nevcairiel:

I find that comparison that everyone seems to be making just silly though. AMD showed TR2 on Air, running at probably 3.2-3.4GHz (probably not even all cores at those clocks). The 28 core Intel used could've easily done those clocks on air as well - thats not a real exciting thing to achieve.

nevcairiel:

I find that comparison that everyone seems to be making just silly though. AMD showed TR2 on Air, running at probably 3.2-3.4GHz (probably not even all cores at those clocks). The 28 core Intel used could've easily done those clocks on air as well - thats not a real exciting thing to achieve. For AMD, TR2 is just a stripped down Epyc anyway, 4 full dies, less memory channel, less PCIe lanes, re-cooked at 12nm (although not using the full potential of 12nm). Intel could've just announced that they are bringing Xeon XCC cores to workstations instead of only on servers - well, they basically did, but they decided to make some silly show out of it, but in the end its basically the same thing.

Witcher29:

The 28 core cpu from intel is still faster then a 32 cpu from amd even in slower speeds, speed isnt always anything these days. Quality over Quantity.

Witcher29:

The 28 core cpu from intel is still faster then a 32 cpu from amd even in slower speeds, speed isnt always anything these days. Quality over Quantity.

sarsar:

But the real issue is the VOLTAGE, nobody said anything about the voltage of that CPU at 5 Ghz! If it takes over 1.5v i think nobody will run it 24/7 at those speeds, a few maybe will buy that chiller and the CPU but they will not run it at 1.5v when they paid 2-3000 $ for it.

Fox2232:

That's not happening unless it's on 7nm. I can do you power consumption under full load graph at different clocks today. But following should be enough to extrapolate 8C/16T: 4.2GHz all cores => ~180W 3.9GHz all cores => ~125W 3.8GHz all cores => ~105W 3.6GHz all cores => ~80W I say that 32C/64T can fit into 300W @3.4GHz and stock 250W limit will put it in 3.0~3.2GHz range once all cores are fully utilized. Basically, above 4GHz this 12nm+ lithography becomes very inefficient. Zen is good uArch, but it is being held back by manufacturing process. If AMD licensed intel's production capacity, even 1st gen Zen would prove to be King of the Hill.

xrodney:

Ok, so 1kW CPU + 1.7kW chiller + 2.5 kW AC unit to keep room at low ambient temperature and it will still likely termalthrottle in minute or so. Thats what I would say (d)efficiency world record. Intel PR at its best.

Fediuld:

It is even worse than that for Intel PR. The chiller used is a Hailea HC 1000. This thing is illegal to be sold in North America and Europe because it uses R22 gas (or R134a depending model type) as freezing medium. Usage of those gasses are banned for 14 years now in the case of R22.

tunejunky:

meanwhile they've snapped up market share and shoved Intel out of one of their most profitable markets with Epyc.

chispy:

In fact it does looks like one of those Hailea 800 watts cooling chiller or similar , maybe even heavily moded for more than 800w , chillers can go down to -21c subsero with the right mixture of gases on the compressor. I

Andrew LB:

Calm down Sir... please take a deep breath. Lets try sticking with the facts instead of nonsense like saying "amd shoved Intel out of one of their most profitable markets". I mean do you honestly believe that? Here's the deal. With the release of Epyc, AMD's datacenter market share went from 0.3% to 1% according to Mercury Research's latest figures. Nobody was shoved, pushed, or even bumped. If any actual contact was made, i'm sure Intel would have thought it was someone trying to rub up on them. lol. To better put this in perspective, AMD's Opteron once commanded over 25% of this same market. And did so much faster than Epyc has so far.