Monday, February 11th 2019
New Intel Core i9-9990XE Sheds Cores in Favor of High Clock Speeds, Benchmarked
Intel is giving final touches to a new socket LGA2066 high-end desktop processor with an interesting model number for its specifications. The new Core i9-9990XE is positioned above the current flagship i9-9980XE. Normally you'd expect it to be the same 18-core "Skylake-X" chip with a speed-bump, however, the i9-9990XE is a unique proposition. It sheds cores in favor of significantly higher clock-speeds than the i9-9980XE.
The i9-9990XE is a 14-core/28-thread processor, based on a binned "Skylake-X" HCC (high core count) die, and uses STIM (soldered thermal interface material) between the die and integrated heatspreader (IHS). It features some aggressive clock-speeds, with 4.00 GHz nominal clock-speeds, and a massive 5.10 GHz maximum Turbo Boost frequency that beats even the Core i9-9900K. Besides 14 cores, the i9-9990XE is configured with 19.25 MB of shared L3 cache, and 1 MB of L2 cache per core. The four disabled cores alone don't help Intel's efforts to dial up clock-speeds. Intel has increased the chip's rated TDP all the way up to 255 Watts!Intel hasn't revealed a price for the Core i9-9990XE yet, and there are indications of the chip being OEM-exclusive. Workstation integrator Puget Systems has put out a handful benchmarks for this chip, in which it beats the i9-9980XE in some HEDT-relevant benchmarks despite 4 fewer cores:
Source:
Tom's Hardware
The i9-9990XE is a 14-core/28-thread processor, based on a binned "Skylake-X" HCC (high core count) die, and uses STIM (soldered thermal interface material) between the die and integrated heatspreader (IHS). It features some aggressive clock-speeds, with 4.00 GHz nominal clock-speeds, and a massive 5.10 GHz maximum Turbo Boost frequency that beats even the Core i9-9900K. Besides 14 cores, the i9-9990XE is configured with 19.25 MB of shared L3 cache, and 1 MB of L2 cache per core. The four disabled cores alone don't help Intel's efforts to dial up clock-speeds. Intel has increased the chip's rated TDP all the way up to 255 Watts!Intel hasn't revealed a price for the Core i9-9990XE yet, and there are indications of the chip being OEM-exclusive. Workstation integrator Puget Systems has put out a handful benchmarks for this chip, in which it beats the i9-9980XE in some HEDT-relevant benchmarks despite 4 fewer cores:
54 Comments on New Intel Core i9-9990XE Sheds Cores in Favor of High Clock Speeds, Benchmarked
2700X: £300 (comes with cooler) :clap: and you can put Zen2 in it in a few months and likely beat the 9900K in everything*
Huh, you could buy a 1920X and X399 Motherboard for the price of a 9900K (here in UK).
*except 720p ultra low (ini tweaked no shadows), the 9900K still gets 12% more FPS here
I do think Intel is pushing it too far here, both because 255W TDP is getting into a territory it's hard to cool without massive cooling, and secondly because they can't supply enough of these.
But I do like the idea behind the product, a HEDT CPU which have as many fast cores as possible, rather than just as many cores as possible by sacrificing a bit of core speed. So I think a HEDT CPU with as many fast cores as possible (and keeping TDP at ~200W) would be a good complement to HEDT CPUs with many more cores.
When it comes to workstation use, the best choice is not as simple as the highest core count etc., it really comes down to the tasks the machine will be running. People commonly think it comes down to multithreading or not, but in reality many heavily multithreaded workloads are sensitive to core speed. There are async multithreaded workloads, where worker threads are basically independent and can scale those workloads almost linearly until you run out of cores or hit a bottleneck. Many typical server loads fall into this category, but also larger batches of rendering or video encoding. For such workloads it's fine to sacrifice core speed, because what matters is total throughput. On the other hand there are many workloads which are synchronized, and while they still benefit from more cores, the synchronization overhead grows quickly, so more core speed is required to achieve good scaling.
While the current HEDT lineup is of course generally good, it's still not good enough to convince me to pull the trigger for my particular mix of development and CAD/modeling needs. And I have worked on Skylake-X, so I know how much better than its predecessors it is. So for me I would probably have to wait longer, until Ice Lake-X/SP arrives in 2020 :(, and then go for the CPU with the highest number of fast cores. As I've said, there are rumors about this chip being an exclusive. The rumors dating from a couple of weeks ago claimed it will only be available through selected system builders and through auction(!). I really hope these rumors are untrue. Launching products and sending them to reviewers is a bad thing if the customers can't buy the thing, regardless of how good/bad it may be. One example of this would be the i7-8086K, a perfectly fine product, except that it was not really available. It is certainly pushing a territory where things get really hot, especially if you're going to run this for hours of load at the time. But the biggest problem will probably be airflow in the case, rather than the CPU cooler itself.
But still remember that GPUs manage to do 250-300W with less optimal airflow, but are not completely silent though.
Well, for a workstation lasting ~5 years of heavy use, overclocking is out of the question. I'm already worried that a cherry-picked CPU pushing it this far like the claimed i9-9990XE would even be risky at stock, so I wouldn't dare to buy it for that reason.
And watercooling only solves the problem of moving heat from the CPU and into the case. Most cases will be a bottleneck long before a decent air cooler (like Noctua NH-U14S) will be a problem. For watercooling to have real potential, I would need one of those special cases with a side-vent for radiators or similar. But my days of watercooling and overclocking are done. I do chuckle though when I see the kids these days building PCs with giant radiators and practically no airflow.:rolleyes:
And how this thing will be efficiently cooled?
Did you check my specs ? I'm using Intel/Nvidia computer.
And you're dead wrong.
Threadripper 2990WX or Epyc 7501, 7551 or 7561 have 32 cores, 64 threads, while intel tops up at 28.
And AMD already has working and demonstrated EPYC next gen 64-core CPU's (7nm chiplets), will launch very soon.
Intel has absolutely nothing to respond to that, not even if they do that glued 28+28 chip, that's still a lose in the "cores war".
But they have the clocks advantage... for now.
Having had a quick look on a AIO's tech page here I couldn't see anything that pointed out TDP handling at all, unless I'm just blind....
I did not mean any 9940X can OC to these speeds reliably (but with a potent cooling solution as would be needed for 9990XE anyway, I believe most 9940X's can match 9990XE clocks), hence the "value" this 9990XE brings: a guaranteed high clocks on a 14-core chip, AND THAT IS IT.
A 9940X is a 165W part, this is 255W. I bet that if one can manually OC many (say 25) 9940X's, and compare the power utilization and temperatures on 25 other 9990XEs, the differences would be statistically insignificant.
I imagine this will have a hefty premium over the 7940x/9940x
Shoe is on the other foot now Intel fans.
Enjoy.
Intel Next-Generation HEVPD (high-end very-pricey desktop) CPU, code-named "Power-Lake" is fast, 5.5Ghz all-core turbo.
Unfortunately every buyer will need a lake + hydroelectric power plant* to run it, and the Multi-Jigawattz chiller it comes bundled with. (Flux capacitor optional, for the 9999X9X999XX version)
*If not feeling that green, a nuclear reactor will do as well.
As long as you dont try to manually push max OC on a 2990WX.
At the mainstream level, I will take higher clocks over more than 8c/16t any day of the week.