Before every major architecture launch from AMD or NVIDIA, the companies typically hold an editor's or tech day. These events usually last about a day (sometimes two) and involve a bunch of press sitting in a conference room while they get peppered with presentations and pepper back with questions. Engineers and key architects are usually present. The goal behind these tech days is to help the press understand, at least from the manufacturer's perspective, what makes their new architecture tick. Benchmarking usually doesn't happen at these events, but typically we get hardware at the event or shortly thereafter. If you're curious, Intel doesn't typically do a tech day - that's what IDF is for.

Two weeks ago AMD held one of these tech days for its upcoming Bulldozer architecture, which will be sold under the FX brand (e.g. AMD FX-xxxx CPU). Although it'll still be a little while before I can talk about most of what transpired at the Bulldozer tech day, there is one thing I'm allowed to share today: overclocking potential.

AMD was conducting overclocking experiments at the tech day and had three different stations setup for us to look at. The first used a sub-$100 closed-loop waster cooling solution from Antec (Kühler series). I can't tell you much about the chip itself other than it is an 8-core FX processor that AMD was able to overclock to 4.8GHz using the Antec Kühler.

Next up was phase change cooling. Armed with a phase change cooler AMD pushed another 8-core FX CPU up to 5.894GHz at 1.632V.

AMD ended on its most aggressive cooling solution: liquid helium. Using liquid helium AMD was able to take Bulldozer to a new world record of 8.429GHz. The resulting overclocked frequency was high enough to get AMD's FX processor inducted into the Guiness Book of World Records for the highest frequency for a computer processor.

None of this tells us much about how Bulldozer will perform unfortunately. The most interesting number is likely the first number (4.8GHz) which gives you the upper bound of what to expect from an overclocked Bulldozer at home without any exotic cooling.

AMD recently announced it started shipping server versions of its Bulldozer CPU and that desktop parts will be available in Q4.

Comments Locked


View All Comments

  • erple2 - Tuesday, September 13, 2011 - link

    That's almost true. Helium's melting point (going from solid to liquid) is actually 1.15K, but it's boiling point (going from liquid to gas) is 4.55K (technically 4.549994K).

    BTW, 0K is defined as "absolute zero", and is part of the SI units. The relationship between Celsius and Kelvin is strictly a change in the zero point: 0°C = 273.15K, 0K = -273.15°C. So 4K is really 4 degrees Celsius away from absolute zero. It's also 4 Kelvin away from absolute zero so ...

    Either way, that's a crazy high multiplier...

    One thing that I've been trying to find out is what the previous record was.
  • omg123 - Tuesday, September 13, 2011 - link

    1. Helium does not melt at 4K, it boils at 4K. At atmospheric pressure helium can only exist as liquid or gas. At absolute zero it is still liquid.
    2. 4K means 4 degrees Kelvin above absolute zero (0K), which is exactly 4 degrees Celsius above absolute zero, not 6.
  • omg123 - Tuesday, September 13, 2011 - link

    Here is the phase diagram of Helium:

    It can be solid only at pressures of at least 2.5 MPa, which is ~25 atm (1 atm is around 10 kPa).

    Just saying :)
  • UrQuan3 - Tuesday, September 13, 2011 - link

    Remember that there is always a delta between the temperature of the cooler/heatsink and the transistor junction temperature. To understand, the spec temperature quoted is normally in terms of the chip's exterior since it can be measured.

    As a random (made up) example, let's say there's a 60 deg C delta. So if the heatsink is 70C, the junction temperature is130C. If the delta is constant, than at 4C, the junction tepmperature is 64C. This makes more since to me since at extreme temperatures, most materials simply stop conducting. While the delta may vary some with temperature, I'm just trying to give a simple example.
  • Stahn Aileron - Tuesday, September 13, 2011 - link

    The Celsius and Kelvin scales use the same interval (1°C = 1°K). So I don't see how 4°K is 6°C away from Absolute Zero... (AT most, it's a 4.15° difference.) I don't even know way you'd use mixed units anyway, now that I think about it...

    Also, helium doesn't melt @ ~4°C. It boils at that temp. If Wikipedia is accurate, helium melts at just under 1°C. (Has anyone really gotten any significant mass of helium frozen anyway...????)

    Anyway, the general rule is, the colder a processor gets, the further you can push it. If you want records set, you do what you must. In this case, AMD said screw liquid nitrogen (the more common cyrogenic coolant with a boiling point of ~77°C) and used helium instead.

    Now my question is, will anyone try something similar with an Intel CPU just to compare absolute overclock frequencies? (We'll ignore actual performance.)
  • izmanq - Tuesday, September 13, 2011 - link

    i thought 1K = 1 celcius, so it's should be 4 degrees away from absolute zero :D :D
  • alpha754293 - Tuesday, September 13, 2011 - link

    a) how can 4 K be 6 degrees Celsius above absolute zero?

    b) It's melting point is only valid at pressures > 2.5 MPa. At atmospheric pressure, it remains a liquid at absolute 0.

    c) The temperature you quoted is the boiling point at 1 atm. (You Celsius number is still wrong, but that's another point.) It also depends on whether you're talking about He-3 or He-4.

    d) They probably used He-4(l) because it's probably cheaper than N2(l).
  • sep332 - Tuesday, September 13, 2011 - link

    It's not about temperature, it's about heat. The chip itself never gets down to 4K. But the heat energy from the chip moves into the helium and boils it, which keeps the chip at a workable temperature (about +160C I think). So being that cold is really just so they can heat it up again with the CPU.
  • Z Throckmorton - Tuesday, September 13, 2011 - link

    Anand do you think it's reasonable to infer from these results (i.e. cherry-picked processors) that average consumers on mid-range air will likely be able to take an FX-8150 to ~4.4GHz (while preserving TC)? While not up to SB -K standards, that's still a respectable ~20% OC.
  • GullLars - Tuesday, September 13, 2011 - link

    My Phenom II x6 1090T is prime stable @ 4,0GHz on stock volt with NH-D14. If Bulldozer clocks better than Phenom II it sounds good to me :)
    What will really make a difference though is if it gets better single thread performance. It's already set to give great multithread performance pr $.

Log in

Don't have an account? Sign up now