Friday, June 6, 2014

R9 290X Windforce 450W OC Review

This review will not cover the FPS this card gives because every review ever made does that so you can go read their results.
I will also not give you temp and power figures because I do not have an AC and use a test bench instead of a case so my temps are lower than in a case and I do not have the equipment required to measure the power draw on the 12V rail.

Cooling Analysis:
The windforce 450W cooler uses 6 heatpipes that are connect to the core though a bit of copper that sticks out of the base of the cooler. The cooler use 3 low profile 74mm fans capable of 4500RPM at maximum speed at which point they are very very loud. The fans are angled to blow air away from the motherboard. The fin stack directly above the core has a trapezoid in the middle that forces hot air to come out of the long sides of the PCB. The memory chips dump their heat into the main fin by use of thermal pads. The VRM dumps it's heat directly into a couple of fins in the secondary fin stack. This cooling is sufficient for gaming and low voltage overclocking(up to +75mv on the core) however Furmark or high voltages will very quickly get the VRM to over 90+C even on my open air test bench. It is also possible that the VRM dumping it's heat into the secondary fin finstack and the heatpipes is lowering the coolers cooling capability. I would much rather have seen a separate VRM heatsink instead of using the core's heat sink space for cooling the VRM.
Overall this a good heatsink for the average gamer but doesn't quite cut it for high voltage (+150mv or more) overclocking. So the heat sink gets a 7/10.

I might be able to produce a guide on how to upgrade this heat sink. I plan to mess with replacing the fans and getting the VRM it's own cooler.

VRM Analysis:
-Phases: 5+1+1 phases from an IR 3567B PWM controller.
-Maximum core current out: 350A @ 125C° which is inline with the AMD reference design
-Core voltage MOSFETs: 70A made by IR
-Mem voltage MOSFET: 70A made by IR
-Core voltage chokes: same length but wider and shorter than reference 150nH Magics that are most likely 70A capable (can't get a spec sheet for them sorry)
-Mem voltage choke: 150nH Magic same as the core
-Aux voltage choke: 2 low current 220nH Magics in parallel
-Core voltage filtering: 7520µF distributed over 16 tantalum capacitors (14% improvement over reference)
-Core voltage filtering: 1880µF distributed over 4 tantalum capacitors (14% improvement over reference)
-Aux voltage filtering: 1880µF distributed over 4 tantalum capacitors (42% improvement over reference)

VRM Verdict:
Core: Epic current capability here a full 87% more current capability than the core so 4.5/5 for that but the inductance on the inductors is really low so that's a 3.5/5 and the phase count is mediocre at only 5 controlled and physical phases so that's a 4/7 which is not a bad score but leaves something to be desired because the reference HD7970 VRM came with 220nH chokes and the option to be upgraded to 360A and 6 phases which most manufacturers did.
Memory: 4GB of GDDR5 is power hungry so it should be no surprise that the 1 phases VRM is capable of pushing 70A to it. But like with the core VRM low phase count means a 1/3 for phase count ripple suppression, a 3.5/5 for inductance and a 3/4 for current.
Aux: It's 40% better than reference so I'll give it a 5/5 also this is not a very important VRM.

Summary: There's enough power here that it's more likely the core will die before the VRM does. Unfortunately what this VRM has in current capability it lacks in ripple suppression and accuracy. If you plan to do any extreme OCing on this VRM I recommend you add an extra 2.5V 1500µF capacitor to every single one of the tantalum caps on the back of the PCB. This VRM is definitely holding the core back but at least it won't burn up.

Tuning Ability:
I have a few issues with the new voltage controller that AMD is using and those are:
Locked memory voltage control
Automatic voltage control based on temperature that you can not turn off without BIOS flashing the card.
To OC this card I used Sapphire Trixx 4.8.2
I managed to finish the Unigine benchmark at as high as 1210/1575 using +200mv with extreme visual artifacts through out the benchmark. I run the GPU at 1100/1496 on +31mv
for gaming. With my maximum artifact free overclock being 1150/1500 on +100mv.
The memory clock is bound to the core clock by the following ratio: memory clock / core clock. This ratio is typically between 1.25 and 1.4 and requires that you find it your self.
For tuning ability I give an 8/10. 1 point off for the temperature control and 1 point off for lack of memory voltage control.

No comments:

Post a Comment