Intel’s 28-Core 5GHz Monstrosity Isn’t Exactly a Standard Consumer Part
At Computex this week, Intel showed off a 28-core CPU running at 5GHz. That kind of clock speed in a 28-core CPU yields tremendous dividends, including a Cinebench-smashing multi-threaded score of 7334, but there were also questions about just how Intel would achieve and hold such tremendous clock speeds.
The problem is simple: CPU power consumption doesn’t scale linearly. The higher the clock speed you want to hit, the higher the voltage typically required to hit it. Higher voltages drive dramatically higher power consumption and temperatures, and because hot spot formation is a critical problem in modern microprocessors (it’s one reason why Intel CPUs drop their clocks when running AVX2 or AVX-512 code), the CPUs maximum frequency is fundamentally limited by the ability to move heat out of the core. That explains why Intel apparently had to go to such lengths to drive its 5GHz demo system.
Anandtech and Toms Hardware both got an up-close and personal look at the testbed, and there’s bad news for anyone hoping to take a system like this home any time soon. The LGA3647 CPU (LGA 3647 is Intel’s enterprise Xeon socket) shows every sign of being cooled by a water chiller — a device that uses water and a refrigerant to facilitate heat exchange in a closed-loop system. In the hierarchy of modern PC cooling, a water chiller is differentiated from a water cooler by using refrigerant (rather than just pumped water), but doesn’t hit the same sub-zero temperatures as a phase-change unit.
The cooler Intel used, a Hailea HC-1000B, draws more than 1,000W of power for its own operation, which apparently precluded running both the CPU and the chiller simultaneously. The Hailea HC-1000B’s output of 1500-4000L of chilled water (as cold as 4 degrees Celsius) is matched by its ability to cool a system of some 1700W of dissipation. Anand reports that a 1600W Corsair PSU was used to power the machine. That’s a $469 PSU all on its own.
Then, there’s cost. The CPU Intel demoed is almost certainly an overclocked Xeon Platinum 8180. That’s a CPU with a base price of $10,000, and it’s maximum frequency isn’t 5GHz — it tops out at the far more sedate 3.8GHz, not far from where AMD’s own Threadripper 2 is currently expected to land (Threadripper 2, at 32 cores, has a current expected turbo clock of 3.4GHz). But the chances that Intel brings this CPU to market for less money than that $10K are pretty low — after all, they’d be leaving huge amounts of cash on the table and they risk seeing the Xeon Platinum market eaten by a faster chip that they’d be selling for much less money.
When you consider the cost of a motherboard, the underlying CPU, the cooling solution required, and the other components, the chance that this finds a home in enthusiast systems is just minuscule. Anandtech thinks $100K for bespoke systems might make sense. Whether that’s true or not, I can’t say, but the idea of building one of these systems in a home rig of any sort is almost certainly a non-starter. And in that sense, AMD and Intel might as well have been talking about two completely different products. Threadripper might not be affordable to a majority of customers, but if you’ve got the cash you can step out and buy one. This chip will probably be much harder to find, even by the rarefied standards of such hardware.
Continue reading
Nvidia RTX Ada Titan Leaks Show Quad-Slot Monstrosity, Dual 16-pin Connectors
This GPU could have ended up costing $3,000 or more, but will probably never be launched.