BMOS speed vs Linux

379 views
Skip to first unread message

aleroise

unread,
May 17, 2015, 9:14:48 AM5/17/15
to bareme...@googlegroups.com
Hi,

I was wondering how fast is BMOS in comparison to Linux for a particular application : do you have kind of benchmark ? Or just a rough idea of the speed ratio ?

Thanks for your reply.

Rémy

Ian Seyler

unread,
May 21, 2015, 9:33:37 AM5/21/15
to bareme...@googlegroups.com, remyl...@gmail.com
Hello Rémy,

BMOS is only a little faster when executing on 1 CPU core. The real advantages show up when using SMP.


-Ian

Benjamin Scherrey

unread,
Jun 26, 2015, 1:02:47 PM6/26/15
to bareme...@googlegroups.com, remyl...@gmail.com
I notice that sysinfo.app reports I have 8 cores when in fact I have 4 cores with 2 threads each. Is there really 8 independent execution units or is it really only 4 + hyperthreading? 

  -- Ben Scherrey

PS: BareMetal is quite a well done a remarkable project. I'm considering implementing a forth-like language/environment for it now.

Ian Seyler

unread,
Jun 28, 2015, 9:51:03 PM6/28/15
to bareme...@googlegroups.com, sche...@proteus-tech.com, remyl...@gmail.com
BareMetal treats hyperthreading like regular cores. With a multicore app I have seen a ~20% performance improvement with hyperthreading enabled.

-Ian

42Bastian

unread,
Jun 29, 2015, 2:01:16 AM6/29/15
to bareme...@googlegroups.com
Ian

> BareMetal treats hyperthreading like regular cores. With a multicore
> app I have seen a ~20% performance improvement with hyperthreading
> enabled.

I wonder, if software can detect if the CPU has 8 real or 4 real plus 4
virtual cores?

--
42Bastian

Ian Seyler

unread,
Jun 30, 2015, 9:48:45 AM6/30/15
to bareme...@googlegroups.com, eb...@monlynx.de
Yes, software would be able to detect this but the method has changed a few times.


Ideally I should have Pure64 gather those details and store it for the OS to use.

-Ian

Benjamin Scherrey

unread,
Jul 1, 2015, 2:02:26 AM7/1/15
to bareme...@googlegroups.com
From my recollection of the Intel documentation I think it's pretty explicit so knowing if you're dealing with a "real" core or a hyperthreading instance shouldn't be difficult. This is actually fairly critical to the work I'm doing with BareMetal as I'm using the cycle clock and cpu performance monitors to try to quantify the impact of non-aligned memory, cache misses, etc. along with general performance and improvements of 64-bit algorithms and overhead of vmx which BareMetal is clearly ideally suited for. Once I get into multiprocessing efforts this distinction will be critical. I'd rather BareMetal make this distinction explicit than implicit but I can't yet say what the impact will be on me until I try it.

Thus far, otherwise, I'm very pleased with BareMetal OS and hope to have some useful or at least interesting contributions to it. I'm curious - is uefi part of the road map for BareMetal OS? 

  -- Ben

Ian Seyler

unread,
Jul 1, 2015, 3:37:12 PM7/1/15
to bareme...@googlegroups.com, sche...@proteus-tech.com
Hi Benjamin,

I'll take a look at the updated Intel/AMD docs to see what has changed. Ideally there will be a system flag in the OS to use HyperThreading or not.

UEFI won't be part of BareMetal but I do plan on adding support for it to Pure64 (the BareMetal boot loader).

-Ian

Benjamin Scherrey

unread,
Jul 7, 2015, 7:11:44 AM7/7/15
to bareme...@googlegroups.com, sche...@proteus-tech.com
Putting UEFI into Pure64 would grant these capabilities to BareMetal or would be restricted to Pure64 only. I presume the latter. What's your thinking regarding utilizing UEFI in BareMetal? I would think it would make things easier regarding hardware support and save effort on device drivers if one wants to work outside a VM, no?

thanx,

  -- Ben
Reply all
Reply to author
Forward
0 new messages