Pages:
Author

Topic: Custom RAM Timings for GPU's with GDDR5 - DOWNLOAD LINKS - UPDATED - page 7. (Read 155475 times)

jr. member
Activity: 176
Merit: 2
Hi All,

any advice memory strap Hynix 8GB H5GQ8H24MJR for Cryptonight V7 ?
I tried create custom from Polaris bios editor one click timing, max I can achieve only 940 h/s with core clock 1200 and memory clock 2250.
I tried to set TRRD = 5 but it will not stable.

Appreciate any help which value need to set higher and which value need to set lower so can get more than 1000 h/s.

Thanks.
jr. member
Activity: 196
Merit: 1
Not even, Unlike AMD, Nvidia's timings are really good already and are already using more bandwidth % than most AMD modded RX 400-500! I think i got something like +0.2~0.3 MH/s on a GTX 1070, it's really not impressive.

What about 1060 with Hynix? Can it be ETHlarged to be on par with other memory types on ethash?
I think it have the same issue.

Hynix in 1060 is bad, and maybe it only could be "saved" by a complete strap mod, not only RRD/FAW mod.
member
Activity: 413
Merit: 17
Not even, Unlike AMD, Nvidia's timings are really good already and are already using more bandwidth % than most AMD modded RX 400-500! I think i got something like +0.2~0.3 MH/s on a GTX 1070, it's really not impressive.

What about 1060 with Hynix? Can it be ETHlarged to be on par with other memory types on ethash?
jr. member
Activity: 196
Merit: 1
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

AFAIK Nvidia timings are not changeable, be it GDDR5 or GDDR5X

I was under the same impression until some time ago. I thought its possible to have something in the works, and it turns out i was correct to myself. It is not possible for the public. There are some stuff, which are confidential and allow some sort of control.

Of course, I didn't say its impossible, but keys are distributed to card manufacturers and probably under very strict NDA

Those card makers have to be able to program their cards firmware, after all Smiley

I'm very curious about this. But I think it wont bee available to common users.

It can be a new biosmod revolution

OhGodACompany has solution to edit and make nvidia bios mods and could thus sell mods, the first problem is the hardware can reject the change. And to actually hardware flash it you need a eeprom hardware flasher and the onboard security can refuse to load the bios, as in soft brick it/refuse to boot GPU. This is because the signature included in whatever bios you modded this in does not allow/include the mod as valid. Our mod is considered as invalid or corrupted.

Mind you this is a problem for GTX cards, not the mining cards.

The second problem:

I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

Not even 10-20% on an 1060?
That would be also worth it.

Not even, Unlike AMD, Nvidia's timings are really good already and are already using more bandwidth % than most AMD modded RX 400-500! I think i got something like +0.2~0.3 MH/s on a GTX 1070, it's really not impressive.

About the first problem:
That's why your company did it by software. I realize you can apply it by software like "ETHlargement" with custom mods (not just 50MH/s, but more). But about the mods directly in the BIOS, I think you could ask for help from OC'ers.

I explain:
When the mods was considered impossible, there was no reason to spend time in validation bypass. But showing it's possible, more people can try it, even devs that are not miners (OC'ers that want to take advantage in some benchmark to win a competition).

If it works, you could still selling the "premium" timings (according your post you have a tool to decode/encode the timings, like OhGodADecode) and more people could make their own mods only with RRD/FAW.

It's good to see GDDR5 is good without straps. We had a black box, so we didin't knew about the best hipothetical performance. Now we know. 0.3MH/s is 1%. Ya, is "meh", and would attract many people asking for more performance (and not donating at least 1 satoshi). Understand you dont release the mod for then.
legendary
Activity: 1302
Merit: 1068
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

AFAIK Nvidia timings are not changeable, be it GDDR5 or GDDR5X

I was under the same impression until some time ago. I thought its possible to have something in the works, and it turns out i was correct to myself. It is not possible for the public. There are some stuff, which are confidential and allow some sort of control.

Of course, I didn't say its impossible, but keys are distributed to card manufacturers and probably under very strict NDA

Those card makers have to be able to program their cards firmware, after all Smiley

I'm very curious about this. But I think it wont bee available to common users.

It can be a new biosmod revolution

OhGodACompany has solution to edit and make nvidia bios mods and could thus sell mods, the first problem is the hardware can reject the change. And to actually hardware flash it you need a eeprom hardware flasher and the onboard security can refuse to load the bios, as in soft brick it/refuse to boot GPU. This is because the signature included in whatever bios you modded this in does not allow/include the mod as valid. Our mod is considered as invalid or corrupted.

Mind you this is a problem for GTX cards, not the mining cards.

The second problem:

I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

Not even 10-20% on an 1060?
That would be also worth it.

