Thanks, nice find

You have probably noticed how the Xeon core count increases every year by moving one step to the left/down.

This year E5-2630v3 got two more cores, from six to eight, and E5-2650v3 got a bump up to ten cores from the previous eight. Next year E5-2640v4 will get an increase from eight cores to ten, but the relative results dictated by the two parameters mentioned earlier probably wont change. The E5-2630vX will in all likelihood continue to have best overall purchase and ownership value.

This means that best practice will be to buy the E5-2630 every time it gets more cores. This year seems to be a good time to buy the v3, and then it will probably be best to wait two years until late 2016 or early 2017 when E5-2630v5 comes out with ten cores.

I'm must confess that I'm a bit tempted to buy two new E5-2630v3. I've seen the prices of used Sandy and Ivy Bridge EP on eBay, and I feel they're overpriced. Especially after VAT and toll is added to the US prices. Correct price for a used v2 should reflect a 50% discount compared to an equal core count new v3 model, and a 66% discount for a used v1, again compared to an equal core count new v3. This is not the case and I can only assume the owners of v1 and v2 are extremely satisfied and are not in a rush to upgrade. This is why I own four lazy and sweaty (78C) ass X5650. They emit so much heat the white ceiling above them is now slightly yellowish, by years end probably brown.

That's a no-nonsense workstation. Great processors and even better graphic card. I read you are crunching Einstein, so I assume they support double precision. Testla seems to shine under those circumstances. It will be easier for you to cool this new rig, compared to your old, which radiates 2x150W (I'm surprised the motherboard didn't melt).

Dual processor workstations are practical, but they're not designed to run at full throttle, all the time. More like short burst of intense CAD activities. I've learned this using my HP Z600 workstations. Multi-core processors need very large heatsinks - one kilo metal, each - in order to bring down the temperatures to acceptable levels during long periods of maximum load.

Cramped workstations seldom have room for these type of heatsinks. I think server boards in chimney-like, very large cases is optimal. They work with natural convection, instead of trying to divert airflows into the rear of the computer case, which 99,99% of cases are designed to do. Why? Here is an interesting article about convection and how to design a passively cooled case.

I am interested to know its performance when crunching CEP2 and UGM, as you seem to be doing. How many points it delivers, how many watts it needs and also the temperatures. Especially for CPU0.

Once again, great processors. The next 4-5 years they'll deliver energy efficient high throughput.

TBMS - I decided to build a computer this year, having never built one before. I took a leap of faith based on what a lot of people on this board said and got myself two E5-2680 V3 engineering samples. I got them both from the same guy who sold them to me for what I consider a really cheap price, about $1350 (at current exchange rates) for the pair, which is about the same as two E5-2630V3s.
This was quite a bit cheaper that what I thought I should pay (perhaps the guy liked me).
Anyway, the server board, DDR4 memory also added up as you mentioned.
I was originally looking at V2 because of the cheaper DDR3 ram too but the chip price offset these more expensive components.

So looking at your purchasing decisions and my experience (as well as others), engineering samples are an avenue that I suggest you explore in the future.
I do have to stress that I think I stumbled across someone who genuinely offered me a deal to make V3 affordable. I still think V2 is way more affordable at the moment.

@ Oldchap

Great article and interesting IPC improvement chart. The biggest IPC improvements, those involving a new micro architecture, happens to coincide with E5-2630 having a core count increase. Another reason to buy that processor when it has odd version numbers. I just wish Microway also had a chart for E5-2630L. Those type of processors tend to be very power efficient, but they lag on performance vs price. Question is, how much?

E5-2630L seems to be similar to D-1540, a member of the new Xeon group of chips. Similar price, core count and frequency. The biggest difference seems to be in the L3 cache and TDP.

Here is a quote from your article: "In general, based on the SPEC integer benchmarks, Intel says that the Xeon D chips will overlap between the high-end of the Xeon E3 line and the low-end of the Xeon E5 line. The most powerful Xeon E5, the E5-2699 v3, will offer 5.1X times the integer performance as the Xeon D-1540, just to give a sense of scale."

What is most important here on WCG, floating point or integer? Maybe Xeon D has poor floating point performance?

It's impossible to say how well Xeon D will compare against ordinary Xeons. We'll have to wait and see, but I doubt Intel will allow it to take market share from its more profitable, higher end Xeons. It's most likely intentionally limited in some way. But it compares well against C2750.

"Depending on the workload, the Xeon D-1540 delivers anywhere from a low of 34 percent more performance than the Atom C2750 to a high of 3.4X and at the upper end of that performance improvement, that works out to a factor of 1.7 better performance per watt."



@ vcd683s

That is cheap! A bargain. Where did you find your processors, eBay?


@ twilyth

Thanks for the advice. I didn't know that.


EDIT: Now that I've had time to think more about this new Xeon family, I'm more inclined to say Intel has buried its head in the sand. The 8-core, with its $580 price tag, is probably 30% overpriced. The E5-2630L has four memory cannels and cost the same. The 8-core Xeon D has only two memory channels and also a 40% smaller cash.

This is the same old story, where a market dominant hesitates to face reality, because it would mean parasitizing on its own, more profitable products. So it does what such companies always have done, releasing an inferior or overprized product. Intel needs to cut prices if it intends to fend of the ARM-armada in the low power server market.

They have been down this road before, when they tried to sell the old netbook-based Atom to the tablet manufacturer, but they decided to use ARM-processors instead, and will do so also in the microserver market if Intel doesn't cut the prices. Intel is fat and content, just like Microsoft was a few years ago. Now Microsoft is scrambling - take our shit, Windows 10, it's free - and soon Intel will too. If an ARM-based ecosystem is successfully built in the small server market, it will serve as a foothold for ARM's assault in the traditional server market, Intel's bread-and-butter.