PowerColor Radeon RX Vega Red Devil Teased

[btarunr]

PowerColor teased its first Radeon RX Vega series graphics card, under its flagship Red Devil branding. The company could use this common board design for both RX Vega 64 and RX Vega 56 SKUs. It combines a custom-design, triple-slot, triple-fan cooling solution, with a semi-custom design PCB. The cooler is an upscale of the company's RX 580 Red Devil, with a massive 2.5-slot thick aluminum fin-stack heatsink, which draws heat from the "Vega 10" GPU and a base-plate conveying heat from the VRM MOSFETs surrounding it, ventilated by a trio of 100 mm fans.

PowerColor's PCB for this card is a variation of AMD's reference "Vega 10" board, in that it's slightly taller towards the front. The company went with two each of DisplayPort and HDMI connectors, instead of the standard three DP + one HDMI layout. The VRM is the same 12-phase setup from the reference-design PCB, with PowerColor sticking to International Rectifier IR6894 and IR6211DirectFETs, and IR3598 phase-doublers, while customizing the chokes. The PCB is enlarged height-wise to give easy access to the BIOS switch. Two inverted 8-pin PCIe power inputs are located toward the end. Given how heavily this board is based on the reference PCB, it remains to be seen if fan idle-power off is featured. PowerColor is expected to launch the RX Vega 64 Red Devil and RX Vega 56 Red Devil before Holiday.



View at TechPowerUp Main Site

 

[punani]

MRW gaming with Powercolor Vega red devil

 

[the54thvoid]

MRW gaming with Powercolor Vega red devil

A tad unfair but I lolled.

 

[Durvelle27]

But when will they be available?

 

[mouacyk]

Lolling

 

[Valantar]

So the PCB looks like it extends pretty much the entire length of the cooler, yet the VRMs are still packed close to the GPU. Is the entire last third of the PCB just empty? If so, why even have it there at all?

 

[Divide Overflow]

So the PCB looks like it extends pretty much the entire length of the cooler, yet the VRMs are still packed close to the GPU. Is the entire last third of the PCB just empty? If so, why even have it there at all?

The PCB doesn't extend the length of the cooler. I think you're looking at the backplate.
Only 2 8 pin power inputs? Powercolor had a chance to raise the bar and beat Sapphire by putting 4 in.

 

[Valantar]

The PCB doesn't extend the length of the cooler. I think you're looking at the backplate.

Nope. That ~2cm gap really doesn't count. If they had made even the slightest effort, they could have fit some of the components forcing them to extend the PCB (such as the power plugs) to above the VRMs, cutting 4-5cm further off it. Lazy board design. Of course, it might be that the PCB below the power connectors is cut away for airflow - if so, good job! That would of course also work. We can't see anything like that from the pictures provided, though.

 

[FordGT90Concept]

All Vega cards have their size dictated by cooling solution. It's a 200-300w GPU so it needs at least that in cooling capacity. Can't fit that much cooling potential in 8 inches without making it super thick so they make the HSF full length like Fury did.

 

[Valantar]

All Vega cards have their size dictated by cooling solution. It's a 200-300w GPU so it needs at least that in cooling capacity. Can't fit that much cooling potential in 8 inches without making it super thick so they make the HSF full length like Fury did.

You misunderstand my point entirely. I'm talking about making the PCB longer than necessary, not the heatsink - in other words, I'm pointing out the stupidity of this not being "open-backed" like for example the Sapphire R9 Fury Nitro. Removing the PCB increases airflow past the heatsink, improving cooling. Try sticking a piece of cardboard behind your CPU cooler, and see how "well" it performs ...

 

[eidairaman1]

You misunderstand my point entirely. I'm talking about making the PCB longer than necessary, not the heatsink - in other words, I'm pointing out the stupidity of this not being "open-backed" like for example the Sapphire R9 Fury Nitro. Removing the PCB increases airflow past the heatsink, improving cooling. Try sticking a piece of cardboard behind your CPU cooler, and see how "well" it performs ...

The pcb disperses the heat too

The PCB doesn't extend the length of the cooler. I think you're looking at the backplate.

Only 2 8 pin power inputs? Powercolor had a chance to raise the bar and beat Sapphire by putting 4 in.

Yeah I saw that x 8 pin connections, is that for LN/LOX cooling??? Lol

 

[Valantar]

The pcb disperses the heat too

The heat dissipation through the copper in the PCB is marginal at best, and more than likely far less per unit of area than the increased airflow from removing the same area obstructing the fan. Removing the PCB lets the fan operate more efficiently, lowering turbulence drastically, avoiding recycling of hot air, reduces airflow resistance to pretty much zero, and makes airflow over the heatsink more linear (rather than turbulent and deflecting every which way to escape out the sides).

