Skip to main content

The Stilt Improves SuperPI Performance on AMD Systems

It is widely known that AMD-based systems aren't particularly performing well with SuperPI performance. This was previously blamed on the architecture, though it now appears that this was simply a BIOS development limitation on AMD platforms.

Finnish overclocker the Stilt has managed to solve these optimization issues by using BIOS developers' guides, and increasing SuperPI performance. The result is a simple patch that can be run on both 32-bit and 64-bit installations of Windows.

The previous time record for a 32M SuperPI run on an AMD Richland chip was 18:15 -- the AMD A10-6800K managed to pull it off within 17:34 using the patch. While not a big difference in time, do consider that the A10-6800K APU was running at just 4.1 GHz as opposed to 5.0 GHz, and they have the same architecture.

Niels Broekhuijsen

Niels Broekhuijsen is a Contributing Writer for Tom's Hardware US. He reviews cases, water cooling and pc builds.

  • KelvinTy
    IMO, 41s is quite a lot... even without the frequency difference...
    By the way, do you really need liquid nitrogen for the A10-6800K to be running at 4.1GHz?
    Reply
  • cscott_it
    Well - that has to do with the particular instruction set. However, didn't enabling that instruction set cause other benchmarks to also drop?
    Reply
  • sarinaide
    The Stilt is becoming a legend, on a 1mb run I get around 3-4s improvement, big deal for any overclocker.
    Reply
  • JPForums
    Assuming linear scaling (which is probably a bad assumption) a 5.0GHz A10-6800K should clock in around 14:38. Of course that linear scaling also suggests the original Richland chip (without the patch) would clock in at 22:15, if set to 4.1GHz. What are the chances someone here will post those numbers so we can see if the scaling is at all close to linear?
    Reply
  • deksman
    Hm...
    4% gain on a BIOS patch.
    The differential is not large, but its certainly a good thing (especially when we are talking about a simple APU refresh.
    It makes one wonder, how many other possible optimizations were missed by manufacturers (who seem to have been giving AMD solutions a hard time left and right in the mobile sector).
    Reply
  • ojas
    LOL i get 17m 23.117s on a 2.77 GHz Core 2 Quad (Q8400).

    Used to get 16m 57s with version 1.5.
    Reply
  • Soda-88
    11056168 said:
    IMO, 41s is quite a lot... even without the frequency difference...
    By the way, do you really need liquid nitrogen for the A10-6800K to be running at 4.1GHz?

    4,1GHz is stock frequency for 6800K
    Reply
  • vmem
    11056382 said:
    The Stilt is becoming a legend, on a 1mb run I get around 3-4s improvement, big deal for any overclocker.

    I think he already is a legend lol... this only makes him more... well, legendary.

    AMD should hire him, great addition to their drivers team. considering SuperPI is a part of the benchmark suite on almost every review site, it's amazing that AMD's drivers team didn't figure this out earlier
    Reply
  • esrever
    11057436 said:
    11056382 said:
    The Stilt is becoming a legend, on a 1mb run I get around 3-4s improvement, big deal for any overclocker.

    I think he already is a legend lol... this only makes him more... well, legendary.

    AMD should hire him, great addition to their drivers team. considering SuperPI is a part of the benchmark suite on almost every review site, it's amazing that AMD's drivers team didn't figure this out earlier

    AMD doesn't really care about 20 year old software that nobody has to use any more. This still doesn't make up for the fact that x87 is pointless towards the future or really even the present. AMD has better things to do than focus on improving performance in a useless benchmark. If they cared at all, they could have put a real x87 instruction decoder on their CPUs.
    Reply
  • vmem
    11057539 said:
    AMD doesn't really care about 20 year old software that nobody has to use any more. This still doesn't make up for the fact that x87 is pointless towards the future or really even the present. AMD has better things to do than focus on improving performance in a useless benchmark. If they cared at all, they could have put a real x87 instruction decoder on their CPUs.

    I agree that SuperPI no longer has any real world applications. however, I must stress again that it is used in almost every benchmark suite on tech review sites. AMD should care about SuperPI and such "useless benchmark" performance because it pays off in marketing. it's just like how Intel's "Intel Inside" stickers have zero relevance, or the boat loads of money they spent buying off OEMs etc. in the end it's all about perception, and even if real-world performance is indifferentiable, the synthetic benchmarks will give people the impression that Richland chips are "worse" than they actually are.

    The say $100-500k a year they can spend hiring the Stilt to do some work for them will be way cheaper than the marketing budget they'd need to make up for slower performance in benchmarks etc. besides, that guy knows what he's doing, it's not like he's just some idiot who stumbled upon this.
    Reply