The Intel Core i9-9980XE CPU Review: Refresh Until it Hertz
by Ian Cutress on November 13, 2018 9:00 AM ESTPower Consumption
One of the key topics in power consumption recently has been whether Intel’s approach to power, or to how it represents its Thermal Design Power (TDP) values, is valid or not. Intel’s take on TDP is that it should represent the sustained power of the processor, which unfortunately does not take into account any of the turbo modes available to the users (or disclose how long those turbo modes should be available for). Part of this is not only confusing, but motherboard manufacturers rarely use Intel specifications for these limits anyway, as you can read in our article covering the practice here.
With the Core i9-9980XE, the typical representation of power is used: stick to the turbo tables unless the system is thermally compromised. In this case the 165W TDP value is a guide, not a limit or a suggestion – it relies on the quality of the silicon and the ability of the motherboard manufacturer to be stable, performance focused, and competitive.
Comparing the Core i9-9980XE to the Core i9-7980XE, the new processor has a higher base frequency by 400 MHz, a higher single core turbo frequency by 100 MHz, and a higher all-core turbo, but uses a newer 14++ manufacturing process and soldered thermal interface material. The peak power consumption numbers are as follows:
Looking at the full chip values, the peak power consumption we observed for the Core i9-9980XE is 192W. This is 9-10W higher than our Core i9-7980XE sample.
If we remove the ‘idle’ power numbers away to see the core-only power, then the Core i9-9980XE uses around 152W just for the cores, which should be around 8.5W per core. The 32-core Threadripper 2990WX by contrast uses around 6W per core.
If we look at the efficiency of each processor, with our power numbers taken during a POV-Ray run:
The Core i9-7980XE gets a performance per watt of 43.3 POV-Ray points per watt - the new Core i9-9980XE scores a little less at 42.7, as for the extra 5% of power, we get a 3.6% increase in performance. For competition, the only HEDT processors coming close are the other Intel HEDT parts, or the 2990WX at the top right of the diagram. Obviously, this is benchmark specific, but an interesting comparison nonetheless.
Facebook
Twitter
RSS
143 Comments
View All Comments
at8750 - Tuesday, November 13, 2018 - link
Hi, Ian.Did Intel officially announce Skylake-X Refresh be manufactured on 14++ node?
But 9980XE Stepping is the same as 7980XE.
Stepping is 4, there is no change.
SanX - Tuesday, November 13, 2018 - link
Sometimes the advantage of these processors with AVX512 versus usual desktop processors with AVX2 is crazy. The 3D particle tests fly like 500 mph cars. Which other tasks besides 3D particle movement also benefit from AVX512?How about linear algebra? Does Intel MKL which seems now support these extensions demonstrate similar speedups with AVX512 on solutions Ax=B, say, with the usual dense matrices?
TitovVN1974 - Friday, November 16, 2018 - link
Pray look up linpack results.SonicAndSmoke - Tuesday, November 13, 2018 - link
@Ian: What's with that paragraph about the Mesh clocks on page 1? Mesh clock is 2.4 GHz stock on SKX, and there is no mesh turbo at all. You can check for yourself with AIDA64 or HwInfo. So does SKX-R have the same 2.4 GHz clock, or higher?Tamerlin - Tuesday, November 13, 2018 - link
Thorough review as always.I'd like to request that you consider adding some DaVinci Resolve tests to your suite, as it would be helpful for professional film post production professionals. There is a free license which has enough capability for professional work, and there is free raw footage available from Black Magic's web site and 8K raw footage available from Red's.
Thanks :)
askmedov - Tuesday, November 13, 2018 - link
Intel is playing with fire by doing incremental upgrades over and over again. Look no further than Apple's new iPads - their chips are better than what Intel has to offer in terms of price-power-efficiency. Apple is going to ditch Intel's processors very soon for most of its Mac lineup.bji - Tuesday, November 13, 2018 - link
Microsoft Windows is known to suck hard when it comes to performance on NUMA architectures and particularly the TR2 processors. See Phoronix for analysis.Why does Anandtech continue to post Windows-only benchmarks? They are fairly useless; they tell more about the limitations of Microsoft Windows than they do the processors themselves.
Of course, if you're a poor sap stuck running Windows for any task that requires these processors, I guess you care, but you really should be pushing your operating system vendor to use some of their billions of dollars to hire OS developers who know what they are doing.
I just bought a TR2 1950X for my software development workstation (Linux based) and I am fairly confident that for my work loads, it will kick the crap out of these Intel processors. I wouldn't know for sure though because I tend to read Anandtech fairly exclusively for hardware reports, dipping into sites like Phoronix only when necessary to get accurate details for edge cases like the TR2.
It sure would be nice if my site of choice (Anandtech) would start posting relevant results from operating systems designed to take advantage of these high power processors instead of more Windows garbage ... especially Windows gaming benchmarks, as if those are even remotely relevant to this CPU segment!
bji - Tuesday, November 13, 2018 - link
Erp I meant 2950X, sorry typo there.Shaky156 - Wednesday, November 14, 2018 - link
I have to agree 110%, gaming benchmarks are more gpu/ipc related more than a multicore cpu benchmark.Somehow seems anandtech could be biased
eva02langley - Wednesday, November 14, 2018 - link
Totally agree, we have come to a time that benchmarks are not even accurately evaluating the product anymore. The big question is how can we depict an accurate picture? Especially if the reviewer is not choosing the right ones properly for real comparison.Well, seeing disparity from Phoronix is raising major concern to me.