AMD spills secret to World Record clock speed
A little liquid helium goes a long way
On August 31, a team of AMD-sponsored overclockers cranked a Bulldozer-based AMD FX processor up to an unearthly 8.429GHz, setting a new world record. This week, The Reg sat down with the leader of that processor-torturing team to find out exactly how they did it.
"This is an extreme technology sport," Simon Solotko, AMD's senior manager of social media, told us. When we asked if social media wasn't a bit of a stretch from overclocking, he said he first got into social media through overclocking connections. Then he grinned. "I also do a number of our more extreme activities."
Any extreme sport requires its own specialized equipment, Solotko said. In the case of extreme overclocking, two of the most important tools are liquid nitrogen and liquid helium.
"It's really cold stuff," Solotko said of liquid helium. How cold is "really cold"? Well, liquid helium boils at -268.93°C (4.22 K). Liquid nitrogen, on the other hand, boils at a comparatively toasty -195.79°C (77.36 K). Solotko referred to liquid helium as "the coldest readily available stuff you can get for cooling."
Liquid nitrogen is all well and good, he told us, but liquid helium can get you close to absolute zero.
When overclocking AMD's Phenom II – an earlier processor – Soltko discovered that it could be operated at extremely cold temperatures. That isn't true of many if not most electronic circuits, which often fail under extreme cold.
Cold can affect timing, Solotko says, and it can also affect power circuits. If you get a voltage regulator too cold, for example, it will stop switching and it will "fry everything." So one of the precautions that Solotko and his team took was to ensure that their AMD FX was well-isolated from other motherboard components.
Extreme cold doesn't merely prevent a fast-cranking processor from overheating, he reminded us. Extreme cold also helps the conductivity of silicon.
To cool the test processor down to as close to absolute zero as possible, Solotko told us, his team first cools the processor with liquid nitrogen. Then they let the liquid nitrogen evaporate so it doesn't turn to ice. And then they use the liquid helium.
Liquid nitrogen turns to ice at -209.86°C (63.17 K), by the way, and liquid helium is about 60°C colder than that.
Once the liquid nitrogen has safely evaporated away from the cooling assembly on top of the processor, "Then we hit it with liquid helium, and we take the temperature further down still."
When we asked him exactly how cold he and his team were able to get their test chip, Solotko said he didn't know. The problem, he told us, is there isn't a thermometer that's accurate to those temperatures. The thermal probes that the team used read around -230°C, Solotko said, but this can't be trusted. "We don't actually know what that means. Our guess is that we're somewhere in the vicinity of 10 degrees north of 0 K."
Another reason that the true temperature of the silicon remains a mystery is that there are temperature gradients between those ludicrously fast-switching transistors and the temperature probes. Solotko cited as examples the first copper layer over the silicon, which he called the "metal lid", and the bottom of the liquid nitrogen- and helium- containing copper "pot".
Solotko described that pot as having a "miraculous" design. "Aaron Schradin, who's one of the team members, designed a cooler specifically for liquid helium. He invented something called the 'Schradin router', and what it does is it allows us to inject the helium straight in, and then it maintains a circular flow of the helium within the unit, also under some pressure," he said.
"And by maining some pressure, it has two advantages: first, the dissipation rate is lower; and second, the theoretical temperature we able to achieve is also a little lower." Win-win.
Although the liquid helium–cooled AMD FX hit 8.429GHz, it didn't stay there long. "Seconds," Solotko said. But at that clock rate, the chip churned through tens of billions of cycles – long enough to take a screenshot of the system running CPU-Z, which Solotko calls "essentially the de facto tool for these incredible runs," and which saves a validation file to disk after runnning an algorithm that determines whether the processor is stable, however briefly.
He did note, however, that he can keep the AMD FX running long enough to run benchmarks at less-extreme but still plenty snappy clock rates. "Oh, absolutely," he said. "With AMD FX we can get benches well into the sevens. Under liquid helium – we didn't do it – but I'd dare say we could run some benches in the eights."
