power efficiency

Takole, po dooooolgem casu se zopet oglasam

Zanimivo se mi zdi, kake reakcije so bile s trani Intela, na tale clanek.

Intel se je seveda ozval, potem jih je pa dobil po nosu ..... tisto v narekovajih je Intelova zadeva, izven narekovajev pa je odgovor. Branja je mogoce malenkost vec, a se svekakor splaca

I take issue with Intel's response to my firm's latest power efficiency test results. Their comments are:

"The report doesn't measure our latest Xeons, or quad cores"
The Intel 5160 is a Woodcrest dual core chip. It is proper to compare the 5160 to an AMD 2222 dual core. If Intel wants to see quad core test results they can loan me a pair of chips. I will run the test at no charge and publish test results within a week.

"We have 2GHz quad cores in the market at 50 watts, 12.5 per core!".
12.5 watts per core is not the whole story. We are measuring power at the wall including power for the northbridge, RAM, fans, etc. In the near future we will publish a power efficiency comparison of an Intel quad core server to an AMD quad core server.

"The report ignores performance, in that you'd use less Intel servers to get the same job done, meaning less electricity is needed."
This statement is incorrect. First, my report has a section dedicated to comparing maximum throughput and power efficiency at maximum throughput. Second, the test data shows that for this type of complex transaction processing workload, at maximum throughput, the Xeon is up to 14 percent faster than the Opteron for calculation type tasks but it is up to 19 percent slower than the Opteron when disk I/O is the bottleneck. It is wrong to make a categorical assertion that fewer Xeons would be needed.

"We stand behind all of our energy efficient claims, period. For those IT managers who don't do their own in-house testing, we recommend that each look at the 100s of independently verified benchmarks and reviews that exist for the most credible assessment."
I feel that Intel is trying to avoid the issue. Virtually all of those benchmarks were single user, desktop, and calculation intensive tests. Most of them did not even report power efficiency. The issue at hand is power efficiency for a client/server workload. I would be very interested in seeing some test results from Intel that report power at the wall for a transaction processing workload. Anandtech ran a power efficiency test. I believe that their findings agree with mine. http://www.anandtech.com/printarticle.a...


Hehe, prav zanimivo branje ..... po moje odgovora ne bo