Not even, Unlike AMD, Nvidia's timings are really good already and are already using more bandwidth % than most AMD modded RX 400-500! I think i got something like +0.2~0.3 MH/s on a GTX 1070, it's really not impressive.
fgm
newbie
Activity: 43
Merit: 0
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

Not even 10-20% on an 1060?
That would be also worth it.
full member
Activity: 162
Merit: 100
Hello

I have Asus 570 Expedition, and cannot have more 26mh/s on Windows or Hiveos but just for one unit...
It's Hynix H5GC4H24AJR.

Original :
2000:BBB000000000000022889D0073EE8D53805515133ECF560C004E26017E0514206A8900A00200312 01C143840C5303F17
1750:999000000000000022559D0031627C489055131339CDD50A004C06017D0514206A8900A00200312 019123037AD2C3A17
1625:999000000000000022559D0010DE7B4480551312B78C450A004C0601750414206A8900A00200312 018112D34A42A3816

Custom :
1500-2000:999000000000000022559D0010DE5B4480551312B74C450A00400600750414206A8900A00200312 010112D34A42A3816

2050/825mv for Mem
1150/750 for Core

That work on many units except one.

There is another strap or tips in this case ?
jr. member
Activity: 196
Merit: 1
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

AFAIK Nvidia timings are not changeable, be it GDDR5 or GDDR5X

I was under the same impression until some time ago. I thought its possible to have something in the works, and it turns out i was correct to myself. It is not possible for the public. There are some stuff, which are confidential and allow some sort of control.

Of course, I didn't say its impossible, but keys are distributed to card manufacturers and probably under very strict NDA

Those card makers have to be able to program their cards firmware, after all Smiley

I'm very curious about this. But I think it wont bee available to common users.

It can be a new biosmod revolution
sr. member
Activity: 2506
Merit: 319
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

AFAIK Nvidia timings are not changeable, be it GDDR5 or GDDR5X

I was under the same impression until some time ago. I thought its possible to have something in the works, and it turns out i was correct to myself. It is not possible for the public. There are some stuff, which are confidential and allow some sort of control.

Of course, I didn't say its impossible, but keys are distributed to card manufacturers and probably under very strict NDA

Those card makers have to be able to program their cards firmware, after all Smiley
jr. member
Activity: 194
Merit: 4
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

AFAIK Nvidia timings are not changeable, be it GDDR5 or GDDR5X

I was under the same impression until some time ago. I thought its possible to have something in the works, and it turns out i was correct to myself. It is not possible for the public. There are some stuff, which are confidential and allow some sort of control.
full member
Activity: 729
Merit: 114
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

AFAIK Nvidia timings are not changeable, be it GDDR5 or GDDR5X

I think laik2 has successfully experimented with Nvidia cards.
sr. member
Activity: 2506
Merit: 319
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.

AFAIK Nvidia timings are not changeable, be it GDDR5 or GDDR5X
legendary
Activity: 1302
Merit: 1068
I'm guessing this thread is expected to be AMD, but GTX 1060-1070ti are also GDDR5, has anyone made leeway there?

I seem to find that the expected timing enhancements do not bring much enhancement. GDDR5X has been a lot more interesting but i suppose that would be OT.
sr. member
Activity: 588
Merit: 251
SRBminer is windows only. I do not plan to start using windows for mining.
PP: At the end , efficiency is the ultimate goal. Chasing higher hashrate sometimes leads to lowering efficiency.

The performance/efficiency tradeoff in ethash is becoming less significant with newer hardware.  With a R9 290 card it was very difficult to maximize performance without increasing power consumption.  With Polaris, where performance is mainly limited by memory and not the core clock, it is much easier to get near peak performance with minimal power.

That said, I think there is room for some tangible improvements in ethash (and probably cryptonight) mining efficiency through optimizing the mining kernel.  For example a Rx 580 doesn't need active 36 CUs to do 30Mh/s; it's closer to 1 CU per Mh.
sr. member
Activity: 652
Merit: 266
SRBminer is windows only. I do not plan to start using windows for mining.
PP: At the end , efficiency is the ultimate goal. Chasing higher hashrate sometimes leads to lowering efficiency.
jr. member
Activity: 196
Merit: 1
Tighten rfc, *wr just a bit, push core to 1200-1266, leave mclk same as ethash( mine are mostly 2150+ ), set voltage ~900mV and start hashing.
This gives ~1kH at just 640W @wall for 6x RX580 mixed.

laik, try SRBMiner with "trial and error" tests to find the best intensity. I'm reaching 1000H/s in RX570 Elpida (CryptonightV7) with no-so tweaked timings. Are your 580 4 or 8GB variants?

