Asetek Liquid Cooling Now in Servers: Part 1

You may have read our press release about Asetek entering the server space. Over the years I have received many questions asking why Asetek is not very visible in the server space like we are in pretty much every other segment in this industry.

Let me start with a small anecdote; some 5 years ago I was doing a presentation for one of the world’s biggest OEMs on our integrated liquid cooling. There were probably 20 persons in the room and 10 of them were from server divisions and the remaining from the desktop side. There was a lot of interaction underway from all divisions. After my presentation, I received really good feedback on our technology and one of the server VPs approached me and said – “Andre, this is a really cool (no pun intended) solution you have. Come back to me when you have a million units in the field and 5 years of statistics”.

Wall Street High Frequency Trading There you go – that’s the answer to why Asetek has not been very active in the server market until now. This market (compared to the desktop market) is, for good reasons, much more conservative and bases its decisions on many other aspects than just performance, such as reliability and TCO. Also, if you have read “Crossing The Chasm” you would know that as a small company you want to start with enthusiasts and early adopters of technology. Asetek IS now at a place where we have more than a million(!) units in the field, we have years of track record from different segments and applications, and we have very good statistics around our quality level and reliability. 5 years ago we had PowerPoints and product life time statistical calculations.

Another thing has changed which makes the opportunity worth going after for us – we now finally see a meaningful need (in human terms: a valuable problem we can solve). We think there are three categories we can play in: 1. Performance 2. Density, 3. Power savings, and of course any meaningful mix thereof.

If we start with the performance play – as you can read in another blog post of mine, overclocking Data Center Rack Densityis actually moving in to the server space. Beyond IBM offering it, we have received exactly the same request from more or less any server OEM out there. In all modesty we think we have a very good grasp on overclocking and consider it a given to play in this space and we have developed a great solution for the Wall Street High Frequency Trading market called WS-HFT (do the math). We have been able to accelerate 4 Xeon CPUs to North of 4.2GHz and the basic technology behind it is our integrated liquid cooling. The extra cost of our liquid cooling has earned itself in on one day of trading. Our solution can be adopted by any OEM or SI.

Moving to the density play, most data centers have lots of racks that are only partially filled. The reason is that most data centers can’t get enough cold air to the racks to keep them cool if they put in more servers. With our high efficiency internal loop liquid cooling solution we can cool the same server with less air. Above we discussed the idea of using the extra efficiency to take a server that only supports 95W CPUs today and populate it instead with 130W CPUs. Another way to look at this is we are using the same amount of air to cool 70 extra watts (2 CPUs per server). Another way to use this extra cooling power is to continue to use 95W processors and cool more servers per rack.

For example, with our solution we can enable 4 high end CPUs in a 1U server and populate the full 42U rack. Think about it – 168 high end CPUs in one rack. Depending on your favorite CPU, you can then multiply each CPU with 6 or 12 cores – that’s 1008 or 2016 cores in one rack! Highest end Sandy Bridge E CPUs are 150W. That’s 25Kw of power just for the CPUs, and we are going to show you that later this year…! If we, in rough terms, assume that under load the CPU power is 60% of the servers’ total power usage we are approaching 50kW racks.

So what’s the trick? The trick is that because we can cool the CPUs more effectively (to lower temperatures) they no longer need the same amount of cold air as with old fashioned cooling methods – simple as that. Like in the above example, our solutions can be implemented in industry standard hardware. Stay tuned for Part II of this blog focusing on energy savings and our RackCDU™ solution.