smilingcrow wrote: stevea wrote:
One problem with these new dual cores is that if you use them with a P55 board you end up with a 32nm dual core that consumes up to 10W more at idle than a 45nm quad on the same board. I wonder how a H55 board with a discrete graphics card compares when using an i3 versus i5-750?
You are making a strawman argument. If you don't want the on-chip GPU then it is ridiculous to choose the i5-661 w/ higher powered GPU, the i5-660 is a much better choice. 14W Lower TPD (73W vs 87W) and more CPU features. The -660 GPU is slower 733Mhz vs 900Mhz).
I didnâ€™t specifically mention the i5-661 so Iâ€™m not clear why you make that argument!
I was thinking about ANY 32nm dual core versus an i5-750 and in conjunction with an entry level VGA card.
Iâ€™d hate to add around 10W to idle power consumption and donâ€™t want to be forced to use an IGP.
You claim these new (i.e. 32nm) duals use 10W more at idle than a 45nm i5/i7 on the same P55 board. The only evidence I see is the Tom'sHW report (on an H55) where they explicitly say the Asus board may be the cause.
Here is a specific report on this issue: http://techreport.com/articles.x/18216/15
"First, when installed in our P55 motherboard, the Core i5-661 had the same power draw at idle as the Core i5-750". This on a Gigabyte P55A-UD6. Obviously the i5-660 would be lower power and same performance in this case.
This appears to show the 32nm system w/ external graphics drawing 6W LESS at idle w/ the -661. http://www.silentpcreview.com/article1013-page5.html
I think you are wrong about the idle power differential, and in any case the -660 should use less power than -661; -661 is NOT a valid reference point for a "not-using-IGP" test. Have any evidence for your "10W worse" claim ?
The IGP has always been â€˜freeâ€™ in the sense that integrated chipsets have never been expensive so I donâ€™t see having it on the CPU is a positive from a financial point of view.
Of course these aren't free, and they have real value. Putting together a headless server, only a few BIOS's will support serial console, so you generally need some minimal low power graphics for boot to succeed. The cheapest separate cards are ~$40 and eat up an expensive PCI-X slot resource and at least 12W at idle. Mobos with integrated graphics (say G35 & G45 and some interface) cost more than their no-vid counterparts by about a $40 margin and use several extra watts. A development station with an i3/i5, IGP+H55 can be cheap, middlin' high performance, modest graphics and low power.
The 32nm process gives the power savings over Lynnfieldâ€™s 45nm process rather than the chipâ€™s design per se. I donâ€™t think itâ€™s any more advanced than Lynnfield in that sense.
Not more advanced, just the opposite, at same clock rate and <= 2 threads the clarksdale has decided architectural disadvantage compared to the lynnfield, (but you get more Ghz/$ on Clarksdale). OTOH Clarksdale has a big advantage over wolfsdale. In isolation of other design criteria this is a pointless fact.
You have to consider all the relevent factors TOGETHER - total price, performance, power, vs requirements. For $200 you can get a E8500, i5-660 or an i5-750. E8500 is the low performance chip among these. The -660 and the -750 have very similar performance until you get to trancoding and multi-threaded benchmarks, at which point the i5-750 runs away; but how many ppl need that ?. Despite your objections the -660 or even -661 is the low power solution, followed by the i5-750 and E8500. Of course IF you use the -660 IGP then the -660 system power is unchanged. For the E8500 there are a lot of on-board solutions for perhaps 5-6 watts at idle. For the i7-750 there are no on-board GPUs so you lose a PCI-X slot, $40+ and at least 12 watts at idle. Even having an extra PCI-X slot unpopulated costs power & money and you NEED 1+ extra for the i5-750.
Clearly an i5-660 system where the IGP graphics are sufficient and where you only need 2 cores (can't use 4) will be cheaper and lower power than either i5-750 or E8500 system. It's performance will be comparable to the i5-750 unless/until you can use the extra cores.
A design that increases memory latency, reduces performance (compared to Lynnfield), increases idle power consumption when discrete graphics are used, increases power density on the CPU die and reduces choice (i.e. Lynnfield canâ€™t use an IGP) is very flawed from my perspective. It makes more sense for the mainstream market but thatâ€™s not my bag.
No it's not "very flawed". You lack perspective. The -660/-661 has faster memory than wolfdales, tho' inferior to the lynnfield. It has performance comparable to same-price Lynnfields except when you can use the extra 2 lynnfield cores and you get the IGP on a similar power budget. Anyone choosing this chip with use of "discrete graphics" is probably a fool. It's a bogus comparison. If you can't use the IGP then it's EXTREMELY likely this is a poor choice for you. The i5-750 would almost always make more sense in that case. If you can use IGP it may be a big win depending on other requirements.