 

[eidairaman1]

The heat dissipation through the copper in the PCB is marginal at best, and more than likely far less per unit of area than the increased airflow from removing the same area obstructing the fan. Removing the PCB lets the fan operate more efficiently, lowering turbulence drastically, avoiding recycling of hot air, reduces airflow resistance to pretty much zero, and makes airflow over the heatsink more linear (rather than turbulent and deflecting every which way to escape out the sides).

https://en.m.wikipedia.org/wiki/Printed_circuit_board

http://www.mwrf.com/technologies/handling-heat-rfmw-circuits

https://www.electronics-cooling.com/1998/05/conduction-heat-transfer-in-a-printed-circuit-board/

 

[Valantar]

https://en.m.wikipedia.org/wiki/Printed_circuit_board

http://www.mwrf.com/technologies/handling-heat-rfmw-circuits

https://www.electronics-cooling.com/1998/05/conduction-heat-transfer-in-a-printed-circuit-board/

Thanks for providing some sources for my claims!

The results show that a typical PCB is not an efficient conductor of heat, leading to significant local variations in the PCB temperature in the vicinity of heat-generating components.

Also, the mwrf.com source seems to be talking about quite different types of electronics than a PC.

Of course, providing the right mix of thermally conductive materials to extract heat from an active source (such as a power transistor) or a thermal pathway (like a transmission line or circuit trace) may not always be so simple. For some designs, the addition of a component that may improve thermal management—e.g., a heat sink to an amplifier—may also thwart efforts at making the design as small as possible.

In the case of a GPU, there are three main heat generators: the GPU itself, the VRM, and the VRAM. All of these are relatively large, exposed and can be cooled by a heatsink with relative ease - which is a far more efficient thermal conductor than any PCB, even if the copper in the PCB is soldered directly to the heat-generating component.

Put simply: if PCBs were efficient heat conductors, the backs of GPUs wouldn't look like this through a thermal camera.

 

[chimonow]

So from what I am seeing from the custom Vega's is that so far they are all base on the RX Fury Nano boards.. So we should expect a RX Vega Nano or at least a custom version of it?

 

[chimonow]

I mean I could care less as Vega is nothing but a overclocked die shrink of Fury but it would be nice to know.

 

[Valantar]

So from what I am seeing from the custom Vega's is that so far they are all base on the RX Fury Nano boards.. So we should expect a RX Vega Nano or at least a custom version of it?

What? No. Not at all. The R9 Nano is way smaller than these boards, and only had a single 8-pin power connector. If anything, they're based on stock RX Vega boards.

I mean I could care less as Vega is nothing but a overclocked die shrink of Fury but it would be nice to know.

While I think I see what you're getting at, that's a gross oversimplification. If it was a pure die shrink, the transistor count wouldn't have jumped the way it did, and there are plenty of features added (although not many relevant for gaming today). Still, AMD made some very questionable choices in designing Vega (not increasing the CU count, among others).

 

[chimonow]

What? No. Not at all. The R9 Nano is way smaller than these boards, and only had a single 8-pin power connector. If anything, they're based on stock RX Vega boards.

While I think I see what you're getting at, that's a gross oversimplification. If it was a pure die shrink, the transistor count wouldn't have jumped the way it did, and there are plenty of features added (although not many relevant for gaming today). Still, AMD made some very questionable choices in designing Vega (not increasing the CU count, among others).

I'm talking about the actual circuit board, Not the plastic shroud.. Straight up based off of Fury nano. Also If you look at Clock for Clock performance of the Vega 64 vs. R9 Fury X you'll find a +/- 3fps difference in games... So yeah..

 

[Valantar]

I'm talking about the actual circuit board, Not the plastic shroud.. Straight up based off of Fury nano. Also If you look at Clock for Clock performance of the Vega 64 vs. R9 Fury X you'll find a +/- 3fps difference in games... So yeah..

Again: no. The board looks similar due to HBM making room for VRM circuitry around the GPU rather than VRAM, but saying that it's based on the R9 Nano is just plain silly. If they're based off of anything, they're based off of the reference Vega board. And yes, "IPC" (or whatever one might equate to that for GPUs - average performance per clock?) is pretty much identical between Vega and Fiji, but that doesn't at all mean that it's a "die shrink" of Fiji. A die shrink is a very specific thing: moving the same basic chip design to a new process with as few architectural changes as possible. Vega has a ton of architectural changes, just not ones that appreciably improve gaming performance in current games (rather, they add various compute capabilities, the HBCC, and so on). As such, calling it a die shrink is just plain wrong, even if the performance per clock is nearly identical. Heck, AMD engineers have been quoted saying that a lot of the added transistor count is to reach Vega clock speeds - if so, that alone means this isn't a die shrink.