Attention Windows OGR crackers & Athlon users, Change to the 8010.463b Client

[Viztech]

I became concerned with some major differences in the benchmarks from the thread that Aberent started on OGR 8009.460 clients. http://forums.anandtech.com/messageview.cfm?start=1&catid=39&threadid=186513 I realized that there are some MAJOR differences with the performance of the 8009.460, 8010.462 and the 8010.463b client which is the latest available.

Here are some results from my herd...
Averages of 3 OGR long benchmark runs per computer.

Athlon 600 (average of two machines)
8009.460 6423 nodes/mhz
8010.463b 7715 nodes/mhz
This is a 20.1% improvement!

Celeron 366@550
8009.460 5549 nodes/mhz
8010.463b 6848 nodes/mhz
Again a 23% improvement!

K6-233@233
8010.462 4714 nodes/mhz
8010.463b 6448 nodes/mhz
A whopping 36.8% improvement!

In checking the client version info at Mika's ~40% of the valid (ver 8009 & 8010 clients combined) OGR stubs are being done with the slower 8009 clients, and who knows how many with the slower 8010.462 clients!

As a bonus, the Athlons saw a 4.3% RC5 speed improvement with the 8010.463b clients. I'll post another thread on that one, as well as a new OGR 463b client bench thread.

Linux users mileage may vary! see this thread for details. http://forums.anandtech.com/messageview.cfm?catid=39&threadid=201585 The 462 looks like it is best on Linux.

viz

 

[Eug]

Heheh, so my Celeron II isn't as amazing as I thought afterall. (Running ~ 6870-6910 nodes/s/MHz.) It didn't make sense that it was a PIII 933 and Athlon killer.

http://forums.anandtech.com/messageview.cfm?catid=39&threadid=186513

Well, at least my "loss" is everyone else's gain.

Happy cracking!

 

[Eug]

Just ran the old 462 client (Thanks Joe O). I confirm a 23-24% improvement over the 462 client, with both of my Celeron machines using 463.

 

[Jani]

With my Celeron 533@624 I got 4,392,418 nodes / sec <=> 7039 nodes / MHz And P3 450@504 3,430,246 nodes / sec <=> 6806 nodes / MHz. Celeron is running without anti-virus and firewall software, so I think that makes the difference.
Client: dnetc v2.8010-463-CTR-00071214 for Win32

 

[Joe O]

I have preliminary results from a Celeron 300a @450 that show a 23% improvement going to 463b from 462.
Also an AMDK6-2 @400 shows a 23% improvement while in use over the best performance under 462 when idle.

 

[Joe O]

The overnight results from a Celeron 300a @450 show a 23.8% improvement going to 463b from 462.
The overnight results from an AMDK6-2 @400 show a 31.8% improvement going to 463b from 462.
The overnight results from a Pentium MMX @200 show only a 10.0% improvement going to 463b from 462.

 

[Viztech]

Just checked Mika's stats- there are still 40% of the clients out there using the slower 8009.460 client, and who nows how many with the slower 8010.462.

We could probably get a 10% Team improvement by upgrading those old clients to the 8010.463b.

So lets get out there and upgrade those clients, because TA is going to win OGR!!

Urmm, thought I was ViRGE there for a minute, dagnabit!

viz

 

[Viztech]

Update-

We're still getting 27% of the OGR work at Mika's done with the slower 8009.460 client! Much better than 40%, but we can do better.

Upgrade, Upgrade as soon as you can!

 

[TwoFace]

Some people are reporting slower OGR with the 463 on linux than with the 462... dunno yet but it seemed a little slow to me so I'm trying the 462 now... Get back to you later on this issue tho'!

 

[Viztech]

Doh!

Slower is not good

I'll modify that title to say Windows users.

viz

 

[rancherlee]

My athlon 700 went from 6500 to 8000 nodes/sec/mhz
23% INCREASE!!!!

 

[narzy]

^ TTT

 

[Viztech]

There's still 23% of our Windows/OGR work being done with the 8009.460 clients...

Upgrade!

viz