AMD Announces the Radeon R9 Fury Graphics Card

[newtekie1]

Most 390X cards have a baseplate that is part of the main finstack above the core which cools the VRAM.

None that I've seen. They have a baseplate that cools the VRAM, but I haven't seen one that is integrated into the main heatsink.

 

[buildzoid]

None that I've seen. They have a baseplate that cools the VRAM, but I haven't seen one that is integrated into the main heatsink.

Both the Tri-X and Windforce have the VRAM hooked up to the main heatsink.

 

[LightningJR]

and that memory actually is responsible for a pretty large amount of heat.

Really? I didn't know this, where did you get this information? I would love to read some info on HBM's high heat output. Links?

 

[Aquinus]

None that I've seen. They have a baseplate that cools the VRAM, but I haven't seen one that is integrated into the main heatsink.

That is how it works on my 6870s, but they're also many years old. Consider the drop in multi-monitor idle usage and lower operating voltages that HBM runs at. I'm reluctant to think that HBM uses any more electricity than GDDR5. In fact, I suspect it would use less because it's clocked lower and runs at lower volatges.

Really? I didn't know this, where did you get this information? I would love to read some info on HBM's high heat output. Links?

I too am interested in where this information came from.

 

[newtekie1]

Both the Tri-X and Windforce have the VRAM hooked up to the main heatsink.

No they don't. The baseplate touches the main finstack, but that's it. There's no TIM between the baseplate and the main finstack, the two aren't soldered together. There is virtually no heat transfer from the baseplate to the finstack.

Really? I didn't know this, where did you get this information? I would love to read some info on HBM's high heat output. Links?

In fact, I suspect it would use less because it's clocked lower and runs at lower volatges.

I too am interested in where this information came from.

I didn't say HBM has a high heat output. I said memory, not specifically HBM, is responsible for a pretty large amount of heat, meaning you just can't ignore it like it isn't putting out any heat at all. And now that the heat load is placed on the main heatsink the main heatsink has to be bigger to handle it.

Yes, HBM is more energy efficient than GDDR5, we know that. Here is a pretty interesting read, I think it might actually be the first write-up of HBM. Look at the chart at the beginning, ~20% of the power consumption for the 6990 is from the memory. That amount of heat can't just be ignored. In the case of the 6990 that would be 60w of power/heat. It doesn't say exactly how much better HBM is, but even if you half that, you're still at 30w of heat. That can't be ignored.

 

[LightningJR]

No they don't. The baseplate touches the main finstack, but that's it. There's no TIM between the baseplate and the main finstack, the two aren't soldered together. There is virtually no heat transfer from the baseplate to the finstack.





I didn't say HBM has a high heat output. I said memory, not specifically HBM, is responsible for a pretty large amount of heat, meaning you just can't ignore it like it isn't putting out any heat at all. And now that the heat load is placed on the main heatsink the main heatsink has to be bigger to handle it.

Yes, HBM is more energy efficient than GDDR5, we know that. Here is a pretty interesting read, I think it might actually be the first write-up of HBM. Look at the chart at the beginning, ~20% of the power consumption for the 6990 is from the memory. That amount of heat can't just be ignored. In the case of the 6990 that would be 60w of power/heat. It doesn't say exactly how much better HBM is, but even if you half that, you're still at 30w of heat. That can't be ignored.

Well you did say "that memory" instead of "the memory" but I understand if it was a mistake.

Yeah the write-up you linked showed an increase of 5.8C when using HBM. I wouldn't say that this is, as you say, "and that memory actually is responsible for a pretty large amount of heat" It is obviously more heat than a custom card would have but not much difference than a blower styles heatsink/fan would have since they actively cool the ram modules on the same heatsink as well.

No change in nm on the core, and large increase in shader count will give you most of the extra heat. The power savings will be mainly from the memory interface I think. The core is not that dissimilar to the 290 series, only thing different is an increase in shader cores and L2 cache (ofc the HBM) The only thing to brig down power/heat is a more mature 28nm process and HBM.

The thing is you can have great performance per watt and still have high heat. You can also have bad performance per watt and have low heat. There are many variables that affect why this can happen.

 

[FrustratedGarrett]

Great review W1zzard! A wide spread out selection of games gives a more accurate performance estimate of this card. Looks like the Fury card is doing well against 980.
The FuryX is beating the 980TI in half these games or so, which isn't bad.

We need Vulkan/DX12! And in these GameWorks games you have most of the physics processing done on the discrete GPU, which is lame, considering it can be done easily on a quad core CPU, under a low overhead API. I don't know....
What's the first DX12 game we're getting this year, if anyone know?