Intel May Not Launch a New HEDT CPU This Year
Normally, Intel refreshes both its desktop and its HEDT product lines on a yearly cadence, with HEDT tending to launch after the desktop refresh. Since Skylake-X debuted in 2017 the two platforms have used the same architecture, with desktop chips launching first. Now, it looks as though Intel won’t have an HEDT refresh this year.
This is less unusual than it might seem. Since launching the 3960X in 2011, Intel has skipped years on several occasions. There was a nearly two-year gap between the 3960X and the 4960X (Q4 2011, Q3 2013) and the 5960X and 6950X (Q3 2014, Q2 2016). Since the 6950X, Intel has delivered a regular cadence of updates, with Skylake X debuting in 2017, the 9th Gen HEDT refresh in 2018, and last year’s launch of the Core i9-10980XE.
In most cases, these pauses happen at platform boundaries. The X79 supported the 3960X and the 4960X, the X99 anchored the 5960X and 6950X, and the X299 has anchored the Skylake X, Skylake X Refresh, and Cascade Lake families. If Intel is holding off for a new chipset, it would explain why the company doesn’t have an HEDT chip ready for this segment.
Frankly, it makes sense for Intel to hold off at this point. Rocket Lake, Intel’s next-gen 14nm architecture (at least according to rumor) isn’t ready yet, and won’t ship until the end of the year. Theoretically, Intel’s next HEDT part could be based on Ice Lake or Tiger Lake (Sunny Cove and Willow Cove architectures, respectively). Rumor has it that Rocket Lake is the 14nm implementation of Sunny Cove while Tiger Lake is the 10nm variant.
We don’t even know if the next HEDT platform will be built on 14nm or 10nm. If Intel decides to follow standard procedure, it’ll be a 10nm CPU with a few strategic changes to differentiate it from Xeon. If it sticks with the bifurcated approach we’ve seen to the enthusiast market, we could expect a 14nm core with higher base clocks compared to 10nm. Then again, we don’t know what the clock delta will be between Tiger and Rocket. In theory, Intel’s 10nm+ should reduce the clock differential between the two nodes, which has grown rather large.
As for Intel’s competition, we haven’t heard a peep out of AMD regarding Threadripper. After last years’ sprint to 64 cores, we’d expect AMD to keep core counts steady this time around and likely focus on improvements to IPC and clock instead. With Windows already unable to use 128 threads in a single process under the default Windows thread scheduler, there’s not much benefit to pushing past the 64C/128T point.
There’s also no info yet on how many cores future HEDT chips may offer. Current solutions top out at 18 for Intel and 64 for AMD, and that’s obviously a gap that it’s in Intel’s best interest to reduce. Thanks to THG for spotting the slide.