pool-gpu-engine and pool-gpu-memclock not evaluated correctly when switching algorithms

[kenshirothefist]

I'm using different Engine and/or Memory settings on a per-pool basis and auto-gpu option (which is basically essential since it enables temp-cutoff functionality). Whenever one of the pool-xxx setting sets a high/low threshold it is remained in the next algorithm switch even if settings are set in a fixed "low-high" manner.

Example:

global settings:
"gpu-engine" : "1000-1000",
"gpu-memclock" : "1500-1500",
"temp-cutoff" : "95",
"temp-overheat" : "85",
"temp-target" : "75",
"auto-fan" : true,
"auto-gpu" : true,

pool_x11:
"pool-gpu-engine" : "1000-1000",
"pool-gpu-memclock" : "1500-1500",

pool_scrypt-N:
"pool-gpu-engine" : "947-947",
"pool-gpu-memclock" : "1250-1250"

Miner starts at x11, all is fine, gpu engine fixed at 1000, mem at 1500, I see
"GPU engine clock autotune is enabled (1000-1000)"

I switch to scrypt-N and I observe this:
"GPU engine clock autotune is enabled (947-1000)"

If temps are low, GPU is forced to 1000 and causes issues. This is wrong, since I specified "pool-gpu-engine" : "947-947" and I would expect
"GPU engine clock autotune is enabled (947-947)"

---

Similar issue is when switching back to x11, I see:
"GPU engine clock autotune is enabled (947-1000)"

If temps are above threshold, GPU is throttled to 947 and decreases performance, but I don't want this, I just wan't Engine speed to be fixed and only being completely cut-off at cut-off temp. Since I specified "pool-gpu-engine" : "1000-1000" I would expect
"GPU engine clock autotune is enabled (1000-1000)"

---

Exactly the same goes with Memory settings.

You should make sure Engine and Memory settings are set and applied exactly as specified by the "pool-gpu-engine" and "pool-gpu-memclock". Moreover, I do believe that all the "pool-xxx" settings should be revised and verified if they are all properly and 100% correctly applied when switching algorithms (maybe it would be wise to add logging when using -D with the output of all active "pool-xxx" settings after they are applied at algorithm switch).

[platinum4]

I second this issue: it's pretty clearly obvious these settings don't take in the pool-area, the only way to lock a GPU's speed or memclock right now is in the lower part of the .conf. This defeats the purpose of having these two available settings if they don't take precedence. Is this a cascade/prioritization issue? We saw similar issues with pool-intensity earlier, to where it would flip down (keccak I:13) but never back up after a algo flip to x11/x13 (I:19). The intensity has since been fixed.

[ystarnaud]

I believe ef722e6 should resolve the gpu-engine issue. I tested this by having one pool use 1100-1100 and the next pool use 1000-1000,1065-1065. Upon switching back and forth between pools, the GPUs all had their respective values.

Unless I'm mistaken, gpu-memclock does not accept ranges, only whole numbers. If you would like the memory clock to be linked to the core clock, you can always try gpu-memdiff.