[Discuss] 8 core CPUs, liquid colling

[markw at mohawksoft.com (markw at mohawksoft.com)]

As for servers, I think this part is a win. My only reservation would be
the 2:1 core vs math processor ratio (2 CPU cores per one floating point
unit)

For servers, you need cores. The part has 8 of them, so no matter how well
the 6 core sandy bridge intel part does, this will do 25% better because
it can do more at once. So, while it is a slightly slower in the grand
scheme in many linear operation, the 2 extra cores more than compensate.

I am really excited about this FX processor. I am under NDA with Intel
about some of the features of sandy bridge, Intel has some neat things in
its westmere and sandy bridge processors and I wonder if AMD has them as
well. Its going to be fun watching Intel and AMD compete again.

All in all, I think AMD is right on here. Focusing on more cores rather
than linear performance. Computing speed, today, is seldom a gating
factor. As applications are becoming more "thread based" more cores will
more than overwhelm linear speed. On a sandy bridge 6 threads may complete
in 10 seconds, but on FX with 8 cores, 8 threads will complete 8.25
seconds if you assume that the FX chip is 10% slower on average.

> I see Micro Center is promoting AMD 8-core CPUs starting at $200
> (FX-8120 $200, FX-8150 $260):
> http://www.microcenter.com/storefronts/amd/Expired/FX_assets/fx_index.html
>
> They seem to be marketing them to D-I-Y builders of performance
> desktops, but I'm wondering if these are any good for servers.
>
> Here's AnandTech's review:
> http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested
>
> They weren't very enthusiastic about it, but their benchmarks are quite
> desktop-oriented. (For example, one of their multi-core benchmarks
> depended on the floating point performance, and this chip share one FPU
> per two integer cores, and thus faired poorly.) I'd be curious how it
> benchmarks when running many mundane server loads. And how it compares
> to other CPUs on a performance per watt basis. (This part apparently can
> vary the clock frequency of each core independently.)
>
> Anyone heard how suitable these are for server tasks or seen articles
> that address this?
>
>
> In other matters, the above review also mentions that AMD is now selling
> a rebranded liquid cooling solution, which reminded me of this Antec
> cooler I see advertised at Micro Center:
> http://www.microcenter.com/single_product_results.phtml?product_id=0361180
>
> It's a bit different from traditional liquid cooling solutions in that
> the it comes as a closed system with the radiator permanently attached
> to the CPU block with a fixed length of tubing. Should be about as easy
> to install as a common CPU cooler. At $50 it is also cheaper than what I
> remember a water cooling setup costing.
>
> The description on a similar Corsair cooler says, "a total noise level
> that's lower than most stock CPU fans." Stock? That's not exactly a high
> benchmark.
>
> I think the interesting opportunity with water cooling is that you can
> transfer the heat to a large passive radiator and avoid the noise of a
> fan, but that isn't what this accomplishes. They don't make any
> particular claims about the actual cooling performance, so beyond the
> flexibility of locating the radiator separate from the CPU block (which
> might be useful in a small case), it isn't clear what advantage this
> offers over traditional coolers.
>
> Anyone tried these out? Can you hear the water gurgling? :-)
>
>  -Tom
>
> --
> Tom Metro
> Venture Logic, Newton, MA, USA
> "Enterprise solutions through open source."
> Professional Profile: http://tmetro.venturelogic.com/
> _______________________________________________
> Discuss mailing list
> Discuss at blu.org
> http://lists.blu.org/mailman/listinfo/discuss
>