• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

FRONTPAGE AMD Bulldozer Breaks CPU Frequency World Record

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
That quote from Chew warrants more explanation - it is confusing and I was there. The 8.4Ghz chip was probably mostly untested or completely untested if thats what they are saying (not sure which they are saying) - I would believe what they say though. They had to run those chips to get the VID off them obviously though.

So while I was standing next to the table, I'm going by memory here, but they were talking about this after they switched out the first two setups they had on the bench. So I believe this was just before the 3rd setup, and it was the 4th CPU they used for the press demo to try to get the record. I thought it was weird as I was standing there Sami and Brian were discussing what to do next and Sami said something about taking a shot in the dark, just gambling on an untested CPU next. I have no idea if they then grabbed a new CPU from the tray, or picked from the trays and trays of CPUs that had already been tested in some way at least... But I thought that was sort of weird since there were so many chips that had already been tested to some degree. There were a lot with VID marked as well as LN2 pretesting to 8GHz or so labeled and sitting in the trays next to them.

So seeing all the labeled CPUs, clearly there was a lot of testing that went into determining which chips behaved which way. Checking VID and recording clocks and voltage to determine trends in frequency results across chips. AMD had sent out CPUs a week before also, where 8 BD chips hit 8GHz - they didn't say 8 out of how many. That all occurred the week before the event reportedly, simply on LN2. An 8GHz chip on LN2 is going to be a contender with 8.3GHz on LHe.

So the obvious question... If they had been through that many chips in pretesting, they had spent so much time learning about the architecture, how it behaves under cold, and recording results for each chip - Why throw a hail mary at crunch time in front of 20 hand picked journalists, plus me?

Maybe the first best CPU's/boards they just tried failed, had condensation problems, died in pretesting... Dunno, a lot of things can go wrong doing subzero. It doesn't totally make sense to put all that time into pretesting, then when it comes to do or die time take a hail mary pass... So what was the reason? Maybe the VIDs they found to hold the most promise in pretesting was enough to give them an idea of what the chips would do.

Ok here is what i did.

I ran tests previously on various vid chips. I found the lowest vid which was rare really sucked, second lowest not so bad and 3rd from lowest scaled decent.

Taking that info i just VID binned in front of media if you recall and labeled all chips. Post to bios with a profile with all power saving off log voltage remove chip and label simple as that......

After i got tired of testing chips we had 24 total with my personal choice of VID which is the third from lowest so not really what you would call leaky.

We tested all chips at 1.8v ( so as not to damage them ) on ln2, 7.7, second sort of a dud 7.2 third 7.7, 4th hit just under 7.9 and we decided to go under LHe at that time.

As far as personal testing prior to event I think i tested 15 chips including my 2 personal B0 samples which I might add were nothing spectacular. So 13 B2 chips in total.

4 did 8 gig on ln2.......

Now onto why we used an untested chip versus chips we had already tested prior. Quite simple beating on chips at home at 2.0 v on 32nm is bad for a cpu.....at that point it is just a fuse waiting to pop.

I actually pushed one at home purposely that was a 7875 chip to find the breaking limits.....and i found it.

Now i'm not sure about you but LHe tanks are $1000 a wack, you burn quite a bit just getting the pot down to optimal temps. My personal motto for live events is this "Failure is not an option"

Knowing this i'd hate to waste half a tank on a previously beaten on chip and since we knew what we needed clocks and volts wise we knew it would not take much to find more 8 gig chips, 4 out of 15 is preety good odds.....
 
We tested all chips at 1.8v ( so as not to damage them ) on ln2, 7.7, second sort of a dud 7.2 third 7.7, 4th hit just under 7.9 and we decided to go under LHe at that time.

My favorite part right there... 7.2Ghz is now considered a dud... what is this world coming to? :D

And on the subject of binning vs random pick... Me no cares one bit... You guys hit 8.4GHz and that's what matters. I can't wait for these things to be available on retail so I can play some (unless you have one of them 24 chips laying around doing nothing!) :sn:
 
Aye, let me just add that I don't give a rip how many chips you binned; I honestly expected the number to be much, much higher. Thanks for sharing!
 
