Phenom II X4 @ +6 GHz by Team Finland

[massman]

Cold bug information:

I have now confirmed from AMD that when running Phenom II X4 with LN2 HyperTransport link has to be in HT 1.0 mode (running 1 GHz) to prevent CPU cold bugging. AMD suggested us using this setting in Chicago event and we had no problems hitting +6 GHz.

Could this affect the coldbug issues with the 65nm Phenom as well?

Can you provide performance scaling information of the HT link? How much decrease in performance does this cause? Moreover, how does this translate in coldbug terms: If HT link is set at, for instance, 1.6G, what will the coldbug be?

As far as I know, the low HT link does not make the performance decrease dramatically.

[Halk]

Tom's Hardware posted pi 1M result for 4ghz

Would you mind linking that, I wasn't able to find it?

I'd like to know when the NDA ends, but I suspect that's under the NDA?

And of course when these go retail, but I'm pretty sure that's NDA too.

[massman]

20/12, if I'm not mistaken

[SF3D]

Could this affect the coldbug issues with the 65nm Phenom as well?

Can you provide performance scaling information of the HT link? How much decrease in performance does this cause? Moreover, how does this translate in coldbug terms: If HT link is set at, for instance, 1.6G, what will the coldbug be?

As far as I know, the low HT link does not make the performance decrease dramatically.

This should not affect 65nm parts. Only these new ones.

We didn't have time to test those things in Chicago and we have to test them in our next bench sessions. HT link doesn't do huge difference, but NB speed does. So if HT is at 1GHz it shouldn't do any harm.

I can test those settings on air today and I'll let you know.

[Sampsa]

20/12, if I'm not mistaken

Yes you are :P

[knightwolf654]

soo, so far if you dont bring down the htlink to ht1.0 settings. the cold bug is at -70C?

[Oliver]

This should not affect 65nm parts. Only these new ones.

We didn't have time to test those things in Chicago and we have to test them in our next bench sessions. HT link doesn't do huge difference, but NB speed does. So if HT is at 1GHz it shouldn't do any harm.

I can test those settings on air today and I'll let you know.

thank you very much

keep pushing it

[massman]

This should not affect 65nm parts. Only these new ones.

@ Macci: Why?

We didn't have time to test those things in Chicago and we have to test them in our next bench sessions. HT link doesn't do huge difference, but NB speed does. So if HT is at 1GHz it shouldn't do any harm.

I can test those settings on air today and I'll let you know.

Sound great.

I have my 9950 BE setup assembled as well, might be cool to compare performance scaling of both platforms .

[SF3D]

@ Macci: Why?

I'm not Macci, but there is a easy answer for this.

In pervious processors there was some particular parts inside the die, which were holding these CPU's back. Now in this latest 45nm revision, they finally got that thing fixed. Now when CPU get colder it doesn't have any parts which would fail under subzero temperatures.

I know the exact answer for this, but can't tell it. Sorry

[Sampsa]

Video coming soon, uploading it to YouTube!

[G-UNIT91]

I'm not Macci, but there is a easy answer for this.

In pervious processors there was some particular parts inside the die, which were holding these CPU's back. Now in this latest 45nm revision, they finally got that thing fixed. Now when CPU get colder it doesn't have any parts which would fail under subzero temperatures.

I know the exact answer for this, but can't tell it. Sorry

Can you post anything else?

also in pvt!

[massman]

I understand that taking away the parts that fail under subzero conditions resolves the coldbug issue in this way that there's a big difference between 65nm (with the failing parts) and 45nm (without ...), hence moving coldbug from around -10°C to -196°C and maybe even lower, but I fail to see what this has to do with the 'lower HT link on 65nm' suggestion.

As told, the HT link helps in preventing the CPU from coldbugging as follows(*): the lower the HT link, the lower the coldbug. If this could be translated in a 'coldbug' function, by which I mean that the coldbug is higher if the HT link is higher, we could conclude that the HT link is one of the factors which play a role in the coldbug prevention scheme. At the moment I have no view on how different the two cores are in terms of design, so I might be missing essential information here, but if the HT link is indeed one of the factors that helps decreasing the coldbug in the 45nm design, why wouldn't it help in the 65nm design? It might not help much, but as we all know -50°C is much better than -10°C. The coldbug won't be removed by using a low HT link frequency, but it might help decreasing it just a tad.

(*): This is a rather delicate assumption I make here. I see the HT link 'trick' as scalable, in other words 1GHz gives a lower coldbug than 1.2GHz, which gives a lower coldbug than 1.4GHz and so on. If that is not the case, and thus the HT link 'trick' only works when it's clocked at 1GHz, the following line of thoughts is not applicable anymore.

[Kabauterman]

on the weekend I have seen also a Phenom II on LN2
and It haves it CB and CBB on -70° :-(

Results are not so fine I think, but it is still under NDA, so I can not say more.
AMD must go other ways than price out there processors with "high" reached LN2 freqs!

