Re: AMD patent about inverse hyperthreading

On 28 Jun 2006 23:57:05 -0700, "David Kanter" <dkanter@xxxxxxxxx>
wrote:

Nobody's explained to me how it would work, and made a good argument
how it would increase performance.

That wasn't your original question. You asked
Do you really think that is feasible? When was the last time someone
installed a 'CPU driver'?

Both George and Tony have pointed out that CPU driver exist and quite
commonly "installed", therefore the idea is feasible.

No, they pointed out they exist. They did not mention anything about
common installation, nor feasibility.

So let me ask you, when was the last time someone installed a CPU
driver? Typically it comes with the OS, right?

Last time I installed one was when I wanted to get Cool 'n Quiet to
work on an Athlon64 system. The driver that came bundled with WinXP
did not support the feature, I needed to download one from AMD's
website and install it.

Will microsoft modify
WindowsXP just to add this feature for AMD? No they won't.

Why not? It's not like they haven't added features to their OS to
support CPU features for both AMD and Intel in the past. WinXP didn't
come with Execute Disable Bit support, but when AMD added it to their
processors MS went along and added it to the OS. Same goes with
adding SSE support for Intel back in the day. Or hell, how about
x86-64? That required a whole new rebuild of an OS and it made it!

It needs to be something relatively seamless to the OS. SMT can be
treated as seamless (just pretend it's another CPU) but your
performance gains will be modest. Ditto for NUMA (if you don't
believe, try benchmarking 4S NUMA systems using Windows).


Hmm...

Linux
http://www.spec.org/cpu2000/results/res2005q4/cpu2000-20050919-04714.html

vs. Windows
http://www.spec.org/cpu2000/results/res2005q4/cpu2000-20050919-04716.html

115/128 (base/peak) for Linux vs. 134/144 for Windows in SPEC
CINT_2000. Windows seems to be doing just fine here on a
4-socket/8-core Opteron system. Compared to these two results:

http://www.spec.org/cpu2000/results/res2005q3/cpu2000-20050819-04509.html

http://www.spec.org/cpu2000/results/res2005q3/cpu2000-20050819-04513.html

We get that Windows is scaling from 42.0 -> 134 when going from
2-socket/2-core up to 4-socket/8-core, or a 319% improvement. Linux
meanwhile goes from 39.2 to 115 on the same two systems, or only a
293% improvement. For this benchmark at least, Windows scales better
in a NUMA environment.

Similarly numbers for SPEC CFP2000_rate can be found here:

http://www.spec.org/cpu2000/results/rfp2000.html

As above we get Windows going from 41.6 -> 119, or a 286% improvement.
Linux meanwhile manages to go from 47.0 -> 129, or a 274% improvement.
Again, Windows is scaling better.

To get the
full effects you really need to have the OS be aware of such things (or
perhaps applications instead of the OS might work, but ideally both
would be aware). Windows XP, the predominant gaming platform is
certainly not aware of any 'reverse hyperthreading'.

Fortunately WinXP is software and can be patched. In fact, it gets
patched quite often! And let's not forget that Windows Vista is set
to become the new standard home-user OS by the end of this year
(maybe, assuming it doesn't get delayed yet again). It is certainly
possible that this is a feature that will only work under Vista.

It has nothing
to do with whether it will have significant benefits or not, which
wasn't what you were asking as well.

No, but it's a rather pertinent issue. Supposedly this reverse
hyperthreading will restore AMD's primacy in the gaming market. How
will it do this if the benefits are minimal?

Well now, that's the real question isn't it. Not whether this is
possible, but whether it's worthwhile, and I don't think any of us are
in a position to judge that at this time.

I still have yet to hear a good theory for exactly how this works in
the context of a system. I have read some of the research in the area
and it certainly doesn't sound terribly promising. Those two factors
combined make me skeptical.

I'll agree with you on that point, I'm rather skeptical about this
providing much of a performance boost in anything other than extreme
situations. However that's not to say that it isn't possible to
implement.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca
.

Relevant Pages

  • Re: Starting Windows 2000 hangs on new install.
    ... | AMD cpu, never cured it and I did swap it for another identical cpu, ... and a fresh shrink-wrapped copy of Windows 2000 would it ... If OK reboot from win2k installation disk (you may have to ...
    (microsoft.public.win2000.setup)
  • Re: CHKDSK NTFS file system corruption boot screen message flooding (SOLVED)
    ... Your Installation Specialist ... Windows XP SP3 detects a disk problem at boot time and runs chkdsk. ... or registry fix for large LBA drives, ... older Microsoft-certified ultra.sys driver. ...
    (microsoft.public.windowsxp.general)
  • Re: 0x000000D1 Message
    ... Uninstall it and let Windows re-install it to see what happens. ... Update then check the hardware section to see if an updated driver ... Your Installation Specialist ...
    (microsoft.public.windowsxp.general)
  • Re: Build Failed Message
    ... I can't build any driver or subset of the OS by ... because there is still something wrong with your installation. ... I received no build.log file in the C:\WINCE600 folder. ... Windows CE 6.0 Platform Builder ...
    (microsoft.public.windowsce.platbuilder)
  • Re: Starting Windows 2000 hangs on new install.
    ... Windows Me is happy with either. ... CPU temperature ... If OK reboot from win2k installation disk (you may have to ...
    (microsoft.public.win2000.setup)