poclbm and phoenix-1.3 no worksize better than -w64 and -w128 (5870,5970)

[marcus_of_augustus]

Okay, here's a tip for free for anybody using the Cypress chip (5870, 5970) ... it seems the new phoenix miner 1.3 goes a bit faster using worksize=64 now not worksize=128 ... and it is same on new poclbm also i think ...i.e.,  -w 64 is now better than  -w 128 ... must be something to with the BFI_INT flag being enabled ... check it out (I waited until new difficulty is set to release this info but have been using this for a few days already).

... you can thank me later with tips  

EDIT: after more testing as per discussion below NO worksize flag or setting (default) is better than -w64 and -w128

[TurdHurdur]

How does it do with no worksize setting?

[marcus_of_augustus]

25Mhash slower here.

Okay, maybe it is just my particular set-up on that rig ...

ubuntu 10.10, SDK 2.1, fglrx 11.2 (AMD drivers built/installed as .deb packages), HD 5970 clock 775/300

[bitjet]

what are you getting for mhps?

[marcus_of_augustus]

what are you getting for mhps?

poclbm flags; -v -w64 -r2 -f1
336.8 Mhash/s

poclbm flags; -v -w128 -r2 -f1
335.1 Mhash/s

phoenix 3 flags; VECTORS AGGRESSION=11 BFI_INT WORKSIZE=64
335.8 MHash/s

[marcus_of_augustus]

How does it do with no worksize setting?

Good question, it is better again ... 338~340 ... so going to edit title of thread.

Can get rid of worksize flag ... less crap on CLI is good.

[TurdHurdur]

How does it do with no worksize setting?

Good question, it is better again ... 338~340 ... so going to edit title of thread.

Can get rid of worksize flag ... less crap on CLI is good.

Same results with my 68[57]0.

[zoro]

w64 is working better for me.
thanks for the tip

[Convery]

On my 5850 clocked 950/300 I get this in phoenix using aggression=13, BFI_INT, VECTORS *;

Workload = 64;
340 MHash/s

Workload = 1024;
363-364 MHash/s

On a Win XP x86 professional using CCC 11.2 and SDK 2.1.
Using a Sapphire 5850 1Gb Toxic, just FYI :3