Alhough Solotko is an AMD staffer and thus might be expected to honor his company's products, he seemed truly impressed by the AMD FX. "It was able to take a lot of voltage, extremely low temperatures, extremely high frequencies," he said. "It was very durable under extreme overclocking. So that was awesome. So it worked well, it scaled well, it responded to cold well – all the right variables."
Not that everything ran completely smoothly the first time the team took the AMD FX down near absolute zero. "The day before the 31st," he said, "we went under liquid helium for the first time and it was a challenge, and we weren't seeing scaling.
"And at some point, you have these moments of doubt. I mean, we were hitting 8.15GHz, and we were wondering 'Has this chip really peaked? Are we at the point of diminishing returns with cold?' Which would be a perfectly reasonable hypothesis."
Then Solotok laughed: "Wasn't true. Wasn't true."
He recounted that the record-breaking event happened not a moment too soon. "The press and the cameras were all there, and we had this spectacular run. We had not seen anything close to that," he said. "We had been maxed out at 8.15GHz right up to that moment. So it was awesome."
Solotko peppered his discussion of overclocking with words like "awesome", "cool", "neat", and "miraculous". It seems abundantly clear that he enjoys this "extreme technology sport". "There aren't a lot of consumer devices that you can rip apart, build on your own, build something that's a lot better than you can buy in the store, and experiment with like a PC," he said.
"What's neat about it is there's a Popular Science element," he enthused. "That's why high-school students watch these videos – because they're learning about conductivity, and cold, and why does it help to cool a processor. There's actually a lot of reasons why it helps: heat, conductivity. It's a neat science problem that addresses a number of high-level concepts."
From what he told us, he wouldn't be as happy working at Intel. In addition to the fact that he's encouraged to do "neat, risky things" at AMD, he told us that "It's fairly common knowledge that when we did this beyond–liquid nitrogen stuff with Phenom II, that other processors would cold-bug out. So that's one of the things that's cool and distinguished: that we can actually do these experiments with these CPUs."
More-recent Intel chips were beginning to exhibit fewer cold bugs, he told us. But he says Sandy Bridge does not scale particularly well with cold. "You'll see the top frequencies in Sandy Bridge in the fives – it does not have this remarkable ability to scale with cold as an attribute of its design."
Another aspect of overclocking is the cameraderie – and from what Solotko told us, there needs to be a feeling of bonhomie, considering that there's little money to be made by becoming an expert overclocker.
"This is not a very profitable endeavor," he said. "It's funny, because you'll see in the forums that people fantasize that someone is making a lot of money or getting a lot of gear overclocking. You talk to the guys who are really doing this? They suffer, they suffer."
Solotko, however, has a day job. "I feel like we're best in class," he said, "but it's still not enough to live on in terms of helping these guys out. I've got a job. I'm fine. So it's just a question of keeping them fed."
A band of (overclocking) brothers
Solotko described his five-man team in enthusiastic terms. "We work with some really good guys," he said. "This time we had Sami Mäkinen, who is a long-time AMD overclocker – he's very smart and precise. And Brian McLachlan, who's like the mirror image of him: extremely pragmatic and gutty, and diligent, and relentless, and he will make sure, and he will make sure, and he will make sure.
"And we had a guy named Pete Hardman who really knows a lot about the silicon process and the platform-engineering process – a very smart guy. And Aaron Schradin, who designs our cooler.
"And then there's me, who's kind of like the coach. I understand all the things that have to go together in a short period of time – how do we get it done, which can be a real challenge."
When we pointed out to Solotko that he also was the one who provided the liquid helium, he laughed. "That is among my skills: how to acquire, and pay for, and get funded for liquid helium."
Solotko and his team may be inspired by the awesome, cool, neat, and miraculous adventures of overclocking, but he's a practical man. He described the AMD-funded overclocking effort as being "at the interface between the extremely practical and pragmatic – which is creating interesting news and marketing processors – and really looking ahead. It's at that interface, and it's a neat compromise."
Awesomely, miraculously cool – and neat. ®