Skip to main content Start reading the article

GeForce GTX 570. Is memory overclocking necessary?

The influence of memory clock rates to performance of graphics card with GPU overclocking

Testbed configuration

MotherboardDFI LP UT X58-T3eH8
ProcessorCore i7 920
MemoryG.SKILL Trident 6GB F3-16000CL9T-6GBTD
Hard drives2 x VelociRaptor WD1500HLFS RAID0
CaseOpen benchtable Banchetto 101
OSWindows 7 64-bit Service Pack 1
VGA driver versionGeForce/ION Driver 266.58
AMD Catalyst 11.1a (Driver Packaging 8.82.2-110119a-112250E-ATI dated 26.01.2011)
MonitorDell Ultrasharp U2711

System configuration:

System config

CPU and motherboard are watercooled.

Overclock

Default VID value for the tested card was 0.95 V. Preliminary tests with stock cooling system revealed the following overclocking results:

Vgpu (monitoring)Fan, %Fan, RPMTgpu
EVGA GeForce GTX 570 732/950 MHz0.950 V64%2370 RPM86C
EVGA GeForce GTX 570 @800/950 MHz0.950 V65%2430 RPM86C
EVGA GeForce GTX 570 @875/950 MHz1.000 V67%2700 RPM87C

Further tests were conducted by adding a universal VGA waterblock Swiftech MSW60 to stock heatsink that cools PCB components. Waterblock installation details are the same as in case of GeForce GTX 470 graphic card, therefore, I see no need to describe the mounting process again.

Watercooled GPU, gave us some more overclocking headroom:

Vgpu (monitoring)Fan, %Fan, RPMTgpu
EVGA GeForce GTX 570 @910/950 MHz1.050 V45C
EVGA GeForce GTX 570 @970/950 MHz1.100 V49C

Further overclocking came harder and card operation was, certainly, not stable enough. To reach 1000 MHz rate for GPU, I had to increase voltage to 1.138 V. For this matter, I also had to unlock maximum voltage limits in BIOS. Those who are interested, read the following article. With card clocked at 1000/950 MHz I managed to pass some benchmarks. The card got 27719 points in 3DMark Vantage Performance profile, and took negligible place at hwbot.org due to having “weak”, according to their standards, processor.

Since 3DMark 11 results has less dependence on CPU performance, our 6784 points with Performance settings allowed the card to almost reach the pedestal. Well, there were nothing more to count on at hwbot.org ratings with a quad-core CPU clocked at 4500 MHz, so we put extreme benching aside and concentrated on tests with more “natural” settings.

To find out how overclocked memory effects overal performance, we set the clock for GPU at 900 MHz with Vgpu = 1.050 V. GTX 570 default memory rate — 950 MHz. There were no problems in passing all test with these settings:

GTX 570 @900/950МГц

Next, we set 900 MHz for GPU and 1050 MHz for memory. However, this required to raise GPU voltage to1.063 V:

GTX 570 @900/1050 MHz

Then we pushed overclocking further till 1150 MHz for memory, GPU was satisfied with same 1.063 V:

GTX 570 @900/1150 MHz

Memory overclock to 1235 MHz did not affect other settings:

GTX 570 @900/1235 MHz

Leaping ahead we have to say that graphic card test results with the settings described above looked kinda “stiffly” — GPU overclocking results were missing. For 1050 MHz memory and Vgpu=1.063 V the graphics processor can reach 950 MHz:

GTX 570@950/1050 MHz

Yet, we set 1250 MHz in opposition to 1050 MHz clocked memory:

GeForce GTX 570 @950/1250

And here are our voltage measurements in case someone is of interested in it:

Vgpu, monitoring0.9500.9630.9750.9881.0001.0131.0251.0381.0501.0751.0881.1001.1131.1251.138
Vgpu, measure0.9600.9720.9850.9981.0101.0241.0361.0501.0631.0881.1011.1131.1261.1391.152
Vmemory1.570

Now we can turn to performance results we gathered. Just for information purposes only, we added test results of RADEON HD 6950 2GB graphic card unlocked to get RADEON HD 6970 performance config.

No comments

Leave a Reply

Avatar