Did you get Ethminer to work with RX 5700? I tried v0.19.0-alpha miner but the miner shut down. Please help.
Hi guys again,
so, I've been messing around with the Ethminer Opencl Kernel, and I think I found out what is wrong with the RX 5700 family for mining ETH on OpenCL 2.0:
(Gonna need some help fixing it)
Let's Start:
After you download the new release's source code here:
https://github.com/ethereum-mining/ethminer/archive/v0.19.0-alpha.0.tar.gzExtract it and find the kernels/cl/ folder,
Edit the ethash.cl file, change this line(236):
mix = fnv(mix, g_dag[buffer[lane_idx]].uint8s[thread_id]); \
to this:
/* mix = fnv(mix, g_dag[buffer[lane_idx]].uint8s[thread_id]);*/ \
compile and run the miner!
That line is responsible for bottlenecking the hole code!
I am able to achieve 300+ MH/s without that line of code!
(LOOOOL not mining properly, of course, I get only rejected shares!)
So:
My thoughts on this, and after some testing, the problem is MEMORY ACCESS!
let's get back to the code:
mix = fnv(mix, g_dag[buffer[lane_idx]].uint8s[thread_id]); \
if I change
for a defined value eg. '0',
It runs @ ~80MH/s (I assume It's the real Speed),
but of course again, REJECTED SHARES, because
is supposed to change according to 'lane_idx'!
Bottom line is, fetching this piece of code's value from memory is slowing down the hole thing!
Conclusion:
I need Help, since I'm no ETH nor OpenCL Expert...
We can assume from this the Theoretical Seed of ~80MH/s for mining ETH on the RX 5700 XT!
(not bad, but I was Expecting 100+ MH/s)
(maybe with specific optimizations, driver upgrade, etc, we could get better performance)