XFX RX 470 4GB Elpida with eliovp's strap that's around the forum for long time with equal TRCDR/A fix, I make around 850h/s with claymore xmr 10.2 , and about 920h/s with gateless gate at 1170core/2000mem/875mv , at about 2-3-4 errors per hour. With claymore xmr 10.2 hashrate is stable with rare low fluctuations to 800-820 , with gateless gate 0.1.3-pre6b fluctuations are a bit more , but averages at about 920-925 , new gateless gate sharp gives me almost the same results , old gateless gate is based on sgminer , so hashrate is about the same as sgminer, for me a little bit more than sgminer-5.5.5-gm-nicehash-8.
edit: With the same strap, same 1170/2000/875 on ETH make 29.5mhs , eth and xmr are stable with these settings , memory can go up to 2050 before significant error rate , but stability goes down with it. Claymore xmr 10.2 settings -h 928 , gateless gate settings 896/8/2 , claymore eth 10.2 -dcri 8
Topic: Custom RAM Timings for GPU's with GDDR5 - DOWNLOAD LINKS - UPDATED - page 3. (Read 155475 times)
August 22, 2018, 05:34:55 PM
August 16, 2018, 09:46:44 PM
I do have the 4gb version too which i can clock to 30mhs no problem. Just the 8gb which is being a pain.
August 16, 2018, 05:32:38 PM
So busy with the mass availability of 8gb Polaris in perfect conditions for less (as low as 290 Australian dollars, thats nearly below 200 usd.... for GOOD variants) also waiting for the Launches next week hesitant on buying some of those Polairs or not..Its not a gamble since prices can't go any cheaper lol.
Anyway, did decipher/experiment with MR0/8 etc. But for ,Hynix Specifically, you need more than that. Some timings are in Nanoseconds, others are based on tCK. I think RASMACTWR is tied to tRFC but funnily until now i'm not sure what rules bind tRC and tRFC in hynix (increasing tRFC and RAS2RAS substantially without adjusting other stats, adds some stability yes but not sure if thats the right thing.)
MISC1 for Hynix AJR 1625 First change the Hex to binary
0x20140475 >> 2014 and 0475 2014 is the MR1 and 0475 is MR0
MR0
*0475> 010001110101, now starting from the right hand side-you need the JEDEC Lookup table from the standard- Write Recovery values are the straight decimal values (101)= 5 so Write Recovery is 5
*The 4 bits after write recovery are CLmrs its equal to tCL-1 , lookup table here doesn't start from zero, but starts at 5, so for our strap, 1110 is 14 adding +5 its 19 but thats CLmrs so tCL is 19-1=18
That was for tCL up to 19 more than 19 is supplemented by an extra bit from MR8, the A0 bit specifically and its added to the left.
*The next bit after CLmrs is useless (TM) ignore it, then the next 4 bits are Write Latency which is also a table lookup and also has an extra fifth bit added from MR8 (because WL values are usually above 20 so the 5th bit is mostly in use) WL 0100 with the 1 bit from MR8 10100 WL look-up table starts from 4 so the decimal value of this Binary (20) is combined with (4), so WL is 24
Hynix doesnt like 20-21 tCL though (changing the binary) for some reason. The higher Hynix straps start with 23
will continue later have to finish my shift!
Anyway, did decipher/experiment with MR0/8 etc. But for ,Hynix Specifically, you need more than that. Some timings are in Nanoseconds, others are based on tCK. I think RASMACTWR is tied to tRFC but funnily until now i'm not sure what rules bind tRC and tRFC in hynix (increasing tRFC and RAS2RAS substantially without adjusting other stats, adds some stability yes but not sure if thats the right thing.)
MISC1 for Hynix AJR 1625 First change the Hex to binary
0x20140475 >> 2014 and 0475 2014 is the MR1 and 0475 is MR0
MR0
*0475> 010001110101, now starting from the right hand side-you need the JEDEC Lookup table from the standard- Write Recovery values are the straight decimal values (101)= 5 so Write Recovery is 5
*The 4 bits after write recovery are CLmrs its equal to tCL-1 , lookup table here doesn't start from zero, but starts at 5, so for our strap, 1110 is 14 adding +5 its 19 but thats CLmrs so tCL is 19-1=18
That was for tCL up to 19 more than 19 is supplemented by an extra bit from MR8, the A0 bit specifically and its added to the left.
*The next bit after CLmrs is useless (TM) ignore it, then the next 4 bits are Write Latency which is also a table lookup and also has an extra fifth bit added from MR8 (because WL values are usually above 20 so the 5th bit is mostly in use) WL 0100 with the 1 bit from MR8 10100 WL look-up table starts from 4 so the decimal value of this Binary (20) is combined with (4), so WL is 24
Hynix doesnt like 20-21 tCL though (changing the binary) for some reason. The higher Hynix straps start with 23
will continue later have to finish my shift!
August 07, 2018, 09:42:37 AM
Quote from: Larvitar
Is tCKE linked with SEQ_MISC1?
Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
regarding tcke, not exactly sure, but 90% it is indirectly dont forget there are other Misc lines not shown. What works though and confirmed by wolf (or thanks to wolf) tCL can tolerate +1 extra difference over the corresponding MISC-1 value. Also, you can change r2w and w2r freely for stability, that makes rules much different and easier, found a 1kh/s hynix (based on 1500 mhz strap though) with this way stable on hynix, non cooled at 2125 (few errors) this card used to do loads of errors @ 2025 mhz! and give 912 h/s with one click.Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
for Deep diving though need to mod a 1625 or 1750 since base tCL is 19, thus 20 when you +1
If your elpida crashes at 1950 with whatever timings, that's similar to mine, whether solder/bga issue or power problems.
Now i fully understand the MR0 ,MR8 etc but the DAMN Binary bits relation to the 4 Hex characters! omg.
My memories are not Elpida, but Micron (not Micron-Elpida). BGA are not a problem, the ex-owner has 4 another cards the same model, and all are bad for memory overclock.
I changed MISC8 to 1500 base, lowered TRP_WRA a little and bumped tRP. To improve hashrate, I did RASMACT fine tuning (yeah, the "trial and error" way). Nice results. For now, 971H/s at 1200/1900. I think we have a "mission complete" for Micron, better than expected results, with much knowledge that can be used to improve Elpida and Samsung.
Elpida RX570 4GB is runing at 1200/2020. 985H/s. I think there is room for improvement over 1KH/s. Maybe getting closer to Samsung (at least in v7)
I find it really strange that your max memory closk is that low, look how high my micron goes on RX560:
https://bitcointalksearch.org/topic/radeon-rx-560-gaming-oc-4g-rev-10-163mhs-bios-3052745
BTW, does anyone have working, stable straps for Hynix and Micron with TRRD=5? Seems Elpida has least problems with it
Me too. But maybe the issue is memory cooling. When the weather is cold, memory can run at higher clock. When temperature rises at 35C+, run at 2000 is a pain.
Quote from: QuirkSilver
Elpida is hilariously funny, glad i started learning with it, no rules (almost!) , so for example for most eplidas an easy solution for low frequency memory (will get hotter though/unstable) try dropping trrd to 4 on the one click strap or SRB bios editior's pimped strap and loosening reads slightly increasing tRFC substantially (thanks to Kueto again who helped with this!) it gives you more stability for negligible speed loss, clock memory to 1875-2025 however, most non cooled memory, will give errors above 1900, still will give 545 h/s -585 h/s (560, will be 1000 h/s++ for 570/80)
Yep. No rules. I'm tuning my RX570 Strix 4GB Elpida, and the Micron techniques are useless lol
August 04, 2018, 03:25:20 PM
This is the first time I've seen this utility. I’m already using OhGodAnETHlargementPill and I’m satisfied with results. I hope that OhGodATool also has such a quality and will help to increase mining performance. Have a success in your great job!
August 01, 2018, 12:07:12 PM
Quote from: Larvitar
Is tCKE linked with SEQ_MISC1?
Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
regarding tcke, not exactly sure, but 90% it is indirectly dont forget there are other Misc lines not shown. What works though and confirmed by wolf (or thanks to wolf) tCL can tolerate +1 extra difference over the corresponding MISC-1 value. Also, you can change r2w and w2r freely for stability, that makes rules much different and easier, found a 1kh/s hynix (based on 1500 mhz strap though) with this way stable on hynix, non cooled at 2125 (few errors) this card used to do loads of errors @ 2025 mhz! and give 912 h/s with one click.Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
for Deep diving though need to mod a 1625 or 1750 since base tCL is 19, thus 20 when you +1
If your elpida crashes at 1950 with whatever timings, that's similar to mine, whether solder/bga issue or power problems.
Now i fully understand the MR0 ,MR8 etc but the DAMN Binary bits relation to the 4 Hex characters! omg.
My memories are not Elpida, but Micron (not Micron-Elpida). BGA are not a problem, the ex-owner has 4 another cards the same model, and all are bad for memory overclock.
I changed MISC8 to 1500 base, lowered TRP_WRA a little and bumped tRP. To improve hashrate, I did RASMACT fine tuning (yeah, the "trial and error" way). Nice results. For now, 971H/s at 1200/1900. I think we have a "mission complete" for Micron, better than expected results, with much knowledge that can be used to improve Elpida and Samsung.
Elpida RX570 4GB is runing at 1200/2020. 985H/s. I think there is room for improvement over 1KH/s. Maybe getting closer to Samsung (at least in v7)
I find it really strange that your max memory closk is that low, look how high my micron goes on RX560:
https://bitcointalksearch.org/topic/radeon-rx-560-gaming-oc-4g-rev-10-163mhs-bios-3052745
BTW, does anyone have working, stable straps for Hynix and Micron with TRRD=5? Seems Elpida has least problems with it
August 01, 2018, 09:38:37 AM
Quote from: Larvitar
Is tCKE linked with SEQ_MISC1?
Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
regarding tcke, not exactly sure, but 90% it is indirectly dont forget there are other Misc lines not shown. What works though and confirmed by wolf (or thanks to wolf) tCL can tolerate +1 extra difference over the corresponding MISC-1 value. Also, you can change r2w and w2r freely for stability, that makes rules much different and easier, found a 1kh/s hynix (based on 1500 mhz strap though) with this way stable on hynix, non cooled at 2125 (few errors) this card used to do loads of errors @ 2025 mhz! and give 912 h/s with one click.Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
for Deep diving though need to mod a 1625 or 1750 since base tCL is 19, thus 20 when you +1
If your elpida crashes at 1950 with whatever timings, that's similar to mine, whether solder/bga issue or power problems.
Now i fully understand the MR0 ,MR8 etc but the DAMN Binary bits relation to the 4 Hex characters! omg.
My memories are not Elpida, but Micron (not Micron-Elpida). BGA are not a problem, the ex-owner has 4 another cards the same model, and all are bad for memory overclock.
I changed MISC8 to 1500 base, lowered TRP_WRA a little and bumped tRP. To improve hashrate, I did RASMACT fine tuning (yeah, the "trial and error" way). Nice results. For now, 971H/s at 1200/1900. I think we have a "mission complete" for Micron, better than expected results, with much knowledge that can be used to improve Elpida and Samsung.
Elpida RX570 4GB is runing at 1200/2020. 985H/s. I think there is room for improvement over 1KH/s. Maybe getting closer to Samsung (at least in v7)
1250 core should bump it over 1k.
in addition, is good to improve by memory optimisation because when OC'ing core the performance gains will be greater
August 01, 2018, 09:20:37 AM
Quote from: Larvitar
Is tCKE linked with SEQ_MISC1?
Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
regarding tcke, not exactly sure, but 90% it is indirectly dont forget there are other Misc lines not shown. What works though and confirmed by wolf (or thanks to wolf) tCL can tolerate +1 extra difference over the corresponding MISC-1 value. Also, you can change r2w and w2r freely for stability, that makes rules much different and easier, found a 1kh/s hynix (based on 1500 mhz strap though) with this way stable on hynix, non cooled at 2125 (few errors) this card used to do loads of errors @ 2025 mhz! and give 912 h/s with one click.Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
for Deep diving though need to mod a 1625 or 1750 since base tCL is 19, thus 20 when you +1
If your elpida crashes at 1950 with whatever timings, that's similar to mine, whether solder/bga issue or power problems.
Now i fully understand the MR0 ,MR8 etc but the DAMN Binary bits relation to the 4 Hex characters! omg.
My memories are not Elpida, but Micron (not Micron-Elpida). BGA are not a problem, the ex-owner has 4 another cards the same model, and all are bad for memory overclock.
I changed MISC8 to 1500 base, lowered TRP_WRA a little and bumped tRP. To improve hashrate, I did RASMACT fine tuning (yeah, the "trial and error" way). Nice results. For now, 971H/s at 1200/1900. I think we have a "mission complete" for Micron, better than expected results, with much knowledge that can be used to improve Elpida and Samsung.
Elpida RX570 4GB is runing at 1200/2020. 985H/s. I think there is room for improvement over 1KH/s. Maybe getting closer to Samsung (at least in v7)
1250 core should bump it over 1k.
August 01, 2018, 08:19:55 AM
Quote from: Larvitar
Is tCKE linked with SEQ_MISC1?
Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
regarding tcke, not exactly sure, but 90% it is indirectly dont forget there are other Misc lines not shown. What works though and confirmed by wolf (or thanks to wolf) tCL can tolerate +1 extra difference over the corresponding MISC-1 value. Also, you can change r2w and w2r freely for stability, that makes rules much different and easier, found a 1kh/s hynix (based on 1500 mhz strap though) with this way stable on hynix, non cooled at 2125 (few errors) this card used to do loads of errors @ 2025 mhz! and give 912 h/s with one click.Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
for Deep diving though need to mod a 1625 or 1750 since base tCL is 19, thus 20 when you +1
If your elpida crashes at 1950 with whatever timings, that's similar to mine, whether solder/bga issue or power problems.
Now i fully understand the MR0 ,MR8 etc but the DAMN Binary bits relation to the 4 Hex characters! omg.
My memories are not Elpida, but Micron (not Micron-Elpida). BGA are not a problem, the ex-owner has 4 another cards the same model, and all are bad for memory overclock.
I changed MISC8 to 1500 base, lowered TRP_WRA a little and bumped tRP. To improve hashrate, I did RASMACT fine tuning (yeah, the "trial and error" way). Nice results. For now, 971H/s at 1200/1900. I think we have a "mission complete" for Micron, better than expected results, with much knowledge that can be used to improve Elpida and Samsung.
Elpida RX570 4GB is runing at 1200/2020. 985H/s. I think there is room for improvement over 1KH/s. Maybe getting closer to Samsung (at least in v7)
July 31, 2018, 05:45:38 PM
[/quote]
Is tCKE linked with SEQ_MISC1?
Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
[/quote]
regarding tcke, not exactly sure, but 90% it is indirectly dont forget there are other Misc lines not shown. What works though and confirmed by wolf (or thanks to wolf) tCL can tolerate +1 extra difference over the corresponding MISC-1 value. Also, you can change r2w and w2r freely for stability, that makes rules much different and easier, found a 1kh/s hynix (based on 1500 mhz strap though) with this way stable on hynix, non cooled at 2125 (few errors) this card used to do loads of errors @ 2025 mhz! and give 912 h/s with one click.
for Deep diving though need to mod a 1625 or 1750 since base tCL is 19, thus 20 when you +1
If your elpida crashes at 1950 with whatever timings, that's similar to mine, whether solder/bga issue or power problems.
Now i fully understand the MR0 ,MR8 etc but the DAMN Binary bits relation to the 4 Hex characters! omg.
July 31, 2018, 07:25:47 AM
I'm quite confused, but knew how it works from the Jedec212 document. I can't set tCL to 20 or 21 to this garbage Hynix AJR even Hynix MJR straps are hilariously made as if hynix is trolling people not to have a soft spot for mining, MJR have 1750 strap (19 CAS) and behold the one next to it is (23 CAS) lol, because cryptonight is affected by writes, both WR, WL are in that MR0 binary inside the damn SEQ-MISC-1.
I tried to borrow straps from other vendors, but the tCKE parameters are different, not sure what excatly would work, then again have to experiment alot. So i rather find that CAS latency.
The lower frequency, i managed to squeeze 975 h/s from that (damaged elpida 570) @1860mhz surely with a tight 1425 strap didn't even hone it correctly though.
Is tCKE linked with SEQ_MISC1?I tried to borrow straps from other vendors, but the tCKE parameters are different, not sure what excatly would work, then again have to experiment alot. So i rather find that CAS latency.
The lower frequency, i managed to squeeze 975 h/s from that (damaged elpida 570) @1860mhz surely with a tight 1425 strap didn't even hone it correctly though.
Nice hash from damaged elpidas
About my lastest tests:
- No matter what I do with straps, at 1950+ I have incorrect ETH shares or compute errors in CN mining). 1930 is fine;
- Tried SEQ-MISQ1 and CAS from 1500 strap. Mining doesn't starts, bsod after closing miner.
- Tried tFAW32 bumped to 6. 12H/s loss. Sticking at 4.
July 30, 2018, 10:40:03 AM
I'm quite confused, but knew how it works from the Jedec212 document. I can't set tCL to 20 or 21 to this garbage Hynix AJR even Hynix MJR straps are hilariously made as if hynix is trolling people not to have a soft spot for mining, MJR have 1750 strap (19 CAS) and behold the one next to it is (23 CAS) lol, because cryptonight is affected by writes, both WR, WL are in that MR0 binary inside the damn SEQ-MISC-1.
I tried to borrow straps from other vendors, but the tCKE parameters are different, not sure what excatly would work, then again have to experiment alot. So i rather find that CAS latency.
The lower frequency, i managed to squeeze 975 h/s from that (damaged elpida 570) @1860mhz surely with a tight 1425 strap didn't even hone it correctly though.
I tried to borrow straps from other vendors, but the tCKE parameters are different, not sure what excatly would work, then again have to experiment alot. So i rather find that CAS latency.
The lower frequency, i managed to squeeze 975 h/s from that (damaged elpida 570) @1860mhz surely with a tight 1425 strap didn't even hone it correctly though.
July 30, 2018, 09:15:32 AM
Quote from: QuirkSilver
Its not a generic shit, the Fujipoly, Laird and Arctic etc. have higher heat dissipation
Yup, these brands are known as good. But I bought really generic shit (lol)Quote from: QuirkSilver
Lavritar, apologies for the abbreviation mistakes, i was writing quickly during night shift. the RASMACT timings are to compensate for having mixed strap i can't fully understand this, but you do that when you get the CAS timings (along with SEQ-MISC-1!!) from a higher strap, but drop tRC for cryptonight, the RASMACT is to compensate (again i'm not sure) when you do this, the equation isn't necessary, at least as long as the sum of rasmac+actr is in the higher than tCR side of course.
don't change tCRCWL, bump tfaw32 to 4 or 6 at least to try.
Nice. That information is useful. You don't need any apologies, your post helped me a lot.don't change tCRCWL, bump tfaw32 to 4 or 6 at least to try.
tFAW32 is bumped to 4. No performance penalty. I'll try 6. About the CAS/SEQ-MISC-1 while we can't understand/decode, the best way to optimize is mix default timings for lower straps (to increase performance per clock) oh higher straps (to increase overclockability). Didn't Nerdralph's tool worked for you? (you mentioned some posts before).
Let's try to find the limit at lower clocks
July 28, 2018, 12:22:59 PM
Its not a generic shit, the Fujipoly, Laird and Arctic etc. have higher heat dissipation, although i also thought each chip barley consumes 2 watts, but for lower end GPUs anything is better than nothing at all, the ram die is non heat conductive, imo , roughly, anything of 6 w/k-m +++ is useful if its attached to the heatsink directly, thats why i really like those gigabyte Aorus, not expensive but memory have an engraved spot in the heatsink and a thin thermal pad is there.
Those Hynix 4gb 580s (3 of them) with one click strap, can go 2100-2200mhz (although the strap is bad) compared to MSI gaming X 570s, same memory not cooled, they do 2020-2090 mhz.
The generic or high end is useless unless heat is dissipated somehow! also i would use thin layered (0.5-1 mm) thermal pad + a bigger shim or heatsink if you have your own idea of attaching thermal pads+shim or whatsoever to the gpu heatsink then thats perfect.
Those generic thin silicon pads are for cheaper GPUs 560, however i will get something better but with fair price/quantiity ( i have 15 GPUs in total half of which are 560s, so I'm a poor miner
)
Lavritar, apologies for the abbreviation mistakes, i was writing quickly during night shift. the RASMACT timings are to compensate for having mixed strap i can't fully understand this, but you do that when you get the CAS timings (along with SEQ-MISC-1!!) from a higher strap, but drop tRC for cryptonight, the RASMACT is to compensate (again i'm not sure) when you do this, the equation isn't necessary, at least as long as the sum of rasmac+actr is in the higher than tCR side of course.
don't change tCRCWL, bump tfaw32 to 4 or 6 at least to try.
Do you do BGA reballing? i have got almost everything for GPU modding (but damn, to buy 1 DIE of Samsung memory, its better just to buy a full 580 8gb Samsung going 220$ here used on ebay)
I have a gpu that BSODs at anything higher than 1865mhz.. its an Asus strix 570 its NOT memory errors, i never get any errors at whatever timings at this frequency....it also shows VDDCI amperage 208 amps in HWinfo!
another is 560 2gb hynix, that crashes and gives errors with whatever timings barely makes 1850mhz!!
Appreciate if anyone can tell me how to decipher those MC-SEQ-1 reading a post by laik from 22nd march 2017 in this thread, still struggling with the crap.
Those Hynix 4gb 580s (3 of them) with one click strap, can go 2100-2200mhz (although the strap is bad) compared to MSI gaming X 570s, same memory not cooled, they do 2020-2090 mhz.
The generic or high end is useless unless heat is dissipated somehow! also i would use thin layered (0.5-1 mm) thermal pad + a bigger shim or heatsink if you have your own idea of attaching thermal pads+shim or whatsoever to the gpu heatsink then thats perfect.
Those generic thin silicon pads are for cheaper GPUs 560, however i will get something better but with fair price/quantiity ( i have 15 GPUs in total half of which are 560s, so I'm a poor miner
)Lavritar, apologies for the abbreviation mistakes, i was writing quickly during night shift. the RASMACT timings are to compensate for having mixed strap i can't fully understand this, but you do that when you get the CAS timings (along with SEQ-MISC-1!!) from a higher strap, but drop tRC for cryptonight, the RASMACT is to compensate (again i'm not sure) when you do this, the equation isn't necessary, at least as long as the sum of rasmac+actr is in the higher than tCR side of course.
don't change tCRCWL, bump tfaw32 to 4 or 6 at least to try.
Do you do BGA reballing? i have got almost everything for GPU modding (but damn, to buy 1 DIE of Samsung memory, its better just to buy a full 580 8gb Samsung going 220$ here used on ebay)
I have a gpu that BSODs at anything higher than 1865mhz.. its an Asus strix 570 its NOT memory errors, i never get any errors at whatever timings at this frequency....it also shows VDDCI amperage 208 amps in HWinfo!
another is 560 2gb hynix, that crashes and gives errors with whatever timings barely makes 1850mhz!!
Appreciate if anyone can tell me how to decipher those MC-SEQ-1 reading a post by laik from 22nd march 2017 in this thread, still struggling with the crap.
July 26, 2018, 07:31:48 PM
How are those aliexpress thermal pads comparable with real stuff?
https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
Are nothing. Big diference from a generic shit to a good product.https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
About the tests:
Increasing RASMACACTWR to 69 and RASMACACTRD to 55 gaves me extra 20H/s. So, at 1150/1900 the RX470 is mining atr 938H/s. Yep, lowering tRC to 60 and TRCDW(A) to 13 gives me extra 4H/s, but I think the stability cost dont worth it.
Increasing RASMACACTWR in Elpida strap gaves me some hashes too. 946H/s at 1150/2020. It wasn't the goal, but good to see the Polaris shining. Nice
942 h/s , @1150 core, @2050 mem, 52W (gpu-z)
777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715
July 26, 2018, 06:40:30 PM
How are those aliexpress thermal pads comparable with real stuff?
https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
Are nothing. Big diference from a generic shit to a good product.https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
About the tests:
Increasing RASMACACTWR to 69 and RASMACACTRD to 55 gaves me extra 20H/s. So, at 1150/1900 the RX470 is mining atr 938H/s. Yep, lowering tRC to 60 and TRCDW(A) to 13 gives me extra 4H/s, but I think the stability cost dont worth it.
Increasing RASMACACTWR in Elpida strap gaves me some hashes too. 946H/s at 1150/2020. It wasn't the goal, but good to see the Polaris shining. Nice
942 h/s , @1150 core, @2050 mem, 52W (gpu-z)
777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715
July 26, 2018, 03:36:09 PM
How are those aliexpress thermal pads comparable with real stuff?
https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
Are nothing. Big diference from a generic shit to a good product.https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
About the tests:
Increasing RASMACACTWR to 69 and RASMACACTRD to 55 gaves me extra 20H/s. So, at 1150/1900 the RX470 is mining atr 938H/s. Yep, lowering tRC to 60 and TRCDW(A) to 13 gives me extra 4H/s, but I think the stability cost dont worth it.
Increasing RASMACACTWR in Elpida strap gaves me some hashes too. 946H/s at 1150/2020. It wasn't the goal, but good to see the Polaris shining. Nice
942 h/s , @1150 core, @2050 mem, 52W (gpu-z)
777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715
July 26, 2018, 02:30:43 PM
How are those aliexpress thermal pads comparable with real stuff?
https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
Are nothing. Big diference from a generic shit to a good product.https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
About the tests:
Increasing RASMACACTWR to 69 and RASMACACTRD to 55 gaves me extra 20H/s. So, at 1150/1900 the RX470 is mining atr 938H/s. Yep, lowering tRC to 60 and TRCDW(A) to 13 gives me extra 4H/s, but I think the stability cost dont worth it.
Increasing RASMACACTWR in Elpida strap gaves me some hashes too. 946H/s at 1150/2020. It wasn't the goal, but good to see the Polaris shining. Nice
July 26, 2018, 09:53:15 AM
How are those aliexpress thermal pads comparable with real stuff?
https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
https://www.amazon.com/s/ref=nb_sb_ss_c_1_8?url=search-alias%3Delectronics&field-keywords=fujipoly&sprefix=fujipoly%2Celectronics%2C267
July 26, 2018, 09:30:53 AM
Same as mine that goes up to 2300...but to go that high, cooling should be made
in such way that metal plate covers not only GPU, but memory, too (and replacing
usually bad factory thermal paste with quality one)
However, 2100 should be easy to achieve
Also, important is to find sweet spot between voltage and temperature (voltage increase
improves stability, but only up to the point where increased heat starts to be a problem)
I bought it from a miner, so it came with some thermalpads and a good biosmod, but not all ICs are cooled (only 5/8) because the other three arent in the cooler contact area. I ordered some heatsinks at Aliexpress but it will take a little time to arrive (3 months. Yeah, Brazil's mail is poor).in such way that metal plate covers not only GPU, but memory, too (and replacing
usually bad factory thermal paste with quality one)
However, 2100 should be easy to achieve
Also, important is to find sweet spot between voltage and temperature (voltage increase
improves stability, but only up to the point where increased heat starts to be a problem)
About the memory strap: The board came to me with Matthev timing. I tried many timings (heavyarms, yours, some for anorak's) and the original one (matthev) was best (at least at 1900~2000 clocks). I did some changes/tests and for now I have it:
777000000000000022CC1C00CE615C41C0571016B30CD50900400700140514207A8900A00300000 010122F3F9D2C3617
Is there a way to improve "performance per clock" for Cryptonight? (considering that it couldn't run at 2000+)
0.5 mm or 1 mm and a copper shim, 1 mm thick or more https://www.aliexpress.com/item/20-pcs-15mmx15mm-0-3mm-0-4mm-0-5mm-0-6mm-0-8mm-1mm-Heatsink-Copper/32826278552.html?spm=a2g0s.9042311.0.0.3dbd4c4dpaqYyI
I experiment with the thicknesses so that the stock heatsink presses the shim and the pad slightly without being thicker than the processor height (been there done that, and the GPU fan was squealing
) the more spacious the copper shims the better, however, i do this on rough estimates, that the memory die consumes 1-2 watts each (reasonable?) the shims not touching the aluminum directly is not a big problem, because copper blown by the gpu fan imho is more than sufficient to cool the memory. Most of the Aorus 580s i have memory cooled from the factory reach 2100-2200mhz (8 gb, 4gb all of them hynix) i found thermal pads being 0.5 mm and perfectly matched with the heatsink, but, as a consequence of mining these pads will dry within months (become brittle)Do you really clock those elpidas to these clocks? lol? how do you cool them? what sort of timings for such clocks!?
I bought these ones (0.5mm): https://www.aliexpress.com/item/20-pcs-15mmx15mm-0-3mm-0-4mm-0-5mm-0-6mm-0-8mm-1mm-Heatsink-Copper/32826278552.html?spm=a2g0s.9042311.0.0.5311b90arAsvdv
And these thermalpads: https://www.aliexpress.com/item/2016-high-quality-10mm-10mm-1mm-100-pcs-Thermal-Pad-GPU-CPU-Heatsink-Cooling-Conductive-Silicone/32675801589.html?spm=a2g0s.9042311.0.0.5311b90arAsvdv
Quote from: QuirkSilver
2- tCR of the stock strap, needs to be loosened, its tied to tCL, tRCDR,tRP, tRRD are tied to tCR (not tRP in elpida, thats why you can have a very low tCR value) so any +1/-1 to any of those, must be subsequently, done tCR value.
example,
tcrdr 23, tRRD 7,tCL 18, tRP 22 tCR=68
trcdr 24 (+1), tRRD 6 (-1), tCL 18, tRP 22 tCR= (68)
tCR=68 should be tRC?example,
tcrdr 23, tRRD 7,tCL 18, tRP 22 tCR=68
trcdr 24 (+1), tRRD 6 (-1), tCL 18, tRP 22 tCR= (68)
Here I'm with this:
TCRDR(A) 24, tRRD 6, tCL 22, tRP 20, tRC 65 (tried to low tCL to 20, but didn't worked)
24 + 5 + 22 + 20 = 71. So I would have to bump tRC to 71?
Quote from: QuirkSilver
3-RASMACTRD,RASMACTWR are tied to ACTRD and ACTWR by the rule, RASMACTRD(/W) + ACTRD(/W)= tCR+1 (check any stock strap, same thing) so when you drop writes (-actwr) you should increase RASMACTWR
Yep, I see some gains by loosening RASMACTWR.870H/s at 50
917H/s at 63
921H/s at 66
But the formula you gave didn't fit here (or I dont understood):
ACTWR (18) + best RASMACTWR (66) = 84
tRC (65) + 1 = 66
Quote from: QuirkSilver
4- Raw Precharge time, tRP must be higher than tCL, it doesnt affect Cryptonight much so its one of the keys to attain high stable straps (by loosening it)
5-loosening tCL is the key to get a higher strap, but now that's the biggest problem, tCL in elpida isn't tied to tCR (that's why you see that one click strap, being good with a low tRC), however, CAS timings (tCL,W2R,R2W) are linked to SEQ-MISC-1
If you changed the tCL of the 1625 strap to the 1750 one without the SEQ-MISC-1 of the 1750 one, will end in BSODs or similar.
Micron one click, took, Elpida's 1500mhz strap and mixed it, so higher tCL mainly, and that works..
Nice!5-loosening tCL is the key to get a higher strap, but now that's the biggest problem, tCL in elpida isn't tied to tCR (that's why you see that one click strap, being good with a low tRC), however, CAS timings (tCL,W2R,R2W) are linked to SEQ-MISC-1
If you changed the tCL of the 1625 strap to the 1750 one without the SEQ-MISC-1 of the 1750 one, will end in BSODs or similar.
Micron one click, took, Elpida's 1500mhz strap and mixed it, so higher tCL mainly, and that works..
tRP is at 20, tCL at 22. So I will try it at 22 or 23.
The 5th tip is intersting. Maybe the issue with tCL at 20 (I tried it today) can be linked with it. I will see the strap and find where tCL 20 are, and copy SEC-MISC-1.
Quote from: QuirkSilver
6-increase tRFC substantially, have minimal impact on CN speeds, but gives more stability (+25% or whatever, noticed no real rules here.)
tRFC is a controverse timing. It gives pain while overclocking memories since DDR2 ;pAbout looseining it: considering I can't go above 2000 for while (maybe because memory cooling, maybe because the shitty ICs) lowering tRFC (from 157 to 129) was a solution to gain some hashes (around 10H/s at 1900).
Quote from: Branko
Micron and micron-elpida are quite a different story, both in timings and max clock.
Also, your estimate about memory chips power consumption is way too low,
"RX 480 uses 40-50W for its 8GB"
Source:
https://www.gamersnexus.net/guides/3032-vega-56-cost-of-hbm2-and-necessity-to-use-it
So for 8GB cards almost half the consumption goes to memory chips
Holy feijoada!Also, your estimate about memory chips power consumption is way too low,
"RX 480 uses 40-50W for its 8GB"
Source:
https://www.gamersnexus.net/guides/3032-vega-56-cost-of-hbm2-and-necessity-to-use-it
So for 8GB cards almost half the consumption goes to memory chips
I don't understand why it comes to us without proper cooling, because it needs. Unafortunately lowering voltage needs hardware mods (could help sometimes).
I really want to thank you three for all knowledge! You are awesome!
Jump to: