Reviews - Intel Core i7 - everything that Core 2 should have been - performance numbe

[Webby]

I will just say that personally (well my brother) has a Phenom running with 1066MHz memory no problem at all. So while I can accept that you may have stability issues perhaps they should be resolved rather than side stepped when comparing architectures? That is meant as a point in general rather than a particular critisism, if you for example were testing crossfired 4870s do you test them in a P35 board or an X48? I would hope the X48 as that would show the true performance. The same I feel is true of the Phenom, while the additional memory speed may not have changed any of the results significantly (with the exception of the memory specific benchmarks) it will have had an effect. I also realise that the Phenom is not the main subject of the review so spending time fiddling to get it working at peak performance was not high on the agenda.

Hopefully that didn't sound to much like I was having a go at you (not meant to be if it does) its just with AMD well behind in the performance stakes already hobbling them a bit more doesn't really seem fair!

[choc]

DDR3-1,066MHz is the limit for the non-EE processors on the Intel reference board, yesm but there's no reason why motherboard manufacturers cannot design their mobos to feature faster-than-default speeds - it happens all the time.

I just emailed the guys at anandtech who used a rather awesome looking Asus Rampage 2 Extreme... that thing is built for overclocking and has some crazy stuff on it. They responded just about as quickly as you guys and say the 920 etc is limited to 1066mhz even on that monster board.

So thats a pretty huge assumption by you guys to publish performance on the chance that some company might one day in the future create a board which can run memory differently... whatever the case it certainly doesn't reflect what will be available to people when the cpus are in stores.

I see you ran the 920 correctly though, why didn't you overclock that system and bench it at 1333mhz memory?

[Tarinder]

But I'm unsure whether anyone has a 'real' '940.

Clocking the '965 down, which we've done, plays havoc with the CPU's EIST.

What we can say for sure is that the '940's performance is between the 920 and 965 EE's. I have no problem in re-running a few 940 tests with memory at 1,066MHz but, really, it won't make much difference to the benchmark scores - perhaps a percent or two.

[shaithis]

Cut them some slack. The architecture is new, it's going to take a while before people get a handle on it's limitations and functionality.

[Tarinder]

I just emailed the guys at anandtech who used a rather awesome looking Asus Rampage 2 Extreme... that thing is built for overclocking and has some crazy stuff on it. They responded just about as quickly as you guys and say the 920 etc is limited to 1066mhz even on that monster board.

So thats a pretty huge assumption by you guys to publish performance on the chance that some company might one day in the future create a board which can run memory differently... whatever the case it certainly doesn't reflect what will be available to people when the cpus are in stores.

I see you ran the 920 correctly though, why didn't you overclock that system and bench it at 1333mhz memory?

Hi,

Please look at the above post concerning the fact that no-one, as far as we know, actually has a proper retail 940 processor, so carrying on this line of thinking all results are kind of invalid now, right?

As I say, we can re-run the 940 with DDR3-1,066MHz and the same timings and see what the difference is.

I will update this section with the results.

[randomgiblets]

Clocking the '965 down, which we've done, plays havoc with the CPU's EIST.

Why on earth do you have it on in the first place?

[Tarinder]

We don't have it on when testing these CPUs but S.Is will, no doubt. The question is how real-world do you want to make it?

I'm more than happy to discuss the merits of testing with you and you can email me - tarinder@hexus.net - at any time.

[randomgiblets]

Thanks Tarinder, you are a nice guy and I really value your reviews on Hexus. I have to be honest however and say I think you dropped the ball with this one man.

Appreciate your courtesy.

[Tarinder]

You're perfectly entitled to your opinion, randomgiblets, but it seems a bit harsh when we're talking about a memory-speed difference on an engineering-sample CPU. I

I've been working on it all night, honestly, so I'm off for some shut-eye. Catch this thread later.

[choc]

But I'm unsure whether anyone has a 'real' '940.

Clocking the '965 down, which we've done, plays havoc with the CPU's EIST.

What we can say for sure is that the '940's performance is between the 920 and 965 EE's. I have no problem in re-running a few 940 tests with memory at 1,066MHz but, really, it won't make much difference to the benchmark scores - perhaps a percent or two.

I think you are right on the availability of 940s but from what we seem to be reading in here there is a right way to fake one and a wrong way to fake one. (Right way = 22x 133 with 1066)

If its wrong its wrong, regardless of the performance impact... you are meant to be telling me what i will get as a user and can we really assume that no benchmark will be affected by more than 1-2%? Memory bandwidth would sure be hit my more than that.

As for EIST... i thought it was common practice to disable things like that for testing to give repeatable results which always use a cpus maximum speed? I know a lot of other sites say they do that.

I'm sorry if i sound harsh but it seems that you are making excuses here for a mistake in your testing methods rather than just admit it was an oversight... anyway i think we know the ins and outs of it now so i'll let you get on with the rest of your stuff i need to hunt down some more reviews and see what others are saying about this cool new cpu. Definitely think i need to buy one of these.

[Tarinder]

Fair enough, point conceded, and the review will be updated with 940 numbers with a 1,066MHz memory speed.

[Biscuit]

DDR3-1,066MHz is the limit for the non-EE processors on the Intel reference board, yesm but there's no reason why motherboard manufacturers cannot design their mobos to feature faster-than-default speeds - it happens all the time.

So if i want to run 1600Mhz ram with a 920 i will have to wait for a board which is capable of doing such a thing or will this kind of functionality be added in a bios update?

[Webby]

I'm not yet convinced it is a motherboard limitation, with the memory controller being on the CPU it could be that it is not possible to change the memory divider at all, Intel have a lot more control now that they can set the supported memory speed in the CPU rather than having chipsets which support a range of memory speeds to allow them to be compatible with a whole host of CPUs.

This could be Intel's way of attracting buyers to the exteme edition CPUs but they will be too costly for most people. Or it could just be them trying to keep the memory voltages within the specified ranges (1.65v max).

[Michael H]

I'm not yet convinced it is a motherboard limitation, with the memory controller being on the CPU it could be that it is not possible to change the memory divider at all, Intel have a lot more control now that they can set the supported memory speed in the CPU rather than having chipsets which support a range of memory speeds to allow them to be compatible with a whole host of CPUs.

This could be Intel's way of attracting buyers to the exteme edition CPUs but they will be too costly for most people. Or it could just be them trying to keep the memory voltages within the specified ranges (1.65v max).

If that were really the case, do you think all the RAM manufacturers would be releasing so broad (and reasonably priced) range of lower voltage DDR3?

Just wait for the retail boards from the big players

[Webby]

I think you are missing my point if the dividers are locked by Intel on the CPU then the is nothing any motherboard can do to change that, much like if you have a non-extreme edition CPU your multiplier is locked, the most extreme of motherboards can do nothing about that, it is hardwired into the CPU and no motherboard settings are going to change it.

Don't forget the extreme edition CPUs have this unlocked so will be able to take advantage of faster memory, and equally subsiquent CPUs may as well. Also plenty of current generation DDR3 motherboards can make use of faster memory so maybe the manufacturers are aiming for them? Look at the available tri-channel DDR3 kits available currently only available in the 1333Mhz or less (on Scan)

[Biscuit]

Well if the case is that the specified memory speeds are limited by the CPU which was my original worry then im deeply dissapointed. I can understand them locking the multi on the chip so that only the extremes can do the serious overclocks but locking the memory seems pretty unfair on the consumer. Intel have too much control over the market atm