Quote from: kerney666 on April 01, 2019, 10:00:04 PM

Quote from: tvukoman on April 01, 2019, 09:18:48 PM

Quote from: heavyarms1912 on April 01, 2019, 12:51:43 AM

Quote from: tvukoman on March 31, 2019, 11:15:51 PM

Just Update win 10 pro (to last version) + driver to 19.3.2 version from 18.6.1 because no need for ppt table and read that this driver is OK. Update OverdriveNtool to last 0.28.beta 1 too.

Strange: CNR on 19.3.2 work just like on 18.6.1 driver but Turtle algo with 19.3.2 driver on Vega 56 (64 bios) or Vega 64 (Asus) drop hashrate from 18 kh/s to 10 kh/s

Did someone try turtle algo on 19.3.2?

Seems like I am not the only one. I had similar results. L28+28 gives 10-11 Khs. Try 26+26, it would work fine around 17-18.5 Khs still not as good as other folks who can get L28+28 with earlier drivers.

heavyarms1912 txs!

Confirmed:

Asus Strix Vega 64, 26+26 on 19.3.2 driver results with 19,4 kh/s (on 18.6.1 result was about 18,5 kh/s)
Reference Vega 56 (64bios), 26+26 on 19.3.2. driver 18 kh/s (on 18.6.1 result was equal)

Here is test on 19.3.2 for CNR algo, after testing all configs, this is the best results with power from the wall (for rig). Testing Vega from 13+13 to 16+15 and 5+5 to 8+7 on rx570 with all combinations. EDIT: Reference Vega cards can do more, I put clock down because overheating (small space about two m2 without windows, just small holes and one small bathroom fan ;-))
╔═════════════════════════════╦═════════╦══════╦══════╦══════════╦══════════╦════════╦════════╦═════════╗
║ Card ║ Memory ║ Algo ║ kh/s ║ Core ║ Mem ║ Config ║ Driver ║ Power W ║
╠═════════════════════════════╬═════════╬══════╬══════╬══════════╬══════════╬════════╬════════╬═════════╣
║ Asus Strix Vega 64 ║ Samsung ║ CNR ║ 2210 ║ 1520/920 ║ 1100/910 ║ 16+14 ║ 19.3.2 ║ 787 ║
╠═════════════════════════════╬═════════╬══════╬══════╬══════════╬══════════╬════════╬════════╬═════════╣
║ Reference Vega 56 Gigabyte x 2 ║ Samsung ║ CNR ║ 1970 ║ 1408/950 ║ 1050/950 ║ 14+14 ║ 19.3.2 ║ 787 ║
╠═════════════════════════════╬═════════╬══════╬══════╬══════════╬══════════╬════════╬════════╬═════════╣
║ Rx570 4GB Nitro+ ║ Elpida ║ CNR ║ 940 ║ 1180/850 ║ 1980/900 ║ 7+7 ║ 19.3.2 ║ 787 ║
╚═════════════════════════════╩═════════╩══════╩══════╩══════════╩══════════╩════════╩════════╩═════════╝

As a comparison, I just fire up my rig with 8 x ref Vega 56 flashed to 64, win10, 18.6.1 driver, L24+24, 19.5 kh/s per gpu, no fuss whatsoever. There are some very strange things happening in later drivers, hard to come up with good theories. I'll bump my dev workstation with 580+V64+VII to 19.3.2 shortly, will be interesting to see if I get the same weird behavior. It's also weird that my 18.6.1 rig and tvukoman's V56-flashed-64 on 18.6.1 is getting very different results.

Devs,

Really appreciate if you can report Hynix mem performance as well where most of us have the unfortunate privilege of owning. All the reports on Samsung memory doesn't translate at all to Hynix for the more taxing algo (CN-turtle as an example).