In this ATC Insight

Zuta_Cool

When making a decision on how to cool the next generation of processors, look no further than ZutaCore's two-phase liquid cooling solution. This product brings so much to the table. Not only can it keep your processors "Cool as a ZutaCore Cucumber," but it can also help improve performance and reduce server energy consumption. 

We tested the product in a technical Pepsi challenge. The test included 32 energy-hungry Dell R650s with dual Intel Xeon Platinum 8368 processors that pull 270 Watts each. We measured the power consumption with the servers while running a 100 percent CPU load for 24 hours while air-cooled. Then they were switched to liquid cooled and the tests were re-run. The results were interesting and certainly not what we expected. 

Performance

During testing for the air-cooled systems we noticed a measurable performance hit in the processors when the CPU load was driven to 100 percent. We did not see any performance degradation when testing with ZutaCore-cooled servers.

The way these processors work is they maintain a steady frequency until the temps reach a certain number. When the processors get too hot they turn down the frequency. When this happens it reduces the power draw which allows the chip to cool. When the temps get back down to normal, performance is restored. 

Energy consumption

The ATC has conducted testing with ZutaCore on both Cascade Lake and Ice Lake processors. The results on the Cascade Lake processors showed noticeable power savings. See my previous article for more on Cascade Lake. 

The Ice Lake processors performed differently than the Cascade Lake. For starters, they use 65 more watts of power. If you multiply 65 Watts by 2 (the number of processors in the servers) and then by 32 (the number of servers), that is a lot more heat to be generated and a lot more opportunity to improve energy consumption. We expected to see greater than 7 percent gains this time, but we found that the ZutaCore cooled system actually used more energy than the air cooled. 

These surprising results were discovered the day before a presentation which threw a wrench in our message. Why were the numbers off? This is where WWT, Dell and ZutaCore dug in to figure out why.

BOMs are important

It is hard to understate how important this next piece of information is when making a vendor selection. The ZutaCore and Dell teams on this project worked together to find the source of the problem. They did not blame the other vendor. There was no yelling. There was simply a search for a solution to the problem. They were incredible to work with on this project. 

At the time of this article, there was a Dell hardware component that reduced the efficiency of the product. This component had a performance minimum that didn't allow the ZutaCore solution to perform at its best. It effectively disabled the power savings that the ZutaCore solution was supposed to provide. As a result, the ZutaCore-cooled product actually used more energy than the air-cooled system. 

How do you avoid using this component? Reach out to your WWT solution experts to help build the proper bill of materials. 

Results

The executive summary of this test is that optimal performance was maintained with the ZutaCore cooled servers which did allow them to consume more power. The air-cooled factory setup servers took a performance hit every time the processors got too hot. While we did not see a reduction in energy consumption for the liquid servers this go around, we expect that will change in our next round of testing.

Next steps

The next round of testing will be performed on Dell servers with Sapphire Rapids processors. We will also perform similar tests with new AMD EPYC processors. All tests will compare ZutaCore and air-cooled on Dell servers. We are very excited about the next set of tests and will have another article with all the details. More to come; stay tuned!

Technologies