Thanks Chew, awesome explanation of how it went - appreciate you putting that down for us. Not every day that people can get a play by play on how a world record is made for a live performance. :) The approach makes a lot of sense now that you explained it, especially given the odds you found out of the chips you tested, not to mention the fact its live and failure isn't an option. Doing subzero live for an audience like that is something very different than beating on hardware in the comfort of home.

Thanks again. Hope AMD got you guys good and drunk to celebrate after we flew out. :)
 
I was so hoping my loyalty would be rewarded and I could finally give some **** back to the intel crowd and their superior attitudes (in my circles anyway :p).

I'm jumpin around my lounge room right now like 10yr old who's just been told he's going to Disneyland, and I'm way to old for that I'm gonna do myself an injury.
 
Only 1 cpu/2 cores(modules) were running, thats how CPU-Z benchmarking is done for max clocks.

There are other tweaks, but one of the basics if you are benchmarking CPU-Z on subzero cooling is to disable extra cores, and you get extra stability to hit absolute max clocks.

With 2 cores, you boot at a speed like 4.5-5GHz. Then you use software to increase the multi or HT-Ref to increase one of the cores to max clocks, and pray that you can save a validation with CPU-Z successfully before it crashes. Some people also do things like set explorer to run on the core at the lower frequency also, and other tweaks.
 
Three of the four modules were disabled for the record attempt. It's common for trying for high frequencies under extreme cooling.

Edit - Whoops...beaten to the punch.
 
Only 1 cpu/2 cores(modules) were running, thats how CPU-Z benchmarking is done for max clocks.

There are other tweaks, but one of the basics if you are benchmarking CPU-Z on subzero cooling is to disable extra cores, and you get extra stability to hit absolute max clocks.

With 2 cores, you boot at a speed like 4.5-5GHz. Then you use software to increase the multi or HT-Ref to increase one of the cores to max clocks, and pray that you can save a validation with CPU-Z successfully before it crashes. Some people also do things like set explorer to run on the core at the lower frequency also, and other tweaks.

Yeah I realize that now.

Good thing I'm not a bencher :D
 
This bodes well for Bulldozer, though it all comes down to IPC.... AMD have stated an increase in IPC over the last gen Stars architecture, but the question is how much?

I guess I'll wait for the benchies....
 
Almost forgot... Skip to 1:45 in the embedded AMD video for a little "easter egg"!

The feed line for the Liquid Helium was ejected out of the container, scaring the crap out of onlookers before Simon Solotko wrangled the line and stuck it back in the dewar where it belonged! ...It only is shown for a split second but you can see the initial response where people jump back. It was really loud and startling, and in person a couple people nearly dove away from the demo. Everyone nearby at the time got a good laugh once we realized no one got hurt when the tube shot out. :)
 
Benchmarkers and extreme overclockers, enjoy the excitement (I am too). But what I am very eager to see is 24/7 overclocked, multi-threaded performance. To me, that's what will really count.

Please don't misunderstand, I have a large measure of respect for you gentlemen overclocker magicians who can find the ultimate speed in hardware. You are high performance people.
 
Last edited:
Almost forgot... Skip to 1:45 in the embedded AMD video for a little "easter egg"!

The feed line for the Liquid Helium was ejected out of the container, scaring the crap out of onlookers before Simon Solotko wrangled the line and stuck it back in the dewar where it belonged! ...It only is shown for a split second but you can see the initial response where people jump back. It was really loud and startling, and in person a couple people nearly dove away from the demo. Everyone nearby at the time got a good laugh once we realized no one got hurt when the tube shot out. :)

Yah that was because I OCed the Lhe dewar, if you hold off all the safeties and increase the pressure you can gain a few degrees.
 
I suppose a dewar is one thing that you can't apply the MOAR VOLTS theory and get gains.
 
Back