What does AMD’s new Epyc 7H12 mean?

It isn’t just a CPU, it is a sign of competitiveness

AMD EPYC LogoAMD’s new 7H12 CPU just made life really hard for Intel’s Whitley next generation Xeon platform. SemiAccurate thinks AMD just one-upped Intel and planted a stake in the ground on wattage.

The Setup:

The story goes like this, AMD’s Epyc/Rome 64-core CPU can’t be matched by anything Intel makes now or into 2020. In it’s base form, the Epyc 7742 essentially doubles the performance of Intel’s top Xeon 8280 at less than half the price, that is a 4x TCO advantage on the CPU side. That is only the beginning though, AMD offers way more features like PCIe4, 128 PCIe(4) lanes vs 48 PCIe(3) for Intel, inter-socket bandwidth that isn’t choked, and has much more memory capacity. Intel can scale to more sockets but then the afformentioned inter-socket bandwidth makes that a moot point. In short it isn’t a fight.

Tepid Response:

Intel responded by shooting their own foot off with the Cascade Lake Xeon 9200 series, a part so popular with OEMs that no one bothered to make a board for it. This $20K+++ CPU, you have to ask about pricing, Intel won’t tell you even if you do though, was dead before arrival. Part of the problem is they lashed two Cascade Lake-SP CPUs together to make a single 9200 adding all the latency issues they were attacking Epyc on last year to the platform.

To get performance that isn’t a large regression from a 2-socket 8280, Intel needed to jack the power up to 400W. This precluded air cooling and forced customers to add liquid to their data center which some are loathe to do. This also means a bespoke chassis which, again, no one other than Intel bothered to make. Add in all the other restrictions like one DIMM per channel and you have a very expensive benchmark attempt that backfired.

The Future:

It was pretty clear that Intel has nothing to counter AMD’s Rome this generation but next year’s Cooper Lake is only three quarter’s away or so. Even with Cooper Lake it will lose on merit to the current Rome, we told you that it would over a year ago. Since then Intel has lept into action and shot off the other foot. The I/O die on Cooper went the way of the dodo, and then the big tell, TDP. When Intel jacks up the TDP of a platform a few quarters before launch, it is a desperation play, OEMs and ODMs don’t have the time to respond.

Worse yet is that once you go over the 250W limit, it is very hard if not impossible to air cool a CPU. Intel’s own Xeon 9221 and 9222 are air cooled but they cap out at 250W, the 350W and 400W 9242 and 9282 are both mandatory liquid cooling. By pushing the TDP up on the Whitley platform, that would be next year’s Cooper Lake and Ice Lake-SP CPUs, Intel is screaming that they have no competitive products. Throwing Wattage at the issue is the only response they have.

Hear that whistling sound? That is an AMD nuclear bomb in 4094 pin package with 7H12 written on the side.

Note: The following is analysis for professional level subscribers only.

Disclosures: Charlie Demerjian and Stone Arch Networking Services, Inc. have no consulting relationships, investment relationships, or hold any investment positions with any of the companies mentioned in this report.

The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and SemiAccurate.com. SemiAccurate.com is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of SemiAccurate.com, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate