Pages:
Author

Topic: AMD Mem Tweak XL - Read/modify timings/pp/straps on the fly - page 54. (Read 58905 times)

newbie
Activity: 31
Merit: 0
This is my lighter timing, hopefully stable:
--CL 19 --RAS 30 --RCDRD 12 --RCDWR 6 --RC 44 --RP 13 --RRDS 5 --RRDL 5 --RTP 4 --FAW 18 --CWL 6 --WTRS 4 --WTRL 9 --WR 15 --WRRD 1 --RDWR 18 --REF 17000 --RFC 248

please test it.

What about Vega 56 Samsung and Hynix ?
legendary
Activity: 1510
Merit: 1003
This is my lighter timing, hopefully stable:
--CL 19 --RAS 30 --RCDRD 12 --RCDWR 6 --RC 44 --RP 13 --RRDS 5 --RRDL 5 --RTP 4 --FAW 18 --CWL 6 --WTRS 4 --WTRL 9 --WR 15 --WRRD 1 --RDWR 18 --REF 17000 --RFC 248

please test it.
latest 19.4.1 drivers. win10
vega64 lc ~1360/1100@875 51.4mhs with claymore 12. But hbm temp is hot 70C and one weaker vega gives sometimes red incorrect shares

22.3 kh/s trtl, testing for pool speed and hw errors
newbie
Activity: 4
Merit: 0
This is my lighter timing, hopefully stable:
--CL 19 --RAS 30 --RCDRD 12 --RCDWR 6 --RC 44 --RP 13 --RRDS 5 --RRDL 5 --RTP 4 --FAW 18 --CWL 6 --WTRS 4 --WTRL 9 --WR 15 --WRRD 1 --RDWR 18 --REF 17000 --RFC 248

please test it.

51.72 on PM4.2C   VEGA56@64BIOS   CORE:1100,875,   MEM:1100,875
jr. member
Activity: 144
Merit: 2
This is my lighter timing, hopefully stable:
--CL 19 --RAS 30 --RCDRD 12 --RCDWR 6 --RC 44 --RP 13 --RRDS 5 --RRDL 5 --RTP 4 --FAW 18 --CWL 6 --WTRS 4 --WTRL 9 --WR 15 --WRRD 1 --RDWR 18 --REF 17000 --RFC 248

please test it.
member
Activity: 658
Merit: 86
Very nice finding!

BTW I managed to reach 51.6mh, but in CN-R I can't get past 2150h/s no matter what i try.
What is the trick there?

at what clocks on cnr?
1100 mem, 1500 core
I just raised the core, hr went up to ~2200, but that is the max.

But maybe the miner sw is not the fastest on CN-R, it's a fork of XMRig.
Going to try TRM

Update: with teamredminer on default setting it's 2300 h/s Smiley

okay.  was gonna mention we're most likely limited on core freq on cnr.
2500+ is possible but with very high core oc of 1.75 Mhz.  Making it lot less efficient.

The current public TRM release isn't the best for these modded timings with tightened latency. That said, it's most probably better than all other options Smiley. We're working on a version that we believe will take better advantage of the new fancy pants tools though. With a high core clk you can compensate and reach those max perf hashrates anyway, but with my current working version I'm doing 2318 h/s using 1407@900/1107@900, so those higher core clks aren't necessary. Note: this is on a Vega 64 LC, which can handle the super tight timings much better than my 8 x Vega 56 flashed 64.

Running a ton of tests on those Vega 56s flashed 64 tonight, feels like this new dimension will increase the complexity a lot. It was never trivial to truly trim those Vega rigs for efficiency and get everything rock solid stable, and with the additional degrees of freedom available now, man...


So, Rig with 6 Vega 64 about 1320 Watts + fan, Mobo.= 1420 and sometimes fluctuates to ...
Imagine the power socket  could handle max  in Ampere ? Could triggered burn at Cable and Socket

Well, it's always the same story when new shiny tools come out: first it's the max-perf circus, trying to break the world record in hashrate on a single Vega 64 (which is fun in itself  Grin), then the more efficiency-focused tuning takes over. We've just seen the beginning of the journey for Vega timings imho.
newbie
Activity: 6
Merit: 0
ANdSqxxLkaYao5EQXWegcpPzyfHE22a6go
R9 390 works?
newbie
Activity: 31
Merit: 0
Very nice finding!

BTW I managed to reach 51.6mh, but in CN-R I can't get past 2150h/s no matter what i try.
What is the trick there?

at what clocks on cnr?
1100 mem, 1500 core
I just raised the core, hr went up to ~2200, but that is the max.

But maybe the miner sw is not the fastest on CN-R, it's a fork of XMRig.
Going to try TRM

Update: with teamredminer on default setting it's 2300 h/s Smiley

okay.  was gonna mention we're most likely limited on core freq on cnr.
2500+ is possible but with very high core oc of 1.75 Mhz.  Making it lot less efficient.

The current public TRM release isn't the best for these modded timings with tightened latency. That said, it's most probably better than all other options Smiley. We're working on a version that we believe will take better advantage of the new fancy pants tools though. With a high core clk you can compensate and reach those max perf hashrates anyway, but with my current working version I'm doing 2318 h/s using 1407@900/1107@900, so those higher core clks aren't necessary. Note: this is on a Vega 64 LC, which can handle the super tight timings much better than my 8 x Vega 56 flashed 64.

Running a ton of tests on those Vega 56s flashed 64 tonight, feels like this new dimension will increase the complexity a lot. It was never trivial to truly trim those Vega rigs for efficiency and get everything rock solid stable, and with the additional degrees of freedom available now, man...


So, Rig with 6 Vega 64 about 1320 Watts + fan, Mobo.= 1420 and sometimes fluctuates to ...
Imagine the power socket  could handle max  in Ampere ? Could triggered burn at Cable and Socket
newbie
Activity: 31
Merit: 0
I changed a few things in the timings from todxx/kerney to (hopefully) improve compatibility with more air cooled Vega 64s w/ Samsung:



--rp 12 --rc 45 --rrds 3 --rrdl 3 --rcdrd 12 --rcdwr 5

This should also work for bios flashed 56 -> 64 w/ Samsung. However, I noticed hashrate drop overnight. If anyone has ideas lmk.

For Vega 56s w/ Samsung and original bios:

--rp 12 --rc 37 --rrds 3 --rrdl 3 --rcdrd 12 --rcdwr 5 --ras 25

Also based off of recommended settings from todxx/kerney, changed "rc" to work with Vega 56 w/ Samsung and original bios. Does not work with Hynix.

Edit: added --ras 25 so that RC = RAS + RP


Vega 56 Hynix, memory is limited for OC
Polor max 930, above gonna crash

Headache for tweak
member
Activity: 277
Merit: 23
Current values on RX470/480/570 & 580. Any help what to change first?

get the optimized values from the public straps with a decoder tool to use with amdmemtweak

Yeah as a noob to all this modding that didn't even make sense. Are you saying to open something like overdriventool and find out what my current straps are, then somehow modify those straps via the winamdtweak? Thanks for your patie4nce.

No, he means find a recommended public BIOS which is good for your card and download it. Use SRBpolaris to open modded BIOS and decode the values. This way you will see what changes you need to dial in to AMDMT
newbie
Activity: 42
Merit: 0
Very nice finding!

BTW I managed to reach 51.6mh, but in CN-R I can't get past 2150h/s no matter what i try.
What is the trick there?

at what clocks on cnr?
1100 mem, 1500 core
I just raised the core, hr went up to ~2200, but that is the max.

But maybe the miner sw is not the fastest on CN-R, it's a fork of XMRig.
Going to try TRM

Update: with teamredminer on default setting it's 2300 h/s Smiley

okay.  was gonna mention we're most likely limited on core freq on cnr.
2500+ is possible but with very high core oc of 1.75 Mhz.  Making it lot less efficient.

The current public TRM release isn't the best for these modded timings with tightened latency. That said, it's most probably better than all other options Smiley. We're working on a version that we believe will take better advantage of the new fancy pants tools though. With a high core clk you can compensate and reach those max perf hashrates anyway, but with my current working version I'm doing 2318 h/s using 1407@900/1107@900, so those higher core clks aren't necessary. Note: this is on a Vega 64 LC, which can handle the super tight timings much better than my 8 x Vega 56 flashed 64.

