1.25v+ on Vega64? - Page 2 - Overclock.net - An Overclocking Community

Forum Jump: 

1.25v+ on Vega64?

Reply
 
Thread Tools
post #11 of 20 (permalink) Old 02-27-2018, 01:19 AM
New to Overclock.net
 
Spacebug's Avatar
 
Join Date: Dec 2013
Posts: 66
Rep: 2 (Unique: 2)
Now, to reiterate, the problem I had with the boost algo lowering clocks when more vcore was used due to higher power consumption was that i could not just increase p7 clocks to compensate.
Sometimes when the game or bench didn't pull as much power from the gpu clocks would increase to unstable frequencies and crash.
Do test this yourself if you want, might be card to card variances in how it behaves and perhaps at what voltage and power consumption the negative performance scaling begin.

This boost problem was lessend a lot for me by hardmod to the current sense circuit, now boost is still active and for me, constantly overboosts from p7 clocks since it senses so low power consumption.
But clocks don't fluctuate much depending on load for me, so it is rather predictable.
I set 1747 p7 clock and get around 1770 when gaming and in benches, +- a few mhz.
So from my experience the boost algorithm seems to be more aggressive and less predictable in its clock changes once you start approaching the higher end of power consumption readings, say 300w and up.

Yeah, gpu prices are insane today, when i bought the card some months after launch i got a demo/open box vega64 for less than intro price of vega56.
Had prices been what they are now I'm not sure i would have taken the soldering iron to it that easily as there is always a chