Hardware Intel's Core i7 920, 940 & 965 processors

Nice review Bindi!

I found the power consumption numbers quite interesting especially the load ones. The tests over at lost circuits are done by measuring the power consumption directly (well before the vrms at least) so as to avoid the numbers being skewed by the motherboards. Their results here perhaps suggest that your load results are due to factors other than the cpu itself?

Perhaps the x58 consumes a lot of power which brings me to my next question. The guys at the tech report say here that the x58 is built on a 130nm process. That can't be right can it?

I'm particularly looking forward to the havendale cpus with integrated graphics late 2009/early 2010.

 

Thanks

Technically, that's better for just looking at the single component, however it's not what you pay for at the wall.

No, that's certainly a mis-quote by TR. It's 45nm High-k MG. 1) why go backwards? 2) Intel doesn't have any 130nm fabbing equipment left 3) AMD can't make a quad core work on 65nm and the 65nm quads Intel have here very power hungry even without a northbridge.

Every single slide and white paper I have from Intel pimps its High-k MG process - you can see from the Tick Tock scale too - last year 45nm Penryn, this year 45nm Nehalem, next year 32nm Westmere, year after 32nm Sandy Bridge.

 

True but it gives us an interesting insight into what role the chipset/motherboard plays in the overall power consumption which is not something that to my mind has been adequately investigated or highlighted in the past on most tech websites. I look at the numbers at lost circuits and then the system based numbers at a variety of sites and can't help but think that either the early motherboards are not very well tuned in terms of power consumption or that the x58 is very power hungry which leads me to:

eerrrrrr I think you have the cpu mixed up with the chipset there? I was referring to the x58. I'd be surprised if that were 45nm as if it were then I would have thought that intel would have made a fairly big deal out of it?

 

Thought I'd share the links here and here which show the idle power consumption to be even better than those on lost circuits. How their numbers can be so different when they all measure directly at the vrm stage is beyond me but that sort of discrepancy seems to be par for the course with tech review sites but oh well. Either results are impressive nonetheless! The potential of this architecture in laptops is immense which being primarily a laptop user is very exciting.

Could you just confirm whether you were talking about the cpu or the chipset as per my previous post? It seems pretty clear to me that the x58 is a real power hog despite not having the memory controller built in which seems to suggest either that it's built on an older process or that QPI is actually surprisingly power hungry. Any comments or insight?

 

Sorry.

X58 is 65nm like P45/G45. X38/48 was 90nm.



It's running 3.2GHz QPI links - I doubt 130nm could sustain that efficiently. The fast QPI and 38 PCI-Express x2 links cause it to be "power hungry" according to the Taiwanese.

AFAIK it's got no specific power saving features too, unless mobos include them (MSI for example with its GreenPower), it's just a "dumb" switching chip that's designed to be fast and efficient.

 

I figured 65nm would be the most likely process but nice to have it confirmed. I look forward to seeing what the power consumption of the havendale supporting version of the P55 chipset will be as that will not house QPI but rather just have a DMI link to the cpu as the gpu and the pci-express controller will both be on the cpu package. That should make for some awesome platform power consumption improvements. Just a shame it's so long to wait!

Interesting. Perhaps a bit further down the line there be some more efficient implementations of the X58

 

The last 130nm chipset done waaaas P965/975X afaik. P965 might have been 90.., but I think the 1333MHz FSBs were supported when 90 was introduced so the NB could clock higher.

Nvidia's current 570MCP (that's like 2 years old) is 130nm still (and super hot at that, but it only houses a HT-1 8-bit width link, not "2" 20-bit 3.2GHz QPIs.)

I doubt you'll get power efficient X58s - it's a workstation chipset not a notebook one, performance is more important.

The P55s will essentially be an "ICH" tbh

 

It irritates me when tech sites get details like this wrong, I mean surely he must have thought 130nm sounded odd but he chose to write it anyway. It's like when the 780G was introduced extremetech claimed that the SB700 was 130nm when in fact it's built on 55nm. How does this simple stuff get through the editing process unnoticed?

Yep and that should allow for a nice upgrade to centrino especially taking into account nehalem's ability to power gate the cores at idle. Very impressive technology IMO.

 

Nice review, Q9450 running 7-zip got 9932 MIPS, but on me it gets 12465 MIPS almost 25.6% better with a very shy overclock. ;-)

 

How shy?

 

For the love of god! QPI isn't a GHZ, its a GT/s, fix your charts!

# Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT enabled)
# Core i7 965 (4x3.2GHz, 4.8GHz QPI, SMT enabled)
# Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT disabled)
# Core i7 965 (4x3.2GHz, Turbo disabled, SMT enabled)
# Core i7 940 (4x2.93GHz, 6.4GHz QPI, SMT enabled)