Last week, we brought you a quick bit of information on Intel’s upcoming Cascade Lake refresh for the High-End DeskTop (HEDT) and Xeon W markets. These chips, as we previously discussed, are a major improvement on their predecessors in price. Performance gains are a bit more workload-dependent, but there are some significant gains there as well, at least in specific markets. Intel is also making price adjustments to its high-end desktop chips via changes to the “KF” product series. Let’s talk about the new CPUs first.
Xeon W and Cascade Lake
There are a number of improvements and features that are common to both the Xeon W and X-Series / Cascade Lake platforms:
- Up to 1TB of DDR4-2933 in one DIMM per channel configurations.
- Intel Turbo Boost 3.0
- Intel Deep Learning Boost
- 2.5G Intel Ethernet
- Wi-Fi 6 AX200+
- Up to 18 cores
- Up to 72 PCIe 3.0 lanes
- Thunderbolt 3
- Intel Optane
On the Xeon W side, the new CPUs support features like ECC, RAS, Intel’s Virtual Raid on CPU, and vPro. The HEDT version of the platform offers fully unlocked processors and support for Intel’s one-touch overclocking solution, Intel Performance Maximizer. Some of these features, like the 2.5G Ethernet controller and Wi-Fi 6 support are new, as is the bump to 72 PCIe lanes in total (up from 68) — but the X299 platform is fundamentally limited in terms of what kind of advantages it can offer at this point.
Performance Improvements
The performance improvement claims are quite interesting, for a very specific reason. Intel has gotten some flak for its approach to the entire idea of “real-world benchmarking” of late. I’ve discussed the topic extensively in previous writing and don’t intend to dig back into it here, but the issue flashed to mind as I looked at the company’s claims for the Xeon W and Cascade Lake X families. First, here’s the Xeon W slide:
Pretty standard claims here. The “up to 97 percent faster” and “up to 2x faster” claims were both made in reference to a Xeon 1680v4. That’s an eight-core / 16-thread Broadwell CPU with a 3.4/4.0GHz clock. The Xeon W-2295 is 3.0 / 4.8GHz with 18 CPU cores. Workloads often don’t scale linearly at high core counts, so a 97 percent – 2x improvement from moving to 18 cores from eight makes sense. Performance testing here was done in Adobe Premiere Pro for 4K video editing, Autodesk Revit for 3D modeling, and when compiling Unreal Engine.
Now let’s look at HEDT:
The projected gains here are much smaller compared with what Intel is claiming for Xeon W. The 7 percent improvement is the straight-line gains from the 9980XE to the 10980XE when tested using Autodesk Maya. The 14 percent gain over the Broadwell 6950X is the odd part. Intel’s testing note only says that this data was also gathered in Autodesk Maya.
In all honesty, we would expect the gain to be higher here. Some 3D rendering applications use the GPU for rendering instead of the CPU, but 3D rendering applications are excellent places to show the improvements of adding more cores. Some of the tests we use at ExtremeTech, including Cinebench and Maxwell Render, both scale vastly better than Maya evidently did in Intel’s own tests.
We’ve reached out to Intel to ask why it chose to highlight this result in particular, but the fact that it did is rather interesting. The other performance claim for HEDT — namely, that it’s 2x faster in image throughput versus previous-gen and 7.9x faster than a three-year-old system — were also made against the 9980XE and 6950X, respectively.
Pricing
Last year, Intel introduced a new lineup of CPUs under the “KF” brand. These chips were still unlocked, like their “K”-part cousins, but they lacked a functional onboard GPU. Intel introduced these parts to help recover bad dies that would otherwise result in dead chips at a time when it was having trouble meeting demand for its chips. Up until now, “KF” CPUs have officially been priced just like “K” CPUs.
Today, that’s changing.
Up until now, the 9900KF has been between $488 and $500 on Intel’s official spec sheet. Now that’s dropping to $463. Other CPUs are seeing commensurate declines, with Intel now offering a small price cut if you buy a CPU without a GPU. It probably won’t change anyone’s buying decisions, but it makes thematic sense for a GPU-less CPU. Next up: Xeon W.
For reference: The current Xeon W-2195, launched in Q3 2017, is a 2.3/4.3GHz CPU with a 140W TDP and a list price of $2553. The new Xeon W-2295 is a 3.0/4.8GHz CPU with faster RAM, double the maximum RAM support (1TB, up from 512GB), a higher TDP (165W), and a list price of $1,333. That’s nearly a 50 percent price cut for a significantly faster CPU, even if we only consider base clock. The raw performance gains are modest, but the price improvements are phenomenal.
Finally, here’s Cascade Lake. We’ve shown this slide already, but we’ve tucked it in again for completeness’ sake. The improvements here mirror those for the Xeon W family — modest clock changes, but huge price cuts.
The Coming Showdown
These new chips are a clear shot over the bow at AMD and its still-unannounced lineup of 7nm Threadripper products. For the past few years, I’ve suggested that readers should keep an eye on the pricing disparities between Intel and AMD. Intel has improved its HEDT positioning before, but never by quite this much.
The big question now is how AMD will respond to it. The upcoming 16-core Ryzen 9 3950X was announced with a $750 price back in June and AMD hasn’t shifted that publicly yet. But the battle lines are clearly drawn. Intel is willing to commit to much lower prices and could still secure wins over a 16-core TR chip with an 18-core Cascade Lake, depending on how the IPC improvements to Threadripper play against the increased clocks on Cascade Lake-X.
Higher core counts will obviously be a tougher comparison, however. Intel can cut its pricing, but it’ll have a hard time matching AMD on cores when the smaller CPU manufacturer can bring 32-64 core solutions to market. Intel, however, may benefit at lower core counts if its per-core performance is as strong or stronger. It has the dominate and default position in the market as well, which always helps — recall that Intel is making this move more than two years after AMD took a price/performance lead in this segment. How will it all shake out? We’ll find out when parts drop.
Now Read:
- Intel’s Cascade Lake X-Series Will Slash Prices 50 Percent
- Intel’s Cascade Lake With DL Boost Goes Head to Head with Nvidia’s Titan RTX in AI Tests
- AMD Delays 3rd Gen Threadripper, 16-Core Ryzen 9 3950X Until November
No comments:
Post a Comment