[informal]

on the weekend I have seen also a Phenom II on LN2
and It haves it CB and CBB on -70° :-(

Results are not so fine I think, but it is still under NDA, so I can not say more.
AMD must go other ways than price out there processors with "high" reached LN2 freqs!

Read a few posts above yours,coldbug is there if you set higher than 1Ghz value for the HyperTransport links.If you set it to 1Ghz it should be fine(ie. no coldgub).

[dread77]

Just a photo from last tests....



Here you can find a summary of our small session.

[FatAlbert]

SF3D, can I ask about mem controller ? I mean, if you set div 1066 how low is timings, 4-5-5-12 was in phenom, how about now ?

It's against NDA ?



I hope this time I'll get answer.

[SF3D]

SF3D, can I ask about mem controller ? I mean, if you set div 1066 how low is timings, 4-5-5-12 was in phenom, how about now ?

It's against NDA ?



I hope this time I'll get answer.

It's not against NDA, but I have no idea. I'm not sure what you mean, but I used SPD timings, cause my memory is so weak. They can't do 1066, that's for sure.

Someone with better ram should answer

[donitsi]

Cool video guys

[FatAlbert]

It's not against NDA, but I have no idea. I'm not sure what you mean, but I used SPD timings, cause my memory is so weak. They can't do 1066, that's for sure.

Someone with better ram should answer

hmm

may Sampsa has better mems and can share ?

Sampsa ?

it's important 'coz if div 1066 work like in P1 it's suck 'coz again for high mhz on mems we have to oc it by htt ;/

and it suck as we know.

[Trouffman]

on the weekend I have seen also a Phenom II on LN2
and It haves it CB and CBB on -70° :-(

Results are not so fine I think, but it is still under NDA, so I can not say more.
AMD must go other ways than price out there processors with "high" reached LN2 freqs!

Ok here is a little transcript of an original post on Syndrome-OC :

Bench Bed
* Gigabyte 790GP-DS4
* DDR2 CORSAIR
* AMD Phenom II 940
* nVidia GTX 280


* no coldbug if we play only with the multi (already said here)
* Coldbug @ -68°c (under Windows) with HT Frequency above 220 (can't remenber the rest of the test (that's not me that had thest but i was looking at)
* Coldboot @ -80°c

max screen was 5ghz (from Jmax), no real scores because he only bench @ 4ghz or near to 4ghz (a little above)

enjoy

[massman]

I have my 9950 BE setup assembled as well, might be cool to compare performance scaling of both platforms .

Will be using this graph in my Dfi review, so I might as well post it here.

I've been checking the performance scaling of the HT Link, which is in case of the Phenom2 very interesting. As most of you probably already figured: very little influence on the performance, however, I was expecting a difference in the 3D results, which seems to be exactly what happens: nearly 100FPS gain by moving from 200MHz to 1800MHz (600 -> 700, using stock 4870X2).

[Caveman787]

so hypertransport only effects performance in gaming?

and 100 fps....your scaling graph looks to me liek you meant it as 1.xx performance scaling of the original.

a 13% increase is 100 fps?

or am I entirely off in my confusion.

[biohead]

I think past 1000 MHz it's hit or miss. There needs to be more tests done to have any meaningful data to discuss.

Besides, do we know if the NB was clocked together with the HT link? Because if so, there's your magic wand.

[AMDDeathstar]

I think past 1000 MHz it's hit or miss. There needs to be more tests done to have any meaningful data to discuss.

Besides, do we know if the NB was clocked together with the HT link? Because if so, there's your magic wand.

The NB is seperate from HT link, and should be equal or perferably higher then HT link.
I don't bother adjusting HT link anymore.
The NB also controls the L3 cache you want to get as high as you can
For 3D stuff I pefer upping PCI-E speed

[massman]

so hypertransport only effects performance in gaming?

and 100 fps....your scaling graph looks to me liek you meant it as 1.xx performance scaling of the original.

a 13% increase is 100 fps?

or am I entirely off in my confusion.

HT link frequency probably affects performance of PCI-e, sata, ide, usb and so on.

Graph is indeed 1.xx of 200MHz; where XX = performance gain in %

200MHz HT link ~ 600FPS, 1800MHz HT link ~ 700FPS.

I think past 1000 MHz it's hit or miss. There needs to be more tests done to have any meaningful data to discuss.

Besides, do we know if the NB was clocked together with the HT link? Because if so, there's your magic wand.

I used the nature benchmark because it's very little hit or miss, it's quite reproducable. More data is always better, however, in this case you should look past the actual numbers and see the performance scaling of the 3D benchmark. 15% increase is significant, the increase or decrease in other tests isn't.

NB was clocked at 2GHz in all tests. I know how to make comparisons