MSI R5670 Cyclone 1G Graphics Card
Author: Zahn Funk
Editor: Shawn Knight
Date: 04-28-2010
Provided by: MSI

The MSI R5670 Cyclone 1G is installed in the following system:

AMD Phenom II 965 X4 BE
MSI 890GXM-G65
2x2Gb Crucial Ballistix Tracer DDR3-1600
Swiftech H2O-220 Compact
2x74Gb WD Raptor RAID0
Optiarc 24x DVD+/-RW
SilverStone Decathlon DA750
Cooler Master HAF 932 AMD Edition
Windows 7 64bit Home Premium

The R5670 easily fits in this motherboard and case with plenty of room to spare, replacing my usual 4890 card. The oversize Cyclone cooler may cause some interference issues if using a Crossfire configuration, particularly on boards with closely spaced PCI-E slots like the 890GXM-G65. However individually it is like any other mid-range card in size and the double riser bracket lends support for the extra weight of the cooler.

Rather than use the version on CD, the latest Catalyst 10.3 driver was downloaded and installed. The most recent version of GPUZ 0.4.2 identifies the 5670 specifications correctly with the exception of ROPs, which are listed as double of actual. The ATI OverDrive portion of CCC shows the fan running at only 28% with an extremely impressive 28 C at idle. I've never seen a GPU temperature that low with a factory cooler. Overclocking can be accomplished through OverDrive in the Catalyst Control Center, the software-limited maximum clocks of 850/1050 are easily reached and pass the built-in stability test. Overclocking through the CCC still allows the power save feature of the card to kick in when in 2D mode, however you're stuck with the low 850/1050 ceiling.

MSI includes their own overclocking utility Afterburner which was co-developed with the makers of RivaTuner, a GPU overclocking tool that has been around for many years. By default Afterburner is limited to the same clocks available in the CCC, however you can open the MSIAfterburner.cfg in the program directory and change the value of EnableUnofficialOverclocking from 0 to 1. I was able to get up to 900/1175 stable using the modified Afterburner config. Doing this does disable the power save feature of the card, so it will never reduce clocks and voltages in 2D mode. To get around it you can save an overclocked profile in Afterburner, and manually switch from default to the overclock profile only when gaming or benching. MSI Afterburner also lets you adjust the fan speed manually. Increasing this to 100% results in a 15 C drop in load temp. Of course this does generate a good bit of noise but it's pretty incredible that a factory cooler is capable of an overclocked load temp of only 40 C. Even with the voltage settings enabled however I was unable to either get a reading or make any change to the level of the card.

As mentioned, at stock clocks of 775/1010 the 5670 idled at an impressive 28 C and only got up to 52 C under load, even with the fan left on the Auto speed setting. When overclocked to 900/1175 the load temp only increased by 3 C however the idle temp jumped 8 C due to the power save not coming on automatically. When left on Auto the fan never moved from 28% even at full load and temp. I realize the 5670 is only a mid-range graphics card however these are still some very impressive temps especially considering how quietly the stock cooler operates.

The 5670 is very miserly when it comes to power usage as well. According to my Kill-A-Watt meter, compared to the onboard 4290 graphics of the 890GX chipset the 5670 used only 14W more in 2D idle mode, and 67W when under full load. The 14W/67W I observed is very close to AMD's claimed 15W/64W power rating for the 5670. Overclocking the card uses a bit more power, 17W/8W more than stock. Just as we saw with the temperatures, the larger idle increase is due to the power saving 2D mode not kicking in when overclocked with the MSI Afterburner utility. By these calculations, when overclocked the 5670 is right around 75W, which is the maximum rated power for a PCI-E slot. Stepping up to a 5750 will require a 6-pin external PCI-E power connection to support the claimed 86W card.

Now let's try some benchmarks.

  Sponsors Links

  Sponsors Links