When AMD’s Lisa Su demonstrated the company’s upcoming 7nm Ryzen CPUs at CES, sharp-eyed viewers quickly realized that there was room for a second chiplet on the processor she was holding. This implies the potential existence of a 7nm Ryzen desktop CPU with up to 16 CPU cores — double the amount AM4 supported when Ryzen debuted. AMD has been coy about its plans, confirming only that it won’t use the gap for a GPU with this generation of Ryzen products.
The safe assumption, at this point, is that AMD will likely roll a 16-core, AM4 Ryzen at some point in 2019. That, in turn, suggests the likely existence of a 12-core chip in a die-recovered part. And based on information submitted to UserBenchmark.com and spotted by Apisak, there are new 12-core parts coming to the AM4 platform.
AMD Eng Sample: 2D3212BGMCWH2_37/34_N
AM4, 1 CPU, 12 cores, 24 threads
Motherboard AMD Myrtle-MTShttps://t.co/9G0Tt11bOe
— APISAK (@TUM_APISAK) January 24, 2019
The CPU in question is a 2D3212BGMCWH2_37/34_N, which doesn’t exactly roll off the tongue. “2” is a reference to the second run of prototypes, “D” means desktop, and “3” likely means “3rd generation Ryzen.” The “B” is a letter that only shows up on Ryzen desktop chips, not Threadripper, and the “C” is for a 12-core chip. “W” is a reference to core cache (we don’t know how much the chip has). H2, according to Reddit, is a designation used for Matisse, and the projected clock is a 3.7GHz boost, 3.4GHz base.
The results themselves are disappointing — as TechSpot notes, integer and FPU performance are quite a bit lower than a set of five Ryzen 7 2700X CPUs from the same database.
In these results, the single, quad-core, and full-core performance of this 12-core CPU is compared against a suite of five 2700Xs that also average a 3.6GHz clock speed. As you can see, there’s an implication that the unnamed 12-core is also significantly slower than the other processor.
But remember — according to the UserBenchmark page, this suite of tests was run on a 4GB RAM DIMM that somehow managed to have 6.1GB of usable memory. If that’s true, it means the test was being run on a single channel of DDR4-2666 (the report claims 1.3GHz, but I’m assuming this is the base clock, not the effective clock). That’s a very small pipe to pull data through, and 4GB of RAM isn’t much for a modern system. Now, toss on the fact that this is an early engineering sample in an unknown motherboard.
One of the reasons I’m not concerned about these figures is because it’s just not unusual to see early engineering samples substantially lag the performance of the finished product. AMD and Intel both keep in close contact with reviewers in the run-up to a launch. If we report a performance issue, the first question invariably asked is “Which UEFI version are you using?” I’ve seen a UEFI flash improve performance by 10 percent or more — and that’s when testing retail processors on retail boards.
The limited speed grade on the CPU isn’t a concern, either. While it’s substantially below the clock rate for the Threadripper 2920X, this could reflect an intent to offer a lower-clock, cheaper 12-core. Alternately, it might just mean that clock rates haven’t been finalized yet, which is in-line with the guidance AMD gave at CES 2019. But again, an engineering sample isn’t expected to necessarily run at full clock, either. If AMD is targeting a late Q2 / early Q3 launch for Ryzen, it has months to bring overall clock rates up.
Either way, this is still more evidence that a 12-core and therefore 16-core Ryzen CPU is likely to debut for AM4.