Originally Posted by Decoman
Please forgive me for not having read all the 8.000+ posts in this thread.
I recently placed an order for a Sapphire 290X card w. AMD stock cooler/fan to be used in a future water cooling setup and I will be buying parts occasionally to set it all up some months later:Sapphire Radeon R9 290X 4GB DDR5 2xDVI HDMI DisplayPort PCIe BF4 - (21226-00-53G) (1000 MHz core clock)
I have a few questions about the card I bought and hope someone can give me some feedback on that. I normally ask around before I buy stuff, but I had been looking for a reference card and saw one at a discount and decided to buy it. Hopefully it will not be impossible to buy another reference 290X card later on. At worst I guess I could try buy a used one.
Q1: Is there perhaps anything I should know about the 21226-00-53G model? The 21226-00-40G seem similar to the version I have.
Q2: How much of an issue is this problem of getting the "black screen" all these months after the product launch in 2013?
Q3: What is safe max voltage for a 290X card?
Q4: Is there any point to flashing to a different bios or modding the bios if I only want to achieve a modest overclock? (20-25%)
1. No clue about the versions, I know that the Tri X, there was an earlier and later model. Biggest difference was that the late model couldn't use the earlier model's BIOS due to changes in VRAM timings.
2. The black screen issue is most commonly associated with bad memory. The prevailing theory at the moment was that the earlier batches of cards had derped Elpida timings resulting in blackscreens at stock. Its pretty rare nowadays (though I daresay its more common with Elpida equipped cards than Hynix), you tend to see blackscreens more now due to unstable VRAM overclocks.
3. There is no real definite answer but I can share a rough guideline: very very rough, if anyone can help me clean it up, it would be much appreciated. Its not really voltage that kills components, its the power draw and heat, voltage alone is just an electrical potential.
The VRM MOSFETS are rated for up to 50A on the large ones and 20A on the smaller auxiliary ones, I am basing as rough estimate (I have limited knowledge on the intricacies of VRM design) that with 5 phases, it is reasonable to assume at a Vcore of 1.25V (stock worst binned GPU) and about 1.18V (VDROOP), 50x1.18x5 =295W of VRM designed power. The TDP is 250W ish with provision for 375W ish(50% power limit). At 375W, you would need roughly (very rough mind you, power consumption is quadratically proportional to voltage and linearly proportional to clockspeed) 1.5V. So with this information, I can surmise that the "safe" 24/7 voltage is leaning towards maybe the 1.35V range (which is about +125mV on a decent binned GPU) on air cooling
where you are roughly still inside the AMD mandated maximum PCB power draw.
However, because the 290X VRM assembly is designed for optimal operation in the 250W zone, VRM temperatures rapidly spike at loads above this. Most would not attempt large overvoltages without watercooling since the act of having cooler temperatures reduces the power draw of the GPU at a given clockspeed, this will allow higher voltages, hence clockspeed, before tripping past the 375W danger zone. Plus, WC also deals with the high heat density of the VRMs beautifully.
4. 20-25% is a pretty large overclock for Hawaii. Most GPUs (barring the worst binned <70% ASIC ones) can seem to reach about 1080-1100mhz on stock voltage (around 1.212V, mine actually need +38mV to do 1100mhz but I think its because mine started off with a low stock voltage with not much headroom) 1150mhz requires at least +125mV (at least for my 2 cards with 78.9 and 79.9 ASIC quality) plus my 1200mhz clock is not stable even at a blistering +175mV (perhaps +200mV would do it). You may get better results under water, notably because your ASIC will draw less power overall thus having more overvoltage headroom (before the VRMs are running too much out of spec).Edited by Dasboogieman - 6/29/14 at 9:22am