We need a new thread for Overclocking for CUDA 4.2

If you have a successful overclock, please post your settings. Please include the card (GTX 570, 580, 670, 680), voltage, Core Clock and Memory.

---

Thx - Paul

Note: Please don't use driver version 295 or 296! Recommended versions are 266 - 285.

Zotac AMP GTX470 (dual fan design) Ref. clocks 607MHz, FOC 656MHz.
2003R2 server, 301.42 driver, Boinc 7.0.28, i7-2600 stock (6 threads used for CPU tasks):

GPU@707MHz, Voltage 1.025 (unchanged)

Performance:
GPU at 99% utilization running a MJHarvey task (the task type also matters, a lot), 566MB GDDR5 used.
Fan at 60%, temps at 59°C
Entire System's power usage, 327W (Corsair 650W PSU).

- was not stable, some MJHarvey tasks completed but the rest failed. This was totally stable for the 3.1app. Turned it down to 667MHz...

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

EVGA GTX 480, with backplate, using MSI Afterburner, voltage 1013, core 801, shader 1602, memory. 1848

Used about 1 GB memory while running, fan & 80%, temp. 75 c., 98 to 99% GPU usage

Windows 7, Seasonic X650, i3 530, 4 GB ram., 3 case fans + 1 PCI card fan( AzenX Blitztorm has 2 fans, 1 exhaust plus 1 blower)

Workunit 3521495
Created 25 Jun 2012 | 13:55:15 UTC
Sent 25 Jun 2012 | 16:06:15 UTC
Received 26 Jun 2012 | 5:08:08 UTC
Server state Over
Outcome Success
Client state Done
Exit status 0 (0x0)
Computer ID 101894
Report deadline 30 Jun 2012 | 16:06:15 UTC
Run time 17,867.75
CPU time 1,226.26
Validate state Valid
Credit 60,900.00
Application version Long runs (8-12 hours on fastest card) v6.16 (cuda42)
Stderr output

<core_client_version>6.12.34</core_client_version>
<![CDATA[
<stderr_txt>
MDIO: cannot open file "restart.coor"
# Time per step (avg over 1430000 steps): 11.898 ms
# Approximate elapsed time for entire WU: 17846.612 s
called boinc_finish

------------------------------------
GTX 670 (Gigabyte Windforce)
GPU = 1249
MEM = 3206
MV = 1175
DRV = 301.42
OS = Win7 x64
BNC = 7.0.25
------------------------------------
GTX 480 (EVGA vanilla)
GPU = 1620
MEM = 1848
MV = 1125
DRV = 301.42
OS = WinXP 32
BNC = 7.0.25
------------------------------------
GTX 295(EVAG vanilla dual card)
GPU = 1512
MEM = 1141
MV = 1125
DRV = 301.42
OS = WinXP 32
BNC = 7.0.25
------------------------------------

---

Thanks - Steve

MSI N560TX-Ti 2GB 900/1800/2004 @1.012
60c at 55% fan (20c ambient)
Win7 SP1 Nvidia Version 301.42

EVGA GTX 580 & Windows 7
Voltage 1.138V
Core Clock 948
Memory 2000

EVGA GTX 570
Voltage 1.1V
Core Clock 907
Memory 1900

GTX 570 (non-reference)
Voltage 1.075V
Core 851
Memory 1900

EVGA GTX 590
Voltage 0.925V
Core Clock 628
Memory 1709

Everything is a little slower than CUDA 3.1 but stats are much improved. I finally broke the 1,000,000 RAC.

---

Thx - Paul

Note: Please don't use driver version 295 or 296! Recommended versions are 266 - 285.

"Everything is a little slower than CUDA 3.1". Do you mean system responsiveness?
On your systems 4.2 should be much faster than 3.1.

You broke the 1M RAC by a fair bit too!

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

"Everything is a little slower than CUDA 3.1" means that all of my clock speeds were reduced from the CUDA 3.1 levels. I had to back everything down about 20 MHz.

The new CUDA 4.2 tasks are not intrusive. The computers continue to work fine with GPUGrid in the background.

---

Thx - Paul

Note: Please don't use driver version 295 or 296! Recommended versions are 266 - 285.

MSI GTX 660Ti PE OC
GPU = 1120 (boost to 1280 during WU)
MEM = 1550 (6200 GDDR5)
V = 1.175
DRV = 305.68
OS = Win7x64 HP
BNC = 7.0.28 (x64)

MSI GTX 660Ti PE OC
GPU = 1120 (boost to 1280 during WU)
MEM = 1550 (6200 GDDR5)
V = 1.175
DRV = 305.68
OS = Win7x64 HP
BNC = 7.0.28 (x64)

I don't own any 660's and didn't know they had an open ended GPU boost like the 670's. If that's an open ended boost, you won the lottery. I have 1 670 that hits 1202 and the second has never gone over 1184. EVGA came out with a new BIOS for the 670's and they don't seem to clock so high, which is on the second card.

---

I don't own any 660's and didn't know they had an open ended GPU boost like the 670's. If that's an open ended boost, you won the lottery. I have 1 670 that hits 1202 and the second has never gone over 1184. EVGA came out with a new BIOS for the 670's and they don't seem to clock so high, which is on the second card.

It seems so. Actually I also haven't seen any other sample that clocks this high on air with stock cooler. On default voltage, it clocks to 1140MHz with Boost to 1300MHz, and with +100mV it goes stable to 1200MHz with boost up to 1380MHz!!!

You can see in attached screenshot with boost (actual clock during load) of 1377MHz in Unigine Heaven benchmark, which is known for being very sensitive for unstable overclock.

But of course, for 24/7 crunching both performance/watt and 100% stability is important, so I am not pushing any extra voltage nor pushing that "last MHz" of stability, hence the 1120MHz clock setting (with boost to 1280MHz). Also, it is virtually noiseless on this setting.

If any one has GTX295's overclocked, please post your settings here also. Thanks in advance, Rick

---

I used to run my GTX 295 (original 2 card EVGA) @1.050 v, 1512 shaders, 100% fan speed.
By 'used to' I mean 6 weeks ago.

---

Thanks - Steve

Thanks for your post Snow Crash. Did those settings work with CUDA 4.2? Rick

---

Yes, that worked fine for 4.2 but I was doing my best to crunch 3.1 (reverse retvari zoltans 4.2 force methodology) because it performed better. If you are in USA and are interested in obtaining my card for cheap send me a PM.

---

Thanks - Steve