 6700 ... what gives?

Couldn't really tell you, other than that there must be a difference in the way we're running the map. Also, you're at slightly lower resolution - but 70fps on all ultra honestly sounds pretty unlikely on a GTX480. Did you FRAPS it?

What did you OC the memory to?

OP from all the benchmarks that i have seen on the web it seems that Kepler is greatly affected my memory overclocking.Increasing the clocks seems to have less of a impact on Kepler.

I think you guys are on to something. I only clocked up the memory to 6400 from 6000. My core clock isn't giving me much in the way of scaling. I have to work on that memory overclock, but it's tricky, because you need to watch for memory correction kicking in and lowering the performance.

Alas, the only game I really care about is BF3, and I'm going to turn on nVidia's Adaptive vsync to lock in smooth gameplay around 60fps. Now if only I could figure out the driver glitches in that game...

 

[Jaydip]

OK so u need to find a sweet spot

 

[BD231]

Couldn't really tell you, other than that there must be a difference in the way we're running the map. Also, you're at slightly lower resolution - but 70fps on all ultra honestly sounds pretty unlikely on a GTX480. Did you FRAPS it?

HBAO was off my bad, I'm at 59 average with in on.

 

[aaksheytalwar]

Still too high for a 480 Oced. 480 Oced can't compete with 670 Oced or even 5850 cf Oced.

 

[Termie]

OK so u need to find a sweet spot

Here you go:

I'm going to settle on +135 offset, +300 on the memory, which gives me about 1194 core boost, 6600 memory.

 

[BallaTheFeared]

I got about 80+fps on avg during the middle part of swordbreaker in SLI, while recording with fraps - with max settings and 4xAA.

http://www.youtube.com/watch?v=iBxwG5uhKwc

 

[Borealis7]

Tremie, your upgrade has actually given me some good info
now i know a GTX670 will roughly be twice as fast as my current single 5850! thanks!

 

[Termie]

Tremie, your upgrade has actually given me some good info
now i know a GTX670 will roughly be twice as fast as my current single 5850! thanks!

That's exactly right - the 670 is just about twice as fast as the 5850, and at least in the benchmarking apps, is able to beat 5850 crossfire, which has imperfect scaling.

 

[BD231]

Still too high for a 480 Oced. 480 Oced can't compete with 670 Oced or even 5850 cf Oced.

You'll get 50 to 60 fps average on ultra with an oc'd 480, ran it three times and hit 59/58/59. Turn off HBAO and you're into the 70's :/. I expected more from the 670.

 

[Ieat]

Your Heaven benchmark seems low. I get about 18% more frames at 1293mhz and 6600 mem. 18% more frames from a 9% higher overclock doesn't seem right. Maybe a cpu bottleneck?

 

[Jaydip]

Here you go:

I'm going to settle on +135 offset, +300 on the memory, which gives me about 1194 core boost, 6600 memory.

Thats great :thumbsup:

 

[f1sherman]

how about subjective feeling of gameplay 670 vs 5850 CF, i.e. smoothness, responsiveness?

 

[T_Yamamoto]

Your Heaven benchmark seems low. I get about 18% more frames at 1293mhz and 6600 mem. 18% more frames from a 9% higher overclock doesn't seem right. Maybe a cpu bottleneck?

Yeah. I sense a bottleneck

 

[Silverforce11]

Metro is not an AMD favored game btw, it's a TWIMTBP title.

Surprisingly good results for the old boys.. 5850s, one of the best gpu ever. You should however notice better gameplay/smoothness with the 670 in starcraft 2 ultra and shogun 2, i noticed a big difference when i ditched the cf setup for this single card in those games.

 

[Termie]

Your Heaven benchmark seems low. I get about 18% more frames at 1293mhz and 6600 mem. 18% more frames from a 9% higher overclock doesn't seem right. Maybe a cpu bottleneck?

Yeah. I sense a bottleneck

Wrong guys. It's the resolution. 1080p is exactly 11% lower total pixels than 1200p.

Your extra boost makes up the rest. Heaven is completely unaffected by CPU - it is the ultimate GPU benchmark. I've tested it many times and found it to be perfectly comparable across CPUs.

 

[blackened23]

Metro is not an AMD favored game btw, it's a TWIMTBP title.

Yeah, I did a double take when I saw that. It was considered nvidia favored for a long time because the cayman series had pretty poor tessellation performance, but the tahitis are really strong in metro (and have dramatically increased tessellation performance in general)