06-25-2015, 04:35 PM
PCI-E specifications state that one 8 pin cable can supply up to 150W of power, with the PCI-E slot on the motherboard providing up to 75W of power.
One Fury X has two 8 pin connectors, which means, according to specification, it should draw at most 150+150+75=375W of power.
Instead of putting a third 8 pin power slot (ol' GTX 480 Lightning had 2 x 8pin and 1 x 6pin connectors, for example), they violate the spec, with unpredictable potential damage to the hardware.
In order to minimize the damage from the violation, they throttle the power draw using software/firmware shenanigans.
And they market this BS as an advantage.
Jeremi told me that it wasn't the first time AMD violated the PCI-E standard, so no surprises here.
Additionally, AMD knew darn well what could come out of the violation, so they decided to additionally minimise risks and go the "liquid cooling 4 stock" route.
Liquid cooling solutions I know of can dissipate 300~2700W of power (ERM-3K3UA a notable example for the latter).
But good luck even trying to install 8 Fury X cards without custom racks.
A card which cannot reasonably be cooled by air is an over-engineered card in my book.
Fury X is the product of staff cuts and desperation, that's what it is.
According to my rough interpolations using data for current GCN 1.2 cards, one Fury X could have made 27.5B p/s for NTLM (vs 35B p/s for Titan X), if it could draw as much power as necessary.
But I highly suspect it will be throttled, so that shouldn't be the stable speed.
Last, but not least, the last thing people care about here on hashcat.net is games.
In them, as long as you can get the GPU to meet the Vsync target (or twice as much for 3D), you're cool.
In other words, required performance is inherently capped.
And quick benchmarks do not reflect the real hourly gaming scenarios.
One Fury X has two 8 pin connectors, which means, according to specification, it should draw at most 150+150+75=375W of power.
Instead of putting a third 8 pin power slot (ol' GTX 480 Lightning had 2 x 8pin and 1 x 6pin connectors, for example), they violate the spec, with unpredictable potential damage to the hardware.
In order to minimize the damage from the violation, they throttle the power draw using software/firmware shenanigans.
And they market this BS as an advantage.
Jeremi told me that it wasn't the first time AMD violated the PCI-E standard, so no surprises here.
Additionally, AMD knew darn well what could come out of the violation, so they decided to additionally minimise risks and go the "liquid cooling 4 stock" route.
Liquid cooling solutions I know of can dissipate 300~2700W of power (ERM-3K3UA a notable example for the latter).
But good luck even trying to install 8 Fury X cards without custom racks.
A card which cannot reasonably be cooled by air is an over-engineered card in my book.
Fury X is the product of staff cuts and desperation, that's what it is.
According to my rough interpolations using data for current GCN 1.2 cards, one Fury X could have made 27.5B p/s for NTLM (vs 35B p/s for Titan X), if it could draw as much power as necessary.
But I highly suspect it will be throttled, so that shouldn't be the stable speed.
Last, but not least, the last thing people care about here on hashcat.net is games.
In them, as long as you can get the GPU to meet the Vsync target (or twice as much for 3D), you're cool.
In other words, required performance is inherently capped.
And quick benchmarks do not reflect the real hourly gaming scenarios.