The Intel Core i9-9990XE Review: All 14 Cores at 5.0 GHz
by Dr. Ian Cutress on October 28, 2019 10:00 AM ESTPower Consumption, Frequencies, and Thermals
Across several articles we have covered why TDP numbers on the box are useless for most users: the loose definition of Intel’s TDP is that it represents the cooling required for the processor to run at the base frequency. ‘Cooling Required’ is a term referring to the power dissipation of a cooler, which isn’t strictly speaking the same as the CPU power consumption (because of losses), but close enough for our definitions here.
For the Core i9-9990XE, that means that when all 14 cores are running in a normal configuration at 4.0 GHz, with no turbo initiated, the CPU is guaranteed to be running at 255W or less. However, in our case, ICC has pushed the processor up to its turbo speed, 5.0 GHz, for an effective ‘infinite’ time. This means we never see 4.0 GHz, and only ever see 5.0 GHz.
In our testing, ICC did at least have some form of ‘Turbo’ enabled, which meant that the chip could run in idle states. At idle, the system would run at 1.2 GHz, but still at the same 1.29 volts that the chip was set to. This lead to a full-system idle power of 266W and a load temperature on the chip of 24C in a 20C ambient room. Unfortunately we could not measure the chip power directly due to some quirks of how Intel manages the power readouts in software. We were able to detect the mesh frequency at idle, which was 900 MHz.
When running a fully multithreaded test, such as Cinebench R20, the fact that every core hit 5.0 GHz was easy to detect. With the advent of features such as Speed Shift, Intel aims to get the CPU from idle to 5.0 GHz as quickly as possible. During a sustained CB20 run, which is possible through the command line, we were able to observe a peak power consumption of the system at 600W, which indicates that at 5.0 GHz this CPU is pulling an extra 334 W over idle – this power naturally being split mostly to the cores but some will be for the mesh and some will be in the efficiency of the power delivery. At full speed, the mesh will rise up to 2.4 GHz.
Naturally, fitting this into a 1U system requires the substantial cooling we described at the beginning – as this cooling is running at full speed even when idle, it doesn’t affect the power consumption when we ramp up the workload. But tying into the temperature, the internal sensors indicated a 81C peak temperature, while still at 1.290 volts. For a 14-core 5.0 GHz CPU, that’s pretty amazing.
For the audible testing, this thing is loud. With ICC’s proprietary liquid cooling solution, in such a small 1.75-inch form factor, in order to take care of those 350-400W that the CPU could draw, nothing short of some fast flow and high powered fans would suffice. This system runs the cooling at full speed both in idle and at full load, which in this instance measured a massive 78 decibels at only 1 ft (30cm) from a closed system. The fact that this is in a 1U form factor should give you an indication that it should be in a rack in a datacenter somewhere, and not in the office. I am not so lucky, and I was only able to perform testing on the system when everyone in my family and next door was out during the day.
We did some testing with AVX-512 tests. The CPU in this instance only hits 3.8 GHz when at full speed, indicating a -12 offset. It would appear that Intel, while pushing the single core frequency through binning, didn’t so much take into account AVX-512, or at least hoped that it would also be as efficient. In this mode we saw the same power consumption at a system level of around 600W, however the CPU thermals did rise slightly to 82C.
Due to the limitations of the motherboard in the system, which was locked down by the system provider, we were not able to attempt additional overclocking. That being said, I’m sure that the OEM partners and system integrators would prefer it if end users did not perform additional overclocking, lest this MSRP-less ‘no guarantee of any more chips’ processor actually bites the dust.
Facebook
Twitter
RSS
145 Comments
View All Comments
Batmeat - Monday, October 28, 2019 - link
Agreed.This chip is auction only. Expect to pay HUGE DOLLARS for this assuming you even have access to the auctions.
