In case you have been living in a cave, SeaMicro makes really dense servers, traditionally with Atom CPUs, but more recently with full fledged Xeons. AMD bought the company earlier in the week, something we think is a good deal. One quarter later, the two announce an AMD part in the middle of what was an Intel stronghold, dense and power efficient servers.
SemiAccurate’s sources on this machine were quite adamant that the chips is a Fusion part, not an GPU-free Opteron, and that leads to some interesting conclusions. First is that they are wrong, and it is an Opteron, or there is simply no such box, but we doubt they are all wrong. Second, we will say that it isn’t a Brazos based part, or Wichita/Krishna for that matter. The latter two are out because they are quite dead, the former is very unlikely because SemiAccurate asked SeaMicro about this last year, and the answer was no. Brazos does not have the performance per watt of an Atom, not even close, and hasn’t been updated enough to matter since. It may crush Atom on GPU performance, but few servers need that, and Llano is a better bet for that market anyway.
So that means Llano in the next SeaMicro box, right? Nope, or probably not, same performance per watt problem vs 45 watt Xeons. What could it be then? How about Trinity, the second generation Fusion part. That chip is much more likely to be the one for SeaMicro, you could possibly fit two 17W Trinitys on a single card too. That would be a quite interesting offering in its own right, and strongly hints that the Opteron/Server Llano part that was knifed when yields were awful is about to make a comeback.
So, we are looking at two options for this Opteron/Trinity, GPU or no GPU. OK, both is a possibly, but we can’t see any sane buyer picking this over the Xeon version if they didn’t want the GPU. AMD may want a flagship, but they wouldn’t pay for the engineering, or drop a profitable line, if they didn’t have a realistic replacement. Money in the door is money in the door, and the new order in Sunnyvale is a lot more finance focused than the old.
This once again strongly hints at another interesting possibility, one of the big server players has found a good use for GPGPU in the server space. OpenCL + Google/Facebook/Amazon has some tasty implications, no? Hadoop + tons of low latency threads = good thing. We will go out on a limb here and say that AMD and SeaMicro would not be doing such a box if someone didn’t want to pay for it, and didn’t need a GPU. Our sources could not shed light on either possibility, but outfits like Onlive suddenly become potential customers too. There is a lot of sales potential there.
Where it will end up, we can’t say for sure. Some of the bits we know are real strongly hint at a specific direction, but the rest is speculation. Who knows what the AMD/SeaMicro box will end up looking like, and who will be buying it? Given the reticence of all the big server players to talk about their infrastructure, we may never know. That said, you might even end up using a few without ever knowing it, funny how that works out.S|A
Latest posts by Charlie Demerjian (see all)
- Thing go bump(gate) in the night for Nvidia’s GP100 Pascal GPU - May 3, 2016
- Cavium’s Octeon TX blends compute and packet moving - May 2, 2016
- Qualcomm releases Zeroth API to developers - May 2, 2016
- Another detail about Qualcomm server SoCs revealed - Apr 27, 2016
- AMD finally really honestly launches the dual Fury - Apr 26, 2016