Today we have the launch of the 2nd Generation Intel Xeon Scalable refresh SKUs. Make no mistake, this is Intel’s competitive side coming out. The original 2nd Gen Intel Xeon Scalable Launch happened in April 2019, about four months before the subsequent AMD EPYC 7002 series Rome Launch. Intel at the time knew it had some breathing room. AMD was four months behind but there was a catch. AMD was transitioning from PCIe Gen3 to Gen4 over a year before Intel will do the same. As a result, AMD platforms had to be re-designed with more robust motherboards to handle the higher platform throughput. Now that those systems have largely hit the market, we are seeing Intel make competitive moves. The first was discontinuing the 2nd Gen Intel Xeon Scalable M SKUs and lowering the price of competitive high-memory L SKUs. This “refresh” brings lower prices and more performance to more directly combat the AMD EPYC threat in the 32 and under high-volume server CPU market segments.
Let us be clear, a few years ago, getting 11-70% better performance per dollar across a line spanning from 8-28 cores we would have called this a new server processor series launch instead of a “refresh.” With 18 new SKUs and a different set of capabilities in many segments, one could call this an interim series launched between the original 2nd generation Intel Xeon Scalable and the next-generation Ice Lake Xeon series. In fact, one could look to the scope of this overhaul as support for the notion that the 2H20 Ice Lake Xeons are not going to launch in the first week of the second-half like the 1st generation Intel Xeon Scalable chips did. With this launch, Intel is effectively re-creating the market bi-furcation that we saw in the previous Xeon E5 and E7 series with the new SKUs making up the mainstream Xeon E5 series. It took a good amount of analysis to get there, but we can walk you through why that is.
In this piece, we are going to look at the new “refresh” SKUs. We are going to dissect Intel’s product strategy. We are then going to share some results benchmarking the Intel Xeon Gold 6248 and the new Xeon Gold 6248R to see just how much of an impact these changes have. We are going to end with some market analysis to outline the impacts both to the Intel v. AMD story but also what it means to the rest of the Intel portfolio. Let us be clear, this is not simply a mid-cycle 100MHz bump across some SKUs like we saw in the Xeon E5 generations. Instead, this is almost a re-positioning of the SKU stack for important market segments.
2nd Gen Intel Xeon Scalable Refresh Video Overview
We have a 13-minute video talking about some of the highlights in this comprehensive article.
The full article has significantly more detail and analysis than the video, but we know some prefer to listen rather than read.
2nd Gen Intel Xeon Scalable Refresh SKU Overview
Here is Intel’s official slide on the new parts which is the only slide we are going to use from Intel in this article. If that Intel Xeon Gold 6258R at $3,950 catches your eye, it should. In our AMD EPYC 7002 series launch article we said about the EPYC 7742: “AMD at around $7000 is essentially saying Intel needs to start their discounting at 73% to get competitive, and that is not taking into account using fewer servers.” (Source: here) Here we have Intel discounting Platinum 8280 by over 60% and calling it the Gold 6258R. That sets the tone for this table and the launch focused on mainstream servers:
Update: We asked for and received an updated version of the above chart from Intel.
Something will immediately stick out to readers. The Intel Xeon Silver 4215R supports Intel Optane DCPMM.
One will also notice that there are four non-R SKUs. First, one can see the Intel Xeon Gold 6256 and Gold 6250 which are lower-core count but frequency optimized SKUs with large caches, high TDP, and Turbo Boost speeds up to 4.5GHz. There is an extended temperature part the Intel Xeon Silver 4210T. One can also see what may be the most intriguing option, the Xeon Gold 6208U which is a “U” or single-socket part that Intel started offering in this generation to combat AMD.
One will also notice that Intel did not add any Xeon Platinum SKUs. These are really designed to be mainstream parts. Something we learned several weeks ago before leaks of these SKUs went public, is that the R parts are designed for one and two-socket servers. When we were preparing for our Supermicro 2049P-TN8R Review, the question arose, should we hold the review for a week or two and use the refresh parts instead. At that point, we were told that the new R parts did not support 4P operation even in the Intel Xeon Gold segment. Intel’s press release confirmed this:
The new processors, labeled with an “R,” “T,” or “U” suffix, are designed for dual-and single-socket mainstream and entry-level server systems. (Source: Intel)
Not only did Intel pass on updating Intel Xeon Platinum 8200 SKUs that are 8-socket capable, but it also removed the 4-socket capabilities from the Xeon Gold range for the first time since the Intel Xeon E5 and E7 series were unified with the Xeon Gold 6100 and Platinum 8100 series in 2017. Here is a slide from our 2017 Intel Xeon Scalable Processor Family: Platinum Gold Silver Bronze Naming Conventions piece:
That 4S capability seems to be removed from the new Gold 6200R and Gold 5200R parts. What this effectively means is that Intel has re-created the bifurcation lines and premiums associated with the Xeon E7 and Xeon E5-4xxx lines from years ago. For example, one can now buy a 2-socket only Xeon Gold 6258R for $3,950 list price or the 8-socket capable Intel Xeon Platinum 8280 for $10,009 which both have the same size cache, clock speeds, and TDP. This is going back to a Xeon E7-like 4-socket and 8-socket “tax” structure. Beyond the sockets for these 4P and 8P capable chips, one can also potentially pay a memory “tax” by purchasing “L” SKUs which has been greatly reduced recently as we covered in our 2nd Gen Intel Xeon Scalable M SKUs Discontinued piece. Intel is extracting value for higher-end configurations but it is worth mentioning that Intel still needs 4-sockets to get to the 64-cores that the AMD EPYC 7742 has. Essentially Intel is competing with its higher-end and higher-price 4-socket capable SKUs against the 2-socket AMD EPYC 7002 64-core parts.
Putting these new chips into context, it is important to keep that distinction in mind. The new “R” parts do not replace the older parts in the SKU stack. Instead, previously announced parts are still potentially used in 4-socket, 8-socket, and high-memory scenarios. Here is the mainstream socketed 2nd Generation Intel Xeon Scalable family with the two new frequency optimized SKUs along with the new refresh SKUs highlighted:
We wanted to point out here that Intel did more than add a few hundred MHz. Take the Intel Xeon Gold 6242 and Gold 6242R as examples where the price remains the same but little else. Intel increased the core count by 4. The base clock went up by 300MHz to 3.1GHz and maximum Turbo Frequencies moved up by 200MHz while the TDP also moved from 150W to 205W. There is another subtle difference. Previous 20-core SKUs like the Intel Xeon Gold 6230 had 27.5MB of L3 cache per processor. Many of the new chips, like the 20-core Xeon Gold 6242R have 35.75MB per chip. That is approximately a 30% increase in total cache or cache per core. Still, Intel is nowhere near the cache that AMD has, for example, the new AMD EPYC 7532 has 32 cores and 256MB of L3 cache or 8MB per core, but Intel is aware of its architectural deficit and is addressing the gap.
Another interesting note here. If you are a 4-socket Intel Xeon Gold buyer and pick the Gold 6248, a 20-core part you pay about $500 more for being able to use the chips in 4-socket servers or about 20% more. That is not all. The more expensive Gold 6248 has only 27.5MB of cache, not getting that Gold 6242R 30% cache per core bump. Base clocks are 600MHz lower at 2.5GHz versus 3.1GHz. Maximum Turbo Clocks are 200MHz lower at 3.9GHz versus 4.1GHz. The sole metric other than 4-socket capability that the older SKU is better at is its 150W TDP versus 205W TDP on the new part. The 4-socket “tax” is not just price, but it is also on specifications with the older parts.
Extending this further to look at how much Intel is charging for each 1GHz of base clock on a core, we can extrapolate these figures:
Just looking at the non-frequency optimized Intel Xeon Gold 6200R SKUs, we now get an average of $42.31 per 1 core at 1GHz base clock. Removing some of the frequency optimized parts such as the Intel Xeon Gold 6244 and Gold 6234 we previously had an average of $58.19. Using that methodology, we get around a 27.3% average price reduction in the mainstream portion of the Intel Xeon Gold 6200R versus the Gold 6200. For each “Cascade Lake” generation’s core at 1GHz in the segment, Intel has essentially discounted the price by more than a quarter. This is, of course, before the company’s deal, program, and contractual-based discounting.
Next, we are going to take a deep-dive into these changes. We are going to follow that up with performance testing of the Intel Xeon Gold 6248R’s in the cover photo versus the Gold 6248 launch SKUs to see if these numbers translate into a meaningful impact. We are then going to discuss the market impact of the refresh and give our final words.