News - AMD FirePro S9150 is "the world's most powerful server GPU"

[HEXUS]

Offers up to 5.07 TFLOPS of peak single-precision compute performance.

Read more.

[Platinum]

Can someone explain to me why this card pulls 235w and a 290x which is pretty much the same chip uses more power than a small country?, I must assume this is clocked a lot lower?

[3dcandy]

The 290x is tuned to run as fast as it can, I guess this will be tuned to use 235 watts. Subtle but important difference

[watercooled]

Can someone explain to me why this card pulls 235w and a 290x which is pretty much the same chip uses more power than a small country?, I must assume this is clocked a lot lower?

Don't believe everything you read...

Pic from: http://www.techpowerup.com/reviews/M...Gaming/22.html

[DanceswithUnix]

Can someone explain to me why this card pulls 235w and a 290x which is pretty much the same chip uses more power than a small country?, I must assume this is clocked a lot lower?

Clocked lower so it can run undervolted. If this is a server part, you want to cram loads of them into a single chassis so performance per watt is more important than absolute grunt.

Don't believe everything you read...

Pic from: http://www.techpowerup.com/reviews/M...Gaming/22.html

If you average 230W, just how much peak power are you consuming? The article implies 225W is the peak consumption, which is easier to power budget for and to wire up as you can get 75W from the motherboard and another 150W from a single PCIe cable.

[watercooled]

I may not have been clear, sorry; I wasn't comparing it to the Firepro (although the average power of the 290X is close to the spec), just that the 290X simply isn't this power-sucking monster it's often made out to be.

[Platinum]

I have a 290x and if that thing is only pulling 230w then I am a small off duty Czechoslovakian traffic warden.

[HalloweenJack]

its a banana


edit:

bit tech agree its on par with a titan for power drawr

http://www.bit-tech.net/hardware/gra...-x-oc-review/8

[kalniel]

Both titan and the 290X are going to be tuned for almost outright performance. The 290X is at least 10% faster than the FirePro card where it is allowed to be.

Plus they will be able to die harvest by now - save the lowest leaking chips for the firepro and sell the higher leaking ones as gamer cards.

[watercooled]

TPU have reviewed quite a few 290X samples from different MFRs, with different boards, etc, and they're all around the same ballpark in terms of power consumption.

TPU measure the power draw of the card itself, removing errors caused by swapping components between tests which can have a huge impact on the results, and which some review sites do without making it obvious.

But yeah, as above, it's common for server parts to have lower clocks than their desktop counterparts, this applies to CPUs too. Depending on the workload, it may be favourable to gain some cores with the power saved by turning down clocks, for example, since performance gained by increasing clock speed isn't linear but adding cores may actually be close to linear (again, depending on workload).

[Butuz]

its a banana


edit:

bit tech agree its on par with a titan for power drawr

http://www.bit-tech.net/hardware/gra...-x-oc-review/8

I'd put my money on the techpowerup review. Bittech are measuring the whole system power draw which is about as much use as a chocolate teapot for us if we are trying to work out how much the actual GPU uses. There's no way for us to tell how much of that ~400w power draw is the CPU or the GPU without bittech telling us, which they don't.

[kalniel]

I'd put my money on the techpowerup review. Bittech are measuring the whole system power draw which is about as much use as a chocolate teapot for us if we are trying to work out how much the actual GPU uses. There's no way for us to tell how much of that ~400w power draw is the CPU or the GPU without bittech telling us, which they don't.

But as the system's the same each time you can possibly compare cards, which is all we're doing here.

[Butuz]

Strictly, you're comparing systems. How much is the card using? 400w? or is the system using 200 and the card using 200? Who knows.

[kalniel]

Strictly, you're comparing systems. How much is the card using? 400w? or is the system using 200 and the card using 200? Who knows.

Of course, but if the question is whether x card draws more than y card then system draw is valid - if the rest of the system is the same then the only change is the card so you can answer the question - in this case, power draw vs a titan for example.