Fastest Linux OGR Dnet client for Tbird

[cirrus1]

Plain and simple - which is the fastest client?

thanks

 

[RC]

version 469

 

[cirrus1]

thanks installing right away

 

[ottawanker]

463 is definately way faster than 469 on AMD athlons/thunderbirds/durons, and a little faster for k6-2s. 469 is way slower..

This is 463:

[Jun 29 01:11:53 UTC] Benchmark for OGR core #0 (GARSP 5.13)
0.00:00:08.05 [13,257,142.88 nodes/sec]
This is 469:

[Jun 29 01:14:32 UTC] OGR: Benchmark for core #0 (GARSP 5.13-A)
0.00:00:17.13 [11,600,247 nodes/sec]

 

[Viztech]

Uhhh, I think he was looking for Linux performance, which I am no expert on.

I seem to remember a thread that had the 462? Linux client faster than the 463. Of course this was long before the 469 came out.

Please correct me if I'm wrong or confirm if this is correct.

viz

 

[sciencewhiz]

I was the one that made the old thread about 462 and 463 in linux. On a PIII, the 462 was much faster than the 463 client, but was still far short of either the client in Windows. 469 fixes this and performance in the 469 client is the same in linux and in windows.

This was all done on a PIII, I don't have a AMD system to test with, though I understand that 463 is faster on a athlon than 469.

 

[RC]

On the AMD platform and using OGR:

Under Linux, client version 462 was faster than version 463.
Now that ver 469 has been released it is faster than 462 under Linux.

For Win32, client ver 463 is still the best client to use.

My Duron/Linux/client_469 is about 5% slower compared to my Athlon_Slot_A/Win32/client_463. The five percent diff could be many things, - processor diff, bios setting I need to tweak on my Duron mobo, the compiler used to build the 469 client for Linux may not be greatly optimized for AMD processors, etc...

I have been in contact with a fellow that has been able to rebuild the Linux kernel using gcc 3.0 and he informs me that there is a noticeable performance improvement for AMD processors.

I downloaded gcc 3.0 and the source code for the dnet client, and took a long look at how much effort would be involved to rebuild the kernel and the client. Don't know if I want to go through that much effort in hopes of gaining a couple percentage points improvement with my OGR dnet client running under Linux.

My Duron/Linux/client_469 is just as fast if not faster than my Athlon_Slot_A/Win32/(client_463 or client_469) for RC5.

Sciencewhiz was the one that originally discovered that client 463 under Linux was much slower than ver 462 for OGR users. About a month later I made the same discovery and Sciencewhiz then pointed me to his thread on the issue. I promptly reported the problem to DNet and about nine months later they issued a new OGR core for Linux.

The fastest Linux OGR Dnet client for Tbird will be version 469.

 

[sciencewhiz]

I'm not sure how much good rebuilding the kernal or client would do. It would ne interesting to know, though. I'm pretty sure that dnet only gives you the source for the cores, not the whole clients, but optimizing the core is what would help the most