Running a ton of tests on those Vega 56s flashed 64 tonight, feels like this new dimension will increase the complexity a lot. It was never trivial to truly trim those Vega rigs for efficiency and get everything rock solid stable, and with the additional degrees of freedom available now, man...


dear developer, thanks for the hark work for the great tool trm, so how to use this tools in vega 56 Samsung and hynix memory, what 's the settings will be for cnr on these cards, thanks
full member
Activity: 729
Merit: 114
Guys, sorry if I missed this info prior, but how are you ensuring that timings are passed to the right gpus? I'd normally go off PCI bus IDs, but how are gpus ordered in this tool? Nothing is jumping out at me as a way to identify gpus and make sure looser timings are applied to gpus that throw errors in the mining software.

it's bus ordered
you can see the bus id on each gpu card properties in dev mgr.
Or alternatively just open OverdriveNTool which is already in sorted in bus order.
jr. member
Activity: 148
Merit: 5
Guys, sorry if I missed this info prior, but how are you ensuring that timings are passed to the right gpus? I'd normally go off PCI bus IDs, but how are gpus ordered in this tool? Nothing is jumping out at me as a way to identify gpus and make sure looser timings are applied to gpus that throw errors in the mining software.
newbie
Activity: 33
Merit: 0
Current values on RX470/480/570 & 580. Any help what to change first?

get the optimized values from the public straps with a decoder tool to use with amdmemtweak

Yeah as a noob to all this modding that didn't even make sense. Are you saying to open something like overdriventool and find out what my current straps are, then somehow modify those straps via the winamdtweak? Thanks for your patie4nce.
jr. member
Activity: 144
Merit: 2
Current values on RX470/480/570 & 580. Any help what to change first?

get the optimized values from the public straps with a decoder tool to use with amdmemtweak
newbie
Activity: 33
Merit: 0
Current values on RX470/480/570 & 580. Any help what to change first?

GPU 0:  Ellesmere [Radeon RX 470/480/570/570X/580/580X/590]
CAS
  CL: 24          W2R: 17         R2R: 5          CCLD: 3         R2W: 31         NOPR: 0         NOPW: 0
RAS
  RC: 87          RRD: 7          RCDRA: 29       RCDR: 29        RCDWA: 17       RCDW: 17
MISC
  RFC: 219        TRP: 27         RP_RDA: 30      RP_WRA: 63
MISC2
  WDATATR: 0      T32AW: 8        CRCWL: 25       CRCRL: 3        FAW: 12         PA2WDATA: 0     PA2RDATA: 0
DRAM1
  RASMACTWR: 38   RASMACTRD: 31   ACTWR: 9        ACTRD: 16
DRAM2
  RAS2RAS: 119    RP: 28          WRPLUSRP: 34    BUS_TURN: 13
GPU 1:  Ellesmere [Radeon RX 470/480/570/570X/580/580X/590]
CAS
  CL: 24          W2R: 17         R2R: 5          CCLD: 3         R2W: 31         NOPR: 0         NOPW: 0
RAS
  RC: 87          RRD: 7          RCDRA: 29       RCDR: 29        RCDWA: 17       RCDW: 17
MISC
  RFC: 219        TRP: 27         RP_RDA: 30      RP_WRA: 63
MISC2
  WDATATR: 0      T32AW: 8        CRCWL: 26       CRCRL: 3        FAW: 12         PA2WDATA: 0     PA2RDATA: 0
DRAM1
  RASMACTWR: 36   RASMACTRD: 30   ACTWR: 9        ACTRD: 15
DRAM2
  RAS2RAS: 112    RP: 26          WRPLUSRP: 32    BUS_TURN: 13
GPU 2:  Ellesmere [Radeon RX 470/480/570/570X/580/580X/590]
CAS
  CL: 25          W2R: 19         R2R: 5          CCLD: 2         R2W: 29         NOPR: 0         NOPW: 0
RAS
  RC: 79          RRD: 7          RCDRA: 29       RCDR: 29        RCDWA: 23       RCDW: 23
MISC
  RFC: 219        TRP: 29         RP_RDA: 33      RP_WRA: 70
MISC2
  WDATATR: 0      T32AW: 8        CRCWL: 25       CRCRL: 3        FAW: 12         PA2WDATA: 0     PA2RDATA: 0
DRAM1
  RASMACTWR: 30   RASMACTRD: 27   ACTWR: 13       ACTRD: 16
DRAM2
  RAS2RAS: 119    RP: 28          WRPLUSRP: 38    BUS_TURN: 13
GPU 3:  Ellesmere [Radeon RX 470/480/570/570X/580/580X/590]
CAS
  CL: 19          W2R: 21         R2R: 5          CCLD: 2         R2W: 24         NOPR: 0         NOPW: 0
RAS
  RC: 83          RRD: 8          RCDRA: 27       RCDR: 27        RCDWA: 19       RCDW: 19
MISC
  RFC: 197        TRP: 27         RP_RDA: 30      RP_WRA: 62
MISC2
  WDATATR: 0      T32AW: 9        CRCWL: 25       CRCRL: 2        FAW: 14         PA2WDATA: 0     PA2RDATA: 0
DRAM1
  RASMACTWR: 35   RASMACTRD: 30   ACTWR: 10       ACTRD: 15
DRAM2
  RAS2RAS: 107    RP: 25          WRPLUSRP: 33    BUS_TURN: 12
GPU 4:  Ellesmere [Radeon RX 470/480/570/570X/580/580X/590]
CAS
  CL: 24          W2R: 17         R2R: 5          CCLD: 3         R2W: 31         NOPR: 0         NOPW: 0
RAS
  RC: 87          RRD: 7          RCDRA: 29       RCDR: 29        RCDWA: 17       RCDW: 17
MISC
  RFC: 219        TRP: 27         RP_RDA: 30      RP_WRA: 63
MISC2
  WDATATR: 0      T32AW: 8        CRCWL: 25       CRCRL: 3        FAW: 12         PA2WDATA: 0     PA2RDATA: 0
DRAM1
  RASMACTWR: 36   RASMACTRD: 29   ACTWR: 9        ACTRD: 15
DRAM2
  RAS2RAS: 111    RP: 26          WRPLUSRP: 31    BUS_TURN: 12
jr. member
Activity: 144
Merit: 2
also I heard there's a tool that can bring down stable voltages by a LOT...
member
Activity: 658
Merit: 86
Very nice finding!

BTW I managed to reach 51.6mh, but in CN-R I can't get past 2150h/s no matter what i try.
What is the trick there?

at what clocks on cnr?
1100 mem, 1500 core
I just raised the core, hr went up to ~2200, but that is the max.

But maybe the miner sw is not the fastest on CN-R, it's a fork of XMRig.
Going to try TRM

Update: with teamredminer on default setting it's 2300 h/s Smiley

okay.  was gonna mention we're most likely limited on core freq on cnr.
2500+ is possible but with very high core oc of 1.75 Mhz.  Making it lot less efficient.

The current public TRM release isn't the best for these modded timings with tightened latency. That said, it's most probably better than all other options Smiley. We're working on a version that we believe will take better advantage of the new fancy pants tools though. With a high core clk you can compensate and reach those max perf hashrates anyway, but with my current working version I'm doing 2318 h/s using 1407@900/1107@900, so those higher core clks aren't necessary. Note: this is on a Vega 64 LC, which can handle the super tight timings much better than my 8 x Vega 56 flashed 64.

Running a ton of tests on those Vega 56s flashed 64 tonight, feels like this new dimension will increase the complexity a lot. It was never trivial to truly trim those Vega rigs for efficiency and get everything rock solid stable, and with the additional degrees of freedom available now, man...
jr. member
Activity: 144
Merit: 2
okay.  was gonna mention we're most likely limited on core freq on cnr.
2500+ is possible but with very high core oc of 1.75 Mhz.  Making it lot less efficient.

Okay, i'm on the track again...
currently 2370h/s @1550MHz core and 1100MHz mem (~180W reported by driver, so sth like 220W on the wall)
full member
Activity: 729
Merit: 114
Very nice finding!

BTW I managed to reach 51.6mh, but in CN-R I can't get past 2150h/s no matter what i try.
What is the trick there?

at what clocks on cnr?
1100 mem, 1500 core
I just raised the core, hr went up to ~2200, but that is the max.

But maybe the miner sw is not the fastest on CN-R, it's a fork of XMRig.
Going to try TRM

Update: with teamredminer on default setting it's 2300 h/s Smiley

okay.  was gonna mention we're most likely limited on core freq on cnr.
2500+ is possible but with very high core oc of 1.75 Mhz.  Making it lot less efficient.
newbie
Activity: 33
Merit: 0
Any findings for us Polaris non bios modded folks on cryptonight people?
Pages:
Jump to: