4GHz i7 920

[Foxery]

Originally posted by: Idontcare
http://aceshardware.freeforums...ghal-nehalem-t628.html

As you can read in that thread, Johan kindly swept aside the suggestion this could be an issue.

His speculation on how this issue is no longer an issue is contained in that thread.

It sounds like he didn't think about it very hard, nor ask anyone for a better answer. Or maybe they're just dodging the issue publically until they hammer it out with Microsoft...

As mentioned in some other thread today, this issue also screws with power management because the cores can never stay idle long enough to Sleep. Johan's best answer is that it's okay because they can power up and down really, really quickly? I fail to believe Intel's engineers are that intensely stupid.

 

[Idontcare]

Originally posted by: Foxery

Originally posted by: Idontcare
http://aceshardware.freeforums...ghal-nehalem-t628.html

As you can read in that thread, Johan kindly swept aside the suggestion this could be an issue.

His speculation on how this issue is no longer an issue is contained in that thread.

It sounds like he didn't think about it very hard, nor ask anyone for a better answer. Or maybe they're just dodging the issue publically until they hammer it out with Microsoft...

As mentioned in some other thread today, this issue also screws with power management because the cores can never stay idle long enough to Sleep. Johan's best answer is that it's okay because they can power up and down really, really quickly? I fail to believe Intel's engineers are that intensely stupid.

I see you have precisely the same impressions from that thread that I had when I finally just walked away from it. It became painfully obvious I was wasting my time.

At any rate we will probably know nothing about it until some forum users start getting the chips in their hands.

Consider the fact no one bothered (Anand, etc) to write up an article about how Santa Rosa fails to enact its turbo mode, but bryanW1995 proved it for us here on the forums. We are pretty much on our own here.

 

[Enthusiast]

the everest results make me drool. i can't wait for these to come out

 

[GregK]

Originally posted by: rge
How do you get exactly 4010 mhz from integers 133 or higher multiplied by integers 20 to 30 (or even with half multi's). I cant get there in excel. Also how did he get 267 stock, is that 133.3?x20, 266 would be 133x20. I can get 4011 using 21x191 (if 133) or 4010.6 using 22x182.3 (if 133.3), but he is showing 4010. Cant seem to get any closer, especially not using higher multipliers.

Edit: Found it, if it is actually 133.4, which explains the 2.67, and 25x160.4 is only way to get exactly 4010, and makes sense with those claiming only 30mhz increase so far on the QPI.
http://diy.pconline.com.cn/cpu...s/0809/1426985_10.html

Hold on... how is this possible?

Core i7 920 has a maximum 20x "normal" and 21x "Turbo" multiplier

920's top multi is 20, and 940's top multi is 22. How's the reviewer getting 4010 with a multi of 25 on i7 920?!!! Either the multi is unlocked or there's got to be an electrical variation that's causing the 4010 to show up, but it's within a tolerance error of, say, 4000 (200x20). But 10Mhz is a LAAAARGE variance. I've never seen that, so that leaves an open question... how the ____ did they do that on a locked chip, unless it wasn't a locked chip?

"4010.6 using 22x182.3" sounds reasonable on i7 940, but the screens clearly say "920" on CPU-Z

and then one has to consider memory multi's. some mobo's have 6x,but most have 8x. So 200x8 = 1600Mhz memory.

Some variations that are close:
211.1x19
250.6x16
267.3x15
From what I hear, the 920's tolerance is 180-190Mhz, so I'd be shocked if any of the above were stable or even worked. Even at 190, the voltage is so high, it shortens the CPU lifespan. So 4010??? Sounds like a false bench. Just because someone has 1 CPU that did that, it doesn't mean anyone else will be able to repeat that feat on their run of the mill 920's.

Question for pro's: on Yorkie's, a higher bus speed with lower multi yields a faster processor than lower bus with higher multi at same speeds. So 400x8 is faster than 320x10. Same memory speed (via adjusting memory multi) of course. Does the same phenomena apply to i7 lineup and bclock?

 

[Idontcare]

Holy thread-necro :shocked:

I would take the the "review" discussed here with a grain of salt now that we have legitimate reviews on the web. This was one of those leaked situations which we had no choice but to salivate over back in Sept because there was practically nothing else going on. Now we have the real deal, no point worrying about figuring out why leaked benches don't make sense or were fraudulent in some regards.

 

[exar333]

Originally posted by: GregK

Originally posted by: rge
How do you get exactly 4010 mhz from integers 133 or higher multiplied by integers 20 to 30 (or even with half multi's). I cant get there in excel. Also how did he get 267 stock, is that 133.3?x20, 266 would be 133x20. I can get 4011 using 21x191 (if 133) or 4010.6 using 22x182.3 (if 133.3), but he is showing 4010. Cant seem to get any closer, especially not using higher multipliers.

Edit: Found it, if it is actually 133.4, which explains the 2.67, and 25x160.4 is only way to get exactly 4010, and makes sense with those claiming only 30mhz increase so far on the QPI.
http://diy.pconline.com.cn/cpu...s/0809/1426985_10.html

Hold on... how is this possible?

Core i7 920 has a maximum 20x "normal" and 21x "Turbo" multiplier

920's top multi is 20, and 940's top multi is 22. How's the reviewer getting 4010 with a multi of 25 on i7 920?!!! Either the multi is unlocked or there's got to be an electrical variation that's causing the 4010 to show up, but it's within a tolerance error of, say, 4000 (200x20). But 10Mhz is a LAAAARGE variance. I've never seen that, so that leaves an open question... how the ____ did they do that on a locked chip, unless it wasn't a locked chip?

"4010.6 using 22x182.3" sounds reasonable on i7 940, but the screens clearly say "920" on CPU-Z

and then one has to consider memory multi's. some mobo's have 6x,but most have 8x. So 200x8 = 1600Mhz memory.

Some variations that are close:
211.1x19
250.6x16
267.3x15
From what I hear, the 920's tolerance is 180-190Mhz, so I'd be shocked if any of the above were stable or even worked. Even at 190, the voltage is so high, it shortens the CPU lifespan. So 4010??? Sounds like a false bench. Just because someone has 1 CPU that did that, it doesn't mean anyone else will be able to repeat that feat on their run of the mill 920's.

Question for pro's: on Yorkie's, a higher bus speed with lower multi yields a faster processor than lower bus with higher multi at same speeds. So 400x8 is faster than 320x10. Same memory speed (via adjusting memory multi) of course. Does the same phenomena apply to i7 lineup and bclock?

Where does FUD like this come from? I am running 180 at slightly over stock voltage on air. Don't post about what you have'nt used if you say things like this. Jeesh...

 

[JackyP]

Originally posted by: Idontcare

Originally posted by: Foxery

Originally posted by: Idontcare
http://aceshardware.freeforums...ghal-nehalem-t628.html

As you can read in that thread, Johan kindly swept aside the suggestion this could be an issue.

His speculation on how this issue is no longer an issue is contained in that thread.

It sounds like he didn't think about it very hard, nor ask anyone for a better answer. Or maybe they're just dodging the issue publically until they hammer it out with Microsoft...

As mentioned in some other thread today, this issue also screws with power management because the cores can never stay idle long enough to Sleep. Johan's best answer is that it's okay because they can power up and down really, really quickly? I fail to believe Intel's engineers are that intensely stupid.

I see you have precisely the same impressions from that thread that I had when I finally just walked away from it. It became painfully obvious I was wasting my time.

At any rate we will probably know nothing about it until some forum users start getting the chips in their hands.

Consider the fact no one bothered (Anand, etc) to write up an article about how Santa Rosa fails to enact its turbo mode, but bryanW1995 proved it for us here on the forums. We are pretty much on our own here.

When I asked on RWT if this might influence the power management of nehalem, I was told it wouldn't. IIRC it was DKanter who said that, so I believed him. That's pretty strange...

 

[Idontcare]

Originally posted by: JackyP

Originally posted by: Idontcare

Originally posted by: Foxery

Originally posted by: Idontcare
http://aceshardware.freeforums...ghal-nehalem-t628.html

As you can read in that thread, Johan kindly swept aside the suggestion this could be an issue.

His speculation on how this issue is no longer an issue is contained in that thread.

It sounds like he didn't think about it very hard, nor ask anyone for a better answer. Or maybe they're just dodging the issue publically until they hammer it out with Microsoft...

As mentioned in some other thread today, this issue also screws with power management because the cores can never stay idle long enough to Sleep. Johan's best answer is that it's okay because they can power up and down really, really quickly? I fail to believe Intel's engineers are that intensely stupid.

I see you have precisely the same impressions from that thread that I had when I finally just walked away from it. It became painfully obvious I was wasting my time.

At any rate we will probably know nothing about it until some forum users start getting the chips in their hands.

Consider the fact no one bothered (Anand, etc) to write up an article about how Santa Rosa fails to enact its turbo mode, but bryanW1995 proved it for us here on the forums. We are pretty much on our own here.

When I asked on RWT if this might influence the power management of nehalem, I was told it wouldn't. IIRC it was DKanter who said that, so I believed him. That's pretty strange...

We know thread migration is real. And we know stepping a core thru power-management cycles requires a non-zero time delay to complete.

So we can logic that in no way is it true of anyone to claim there is zero impact to performance and/or power management. We know there is a non-zero impact here, the question we are all left to answer for ourselves is the magnitude of the impact.

The issue I encountered on the aces thread was the wholesale dismissal of the very topic itself. Which is worse than simply getting no response. In your DKanter may have said "no impact" while thinking in his mind "well there is some impact, but it is mostly negligible, mostly". You never know the context of thinking process when assessing other's responses second-hand.

For santa rosa, bryanW1995 showed the magnitude of the non-zero impact to actually be so large as to negate the entire feature. At best he could coax the higher multiplier out of the system for just a fleeting instance thanks to thread migration.

Now the PCU on Nehalem may very well reduce this impact from the 100% we see on Santa Rosa down to something imperceptible except by the raw benchmark numbers, but anyone who says the impact is zero or outright denies the existence of the phenomenon is usually immediately binned to the "wasting my time" category.

 

[GregK]

Originally posted by: ExarKun333

Where does FUD like this come from? I am running 180 at slightly over stock voltage on air. Don't post about what you have'nt used if you say things like this. Jeesh...

Awww jeeez! I feel so warmly loved!

If only I wasn't so blatantly misquoted. You running 180 doesn't prove you can hit 250 or even 211. Some times a barier is just that - a barier. Unless you go to some unreal sub-zero solution, that no one can practically use, there is a speed limit to QPI. If that speed limit is below 200, then that's that.

The question (actually 2 questions) was how did they push it to or above 4Ghz, and has anyone been able to do that on air with the real deal (will 200x20 work with a really good cooler(CM's V8) and some more voltage)? and will it be stable enough to trust it to run some 24x7 visualization tasks?

Second question was whether or not a notable speed gain is there at for example 184x18 vs. 166x20. Meaning, does higher QPI produce a faster machine at same final CPU frequency, like the previous generation did with higher FSB/lower multi.

EDIT: Best OC tests I've seen, are still 3.8 stable, and that's it. So has anyone figured out how to push this thing to and beyond 4 and make it work well?