Titan work restart time ...

[GenTarkin]

Some definitive testing needs to be done in regards to this.
The claim of Titans being really slow on block changes (which I assume has the same penalty as work restarts) , I have been skeptical of.
Prohashing ops claim it as bad as 6-7s before Titan is running full speed.
To test this, I made my worker statically mine krugercoin when its diff was something like .03
My Titan easily sustained a measured 320MH+ in both the 5s avg, avg and pool avg stats as reported by bfgminer.
By the time the worker had registered its speed pool side it was showing 355MH!
For a majority of the work, the Titan processing through SUB 1s block times!!!!
If its true the Titan is severely hampered in speed then w/ fast block changes & work restarts .... maintaining 320MH / 355MH reported pool side ... should not be anywhere near possible when it was doing nearly sub 1s block switches!!!!

Something just doesnt add up w/ these claims about the Titan being slow on block changes / work restarts.
If these claims were true then theres NO WAY the pool worker would have shown 355mh, theres no way bfgminer would have reported 320+ pool average as measured by accepted shares poolside!

So, Im confused lol!

[GenTarkin]

Well, I think I determined the length of restart times.
It seems the dies are still actively working on work, just older work until they get updatd to new work. So, the cubes that are lower on the list in bfgminer have a higher chance of stales, due to getting issued work update many seconds after the first cube is.
Anywho, Ive been experimenting w/ some stuff in the titan driver for bfgminer. I have a 4 cube Titan, I brought its work restart / update times down from 4-6 seconds to 3-4 seconds. I dont know if theres any negative effects on how Ive changed the code. Time will tell.

[excelerator]

That is so cool and congratulations!

[Steve Sokolowski]

Well, I think I determined the length of restart times.
It seems the dies are still actively working on work, just older work until they get updatd to new work. So, the cubes that are lower on the list in bfgminer have a higher chance of stales, due to getting issued work update many seconds after the first cube is.
Anywho, Ive been experimenting w/ some stuff in the titan driver for bfgminer. I have a 4 cube Titan, I brought its work restart / update times down from 4-6 seconds to 3-4 seconds. I dont know if theres any negative effects on how Ive changed the code. Time will tell.

If you have time, could you write a detailed explanation of this in the "news" forum? That would be amazingly helpful to explain why Titans don't work as well as other miners.

Make sure to include a link to your firmware at the bottom. I'll sticky the post so that it both informs customers and increases sales of your firmware.

[GenTarkin]

Well, I think I determined the length of restart times.
It seems the dies are still actively working on work, just older work until they get updatd to new work. So, the cubes that are lower on the list in bfgminer have a higher chance of stales, due to getting issued work update many seconds after the first cube is.
Anywho, Ive been experimenting w/ some stuff in the titan driver for bfgminer. I have a 4 cube Titan, I brought its work restart / update times down from 4-6 seconds to 3-4 seconds. I dont know if theres any negative effects on how Ive changed the code. Time will tell.

If you have time, could you write a detailed explanation of this in the "news" forum? That would be amazingly helpful to explain why Titans don't work as well as other miners.

Make sure to include a link to your firmware at the bottom. I'll sticky the post so that it both informs customers and increases sales of your firmware.

Well its more of like a shot in the dark type modifcation =P ... I disabled some KNC code that seemed to not be doing anything except using unnecessary time when work updates were issued to dies.
Unfortunately being that this mod is part of the bfgminer titan driver ... I cant exactly sell it under my licensing terms since bfgminer is GPL code.
Also, whether it has any actual merit poolside is hard to tell. But, Im pretty sure my rejects/stales are down considerably. Pre-patch my titan was at around 2-3% rejects on prohashing. Now it sits between 1-1.5% ... but I also realize maybe we were mining completely different coins before and after. So its hard to tell.
I've been using h=3 w/ this mod as well.

Anywho ... not sure what Im gonna do yet =P ... Ill keep the thread updated if I decide to proceed any further =)

[gaanthony]

I'm definitely interested in you proceeding further.

[GenTarkin]

Out of curiousity can anyone share what their couple day average reject % is as reported by bfgminer w/ a Titan pointed just at this pool and mining all coins w/ h=5 or lower?
Thanks

[excelerator]

Out of curiousity can anyone share what their couple day average reject % is as reported by bfgminer w/ a Titan pointed just at this pool and mining all coins?
Thanks

Is there a log file on the Pi that records this and I can forward to you?

[GenTarkin]

Out of curiousity can anyone share what their couple day average reject % is as reported by bfgminer w/ a Titan pointed just at this pool and mining all coins?
Thanks

Is there a log file on the Pi that records this and I can forward to you?

No, its just something you would have to glance at the webgui for after the Titan and bfgminer have been running solidly for a couple days.

[gaanthony]

0.8967% on one and 1.5226% on another.