I'll try your tips. Thank your for your support!
sr. member
Activity: 652
Merit: 266
Tighten rfc, *wr just a bit, push core to 1200-1266, leave mclk same as ethash( mine are mostly 2150+ ), set voltage ~900mV and start hashing.
This gives ~1kH at just 640W @wall for 6x RX580 mixed.
full member
Activity: 729
Merit: 114
The stock votlage for the 580 is 1150mv for 1380Mhz core. Same as the RX 480. The 480 runs at 1300Mhz core. It can be safely undervolted to 1050mv. The 580 can be undervolted to 1075mv (this particualr one, the 2nd 580 i have can go 1050mv). Those cards can do 900-1050mv if they are underclocked as well (the 470 can do 975mv at 1225Mhz core).

So you already know why the power consumption is high.  And that makes your efficiency at 1070/140w around ~7.7 while it can be 9.5-10.
RX570/RX580 can do 1200-1250 @ 850-875mV.  In fact I run all my RX570/RX580 in those range.  Some cards need 887-900mV.
eg. https://anorak.tech/uploads/default/original/2X/d/d540bbb14e2c506f2c8efe62c3d475cedd44885f.JPG (HWinfo shows chip power is 62w.  Actual wattage from wall is 95-100w)
jr. member
Activity: 194
Merit: 4
Since its stock for that model, i have it undervolted by 75mv, down to 1075.
GPU-Z says 94w.100w idle, 220w at wall. With a Corsair TX850M PSU.The efficiency is about 90-93% (cant measure it actually, trusting specs).

220w at wall? That's quite a bit. what are the other components?  When I said 100w that's from the wall for the GPU only.

Maybe i didnt write it properly:

GPU in GPU-Z while idle: 5w
GPU in GPU-Z while mining: ~94-100w
Whole system while idle: 100w
Whole system while mining: 220w
My RX470 4GB make 1040 h/s with ~60W in GPU-Z ... 100W is to much in GPU-Z, but 100W from the wall is ok.

No idea how it pushes 60w reported in GPU-Z and 100W on wall. Maybe my bench is a bit unreliable for GPU-Z numbers?

Anyway, tested again.
100w system idle.
XMR_Cast-Vega - 1050h/s, 215w wall.
XMR_RIG - 1063h/s, 215w wall (1% devfee)
sgminer-gm 5.5.6 - 1073h/s, 220w wall.
GPU-Z reported about 4w less on the Cast-Vega, but the load was a bit more inconsistent.


EDIT: Yea, benching is PITA.

Idle system dropped to 95w now.
Cast-Vega, Rig, and Stak are doing ~210w at wall now, at same speed.
Sgminer is doing 215w at wall, same speed. The power consumption numbers are a bit unreliable.

Your core voltage might be high to run at 1380 clocks.  I am guessing higher than 1.05v? 
Vegas do 1950-2000 @ 160w from the wall. 
RX470/RX480/RX570/RX580 can do 900-1050 @ 100-105w from the wall (varies based on memory timing straps used)

The stock votlage for the 580 is 1150mv for 1380Mhz core. Same as the RX 480. The 480 runs at 1300Mhz core. It can be safely undervolted to 1050mv. The 580 can be undervolted to 1075mv (this particualr one, the 2nd 580 i have can go 1050mv). Those cards can do 900-1050mv if they are underclocked as well (the 470 can do 975mv at 1225Mhz core).
full member
Activity: 729
Merit: 114
Since its stock for that model, i have it undervolted by 75mv, down to 1075.
GPU-Z says 94w.100w idle, 220w at wall. With a Corsair TX850M PSU.The efficiency is about 90-93% (cant measure it actually, trusting specs).

220w at wall? That's quite a bit. what are the other components?  When I said 100w that's from the wall for the GPU only.

Maybe i didnt write it properly:

GPU in GPU-Z while idle: 5w
GPU in GPU-Z while mining: ~94-100w
Whole system while idle: 100w
Whole system while mining: 220w
My RX470 4GB make 1040 h/s with ~60W in GPU-Z ... 100W is to much in GPU-Z, but 100W from the wall is ok.

No idea how it pushes 60w reported in GPU-Z and 100W on wall. Maybe my bench is a bit unreliable for GPU-Z numbers?

Anyway, tested again.
100w system idle.
XMR_Cast-Vega - 1050h/s, 215w wall.
XMR_RIG - 1063h/s, 215w wall (1% devfee)
sgminer-gm 5.5.6 - 1073h/s, 220w wall.
GPU-Z reported about 4w less on the Cast-Vega, but the load was a bit more inconsistent.


EDIT: Yea, benching is PITA.

Idle system dropped to 95w now.
Cast-Vega, Rig, and Stak are doing ~210w at wall now, at same speed.
Sgminer is doing 215w at wall, same speed. The power consumption numbers are a bit unreliable.

Your core voltage might be high to run at 1380 clocks.  I am guessing higher than 1.05v? 
Vegas do 1950-2000 @ 160w from the wall. 
RX470/RX480/RX570/RX580 can do 900-1050 @ 100-105w from the wall (varies based on memory timing straps used)
Pages:
Jump to: