Gigabyte Unveils SuperOverclock VRM Boards for Graphics Cards

[btarunr]

Gigabyte showed us its new range of auxiliary VRM boards for graphics cards, designed for professional overclockers. The board has to be manually soldered onto your graphics card, it supplements your card's GPU power domains. The card features a high-current power stage, which draws power from five 8-pin PCIe power connectors. The power stage is actively cooled by a fan-heatsink. While your GPU may never need that much power, it makes near darn sure that it doesn't suffer from voltage droop, which destabilizes your record-seeking OC. Such devices have been attempted by other VGA makers in the past, famously, the EPower Board by EVGA. The company also showed off its first liquid-nitrogen evaporators for GPUs.



View at TechPowerUp Main Site

 

[Assimilator]

750W for a graphics card? Not even the 295X2 needs that much power...

 

[buildzoid]

750W for a graphics card? Not even the 295X2 needs that much power...

at 1.5V an R9 295X2 would need quite a bit more than 750W so you would still need 2 of these for an R9 295X2. Really this is great for people who want to Vmod the GTX 590 since just one of these should be able to feed it upto 1.3/1.4V.
I wonder where I could buy one of these.

 

[FireKillerGR]

It isnt about the NEED of 750W its about being ABLE to offer that W on hi-end cards in case of need.
Reference cards like 780, 780 ti, Titan, Titan Black, 290x, etc. have bad vrms, so that way you get higher clocks with voltage above 1.4-1.5v.

 

[buildzoid]

It isnt about the NEED of 750W its about being ABLE to offer that W on hi-end cards in case of need.
Reference cards like 780, 780 ti, Titan, Titan Black, 290x, etc. have bad vrms, so that way you get higher clocks with voltage above 1.4-1.5v.

The 290X VRM might not be accurate but it isn't bad. At least it doesn't blow up like the some Nvidia VRMs. I do wish AMD had used more phases(not for more power but for ripple suppression) or at least higher inductance(again to suppressing ripple) chokes instead of using the new controller to reduced power draw. If I had more money I would try adding more capacitors or replacing the chokes on my R9 290X because honestly the VRM has enough current output it's just achieved by using inductors that are too damn weak to get good ripple suppression.
Just for comparison the R9 290X uses 150nH chokes across five phases when most other high end cards use 220nH chokes on six phases or more phases. The HD 7970 used 220nH 60A chokes on 6 phases and some HD7970 records are still held by the reference cards even after the MATRIX and Lightning came out. So really if AMD had just used bigger chokes the R9 290X VRM would have been OK. I even found chokes that have lower electrical resistance push the same current and have higher inductance but their just too big compared to the reference R9 290X chokes and would have required a longer or wider PCB.

 

[FireKillerGR]

The 290X VRM might not be accurate but it isn't bad. At least it doesn't blow up like the some Nvidia VRMs. I do wish AMD had used more phases(not for more power but for ripple suppression) or at least higher inductance(again to suppressing ripple) chokes instead of using the new controller to reduced power draw. If I had more money I would try adding more capacitors or replacing the chokes on my R9 290X because honestly the VRM has enough current output it's just achieved by using inductors that are too damn weak to get good ripple suppression.
Just for comparison the R9 290X uses 150nH chokes across five phases when most other high end cards use 220nH chokes on six phases or more phases. The HD 7970 used 220nH 60A chokes on 6 phases and some HD7970 records are still held by the reference cards even after the MATRIX and Lightning came out. So really if AMD had just used bigger chokes the R9 290X VRM would have been OK. I even found chokes that have lower electrical resistance push the same current and have higher inductance but their just too big compared to the reference R9 290X chokes and would have required a longer or wider PCB.

Good point but the main issue is still the VRM, I dont think they can survive more than 1.55 V because even if they can, their freq will go down. I ran 290x @1.55v and high clocks, mentioning that because its different for the VRMs to run a card @Default clocks and with 1.5v than 1500 MHz and 1.5v; but I bet you know that already.

Now to support your point of view, yeah AMD's VRMs are better because many Titan, 780 Ti and cards with similar PCB have died with voltage <1.5 and in some cases with lower than 1.4 V as well. I never saw a card dying by inductors but mostly cause of cheap/no hi-end mosfets and thats because IC, handles (values and everything) the Vdroop and voltages but mosfets do the real job, thats why mosfets get hot.

Last but not least, you couldnt make a 290x work with Evga Epower because you had to surpass the ΙΩR IC. That IC cuts the rest voltages of the card (I dont know if all or just VCCIO) in case you solder epower. GB did a nice trick to avoid that but I cant get into any more details because I dont know if they want me to.

 

[buildzoid]

I'm writing a research paper on semiconductor power scaling so I'm well aware that 50% frequency nets a 40% power draw increase before you increase voltage. But the loss in frequency past 1.55v is caused by the caps and inductors losing all the energy they store too quickly meaning that ripple goes completely insane. Which is why I would first try doubling or trippling the VRMs output capacity before trying to replace the whole VRM.
Also drawing too much current through an inductor causes it's inductance to drop massively causing v ripple t get so big that whatever it's feeding will error the good news is that a reboot is eenough to fix that most of the time.

Sorry for spelling I'm on a phone