Intel’s second generation of “scalable” Xeons dubbed Emerald Rapids, which is set to feature a maximum of 64 “Raptor Cove” cores, similar to those found on Raptor Lake, as opposed to Sapphire Rapids being limits to a maximum of 60 “Golden Cove” cores as previously seen on Alder Lake.
So basically, it’s an actual refreshment.
Though in terms of peak performance, Sapphire Rapids falls quite a long way behind AMD’s current crop of Zen 4 based EPYC processors, and with the release of new Threadripper processors there’s certainly no hope for team blue superiority for the foreseeable future.
We seem to be closer to the official announcement of Emerald Rapids than previously anticipated judging by a Twitter leak showcasing an engineering sample of Intel’s Xeon 8592+ Platinum, featuring 64-cores.
The CPU features a base clock of 1.9 GHz and can boost up to 3.9 GHz. However, final clock speeds for the retail version will likely change.
Emerald Rapids intends to bring significant upgrades to the Xeon line, such as expanded LLC cache, faster DDR5 memory support, CXL Type 3 support, and a notable double-digit improvement in performance per watt, which is more than likely merely in the teens rather than any substantially high figure.
In terms of power draw of which Intel is notorious for falling behind AMD in this regard, the base TDP is fixed at 350W, yet it can be adjusted to 420W, making it capable of reaching peak power draws of 500W.
While it seems the peak power could hit 922W, this might be specific to this pre-release sample or setup, but given the severe performance difference between Intel Xeon and AMD EPYC / Threadripper, I wouldn’t be all that surprised if Intel once again tried to push a 900+ wattage griller.
This processor boasts 448MB of combined cache, featuring 120MB in L2 cache and 320MB in L3 cache. It’s a significant improvement over the flagship Sapphire Rapids SKU (8490H), which has just 232MB of cache.
If we’re being honest the performance difference from Alder Lake to Raptor Lake most likely won’t be substantial enough, even with an additional 4 cores on offer, AMD are able to offer almost identical single core performance in a much more efficient package, now with a maximum of 96-cores on offer with less wattage consumed.
I just don’t even see the point in Intel even pressing on with Emerald Rapids, they would’ve had a chance had they not delayed Sapphire Rapids for three consecutive years.
It’s rather hard to gage expected performance figures, the 60-core Platinum 8490H is simply getting swamped, even by the 64-core EPYC 9554 with both of which in dual socket configuration and the performance difference only extends further when comparing AMD EPYC processors with greater than 64-cores as well.
The main difference between Raptor Lake and Alder Lake was the adjustments made to its cache hierarchy, but even synthetic performance evaluations are skewed given the increased amount of E-Cores on offer versus the previous generation, there are no performance evaluations made between the two on a purely Performance core basis, but I’m still hopeful that a 64-core Raptor Cove based rendition would at least provide a >20% bump in performance over the 8490H, which would put Intel around that of AMD’s 64-core EPYC 9554 in terms of flagship performance.
Furthermore, another leak has surfaced revealing a 48-core, 96-thread variant known as the Platinum 8558U via Geekbench.
This processor seemed to maintain a steady 2.0 GHz base clock without an operational boost capability. The CPU is reported to possess 356MB of combined cache, showcasing a substantial increase of over 150MB compared to its counterpart from Sapphire Rapids with the same core count. (260MB L3 Cache + 96MB L2 Cache vs. 105 MB L3 + 96 MB L2).
It’s nice that Intel are actually able to move forward with their “scalable” Xeon without tripping over their feet, given the several years of delays of Sapphire Rapids and a hardware bug that apparently halted its production temporarily, but I genuinely don’t see a point if Intel’s flagship 64-core monstrosity roughly matches the performance of an equal 64-core Zen 4 based offering while consuming near twice the power consumption. I don’t see why any customer would opt to purchase Intel’s Xeons unless otherwise bribed with discounts and benefits, bribery is a tactic that comes secondhand to Intel after all.