mrvco - Monday, October 28, 2019 - link
But the chip isn't selling for all that much at auction apparently... I was expecting much more than 2849 euros if this thing really is the golden ticket for HFT. From a financial perspective, this isn't worth Intel's time or effort relative to letting top-tier partners and resellers buy up the whole supply of 9990XE's to do the binning on their own. Regardless, it's less expensive than a Super Bowl ad I suppose and probably more effective considering their target audience.Cygni - Monday, October 28, 2019 - link
It is a niche product for a niche market... one that the article talked about at length. "Bragging rights" doesn't come into play when it is a tool for making money.willis936 - Monday, October 28, 2019 - link
It just doesn't make sense. If single threaded performance is king then why do you need 14 cores running at 5 GHz? If multithreaded performance is king then why not go wider? The low latency case doesn't make sense. You could assemble multiple systems focused on single threaded performance for less money than this 14 core auction-only chip costs. When people say it's only for bragging rights, they are not wrong.edzieba - Monday, October 28, 2019 - link
To GREATLY oversimplify HFT loads:you want as many cores as you can get, but those cores must respond as quickly as possible. You're effectively packet-watching: as soon as you see a packet you need to read it, determine if it is to be acted upon, determine how it is to be acted upon, and then respond, and you need to do it faster than everyone else. Everyone else has as close as is possible to the same network latency (e.g. stock exchanges employ huge fibre loops to ensure every endpoint has the same light-speed lag), so if you can run at 5GHz vs. 4GHz of your competitors you can respond to any given packet before they can. It's only a hair faster, but if you're jsut barely first you're still first so your transaction request is the one the stock exchange acts upon and not everyone else's.
You want more cores because each core can run its own worker with its own algorithm (or more workers on the same algorithm to offset by packet arrival). You never want to be operating at a lower frequency than everyone else because it means NOTHING that you have 64 cores if every one of your cores is consistently too slow to beat out everyone else.
You want all those cores on one die because you need to remain consistent in response (i.e. not have one worker working at cross-purposes against another) and inter-socket or worse inter-machine latency will kill you dead in the race to respond.
Processwindow - Monday, October 28, 2019 - link
If high frequency is the ultimate goal and money is not an issue, why staying water.cooling.and not going straight to LN2. LN2 is cheap by wall street standards and it would allow to go higher than 6 GHz for sure. Btw in the great explanation you give regarding the need of high frequency CPU in HFT , I see nowhere optimization of the network card whatever it is and of its firmware.JoeyJoJo123 - Monday, October 28, 2019 - link
>If high frequency is the ultimate goal and money is not an issue, why staying water.cooling.and not going straight to LN2. LN2 is cheap by wall street standards...These kinds of machines are used by automated systems doing stock trading thousands of times per second. (It's why being a day-trader is absolutely worthless because automated software can do your task thousands of times faster and more accurately given historical trends.) LN2 isn't "expensive" but this is a machine that needs to have the highest single-threaded performance possible, with as many cores as possible, yet still run 24/7 to keep up with the market. The system can never sleep, as we're talking potentially hundreds of thousands of dollars lost per few seconds of downtime. And that's why LN2 isn't used--It evaporates and while LN2 cooled systems can overclock higher than non-LN2 systems, it's at such a bleeding edge of instability that it in-and-of-itself will cost the stock-trader money whenever it inevitably gets a cold-bug and needs to reboot or if the thermal transfer between the pot and the IHS cracked at ultra-low temperatures and the temps are starting to rise, or if there's condensation around ICs, or if the power delivery system is starting to fail because it's been ran over-spec for the last 50 days, etc.
Processwindow - Monday, October 28, 2019 - link
What are you talking about? MRI tools in every big hospital in the world is working 24/7 with LN2 . Every semiconductor fab in the world use LN2 in manufacturing environnement. LN2 is not an exotic material.So you tell me you can spend 10s millions of dollar to gain a few ns in HFT but you don't want to go beyond water cooling to gain a few GHz on your CPU Something is wrong here.
Opencg - Monday, October 28, 2019 - link
You clearly don't understand the difference between those machines and these. You just don't have a mind for it. Do not try to argue. Don't even try to think about it man. You are useless.The difference is that those machines were designed with LN2 cooling in mind for sustained operation. Computers were not. Having a skilled overclocker precisely control a benchmark for a (relatively) short time is nothing like having a machine designed to automatically consistently do it over long periods. Developing a computer that could do his would not only be VERY expensive but there is also a risk of it simply not working consistently enough in the end anyway and you are back to square one.
Processwindow - Monday, October 28, 2019 - link
You are the one clearly not understanding the situation. It seems financial companies involved in HFT can spend 10s millions us$ to gain a few ns . But they wouldn't want to spend those same dollars to get 1 or 2 more GHz with special custom LN2 or whatever other ultra low temp cooling system? Something doesn't add up here.