Slashdot Mirror
New Xeon CPU Hot and Underpowered
Kasracer writes "Web Sites The Register and GamePC received several of the new dual-core processors from Intel, dubbed 'Paxville', and ran a battery of tests on them. What did they find? From the article: 'There's no doubt about it, Intel's dual-core Xeons are their most power hungry Xeons to date ... Even when idling, two dual-core Xeons consume nearly 400W of power at any given time, which is amazingly high, even by Intel's standards ... their new dual-core chips (while powerful in their own right) simply are bested across the board by AMD's dual-core Opteron processors. Even worse, the Opterons typically perform much better while running at slower clock speeds and only having half the amount of on-die L2 cache to utilize.'"
Apple will most likely be using Pentium M's, as they are currently planning on phasing in the Intel chips from the ground up, whilst leaving the Power Macs to run on the G5's for the moment IIRC.
So there is a good chance they already know that Intel has something far better in store for them to use in 2007/2008
This new xeon Chip is sucking down more juice than the three macs I have In my house , It is 100W less than my server and PC
The only things certain in war are Propaganda and Death. You can never be sure which is which though
So you're saing having a chip that uses 50% more power than its direct competitor and doesn't even beat it in a single benchmark isn't terrible? The Opterons beat the Paxvilles in every single test. Are you reading the same articles I am?
It's less a matter of design flaws as marketing flaws, I think. The P4 sucks in a lot of ways, but it's also very well-done in a lot of ways. The big problem is not so much netburst as it is Intel's inertia. Intel's working on a way to transition their product matrix from netburst on the desktop and in server space to something based on the pentium-M. The faster pentium-Ms beat the fastest P4s hands down in most benchmarks, but there are no good desktop chipsets for it yet, etc. Eventually Intel will release a dual-core P-M-based design, hopefully with an on-die memory controller, and then the Opteron will finally have some real competition.
Even if Intel eventually hits a process wall, they'll still be able to rest on their huge manufacturing capacity. For the past 5ish years, Intel has been building bigger chips than AMD, mostly in terms of L2 cache. It's not necessarily the best way to improve performance, but it's fairly easy, and leverages Intel's manufacturing strengths. Intel can afford to crank out dies 50-100% larger than an equivalent AMD die, and still make money at it, and still not run out of capacity. The reason Intel has been having shortages of late in the chipset arena is not lack of capacity but bad capacity management. They mis-read the market 3ish months ago and are paying for it now.
Anyway, here's hoping AMD's 65nm transition goes as smoothly as their 90nm one.
High-speed Road Trip (18.000KPH)
As a small-time, low power switch-mode power-supply designer, I would like to object that it is very difficult to accurately measure the power draw of a modern desktop CPU, unless you design a whole motherboard specially dedicated to power measurements.
;-)
These beasts have peak current draws in the range of 100A @ 1.5V.
If you were to allow for a maximum of 1% voltage loss (15mV) across a measuring shunt resistor, this would mean 0,00015 Ohm Resistance - the equivalent of a piece of copper Wire with a cross-section of 1.0mm^2 and a length of 7mm. Good luck attaching a 1% accurate kelvin sense connection to this. You will also have to design a high performance multiplier circuit to make accurate RMS power measurements in the MHz Range.
But even then, and not considering mechanical difficulties like buried traces/planes, you cannot place this shunt betwwen the cpu and the Power supply (i.e. 1.5V switcher and bulk 1.5V capacitors), because the added inductance and skin effects would probably cause the CPU to malfunction.
So your best bet is to place shunts between the multiple switching coils and 1.5V Caps, which would probably work. But then you are only measuring the sum of the CPU current draw + the dissipation losses in the traces and the capacitors - which arent neglible, as a lot of people learned recently, as the overheated low-quality organic caps died on their motherboards.
But what is it worth ?
As a computer user, I care only for four things :
- The reliability
- The noise coming out of the box
- The cost of the power going into the box
- The cost for extra air-conditioning, or the savings on heating -
depending on the season and where you live
For all of these, the amount of power drawn from the wall plug is a very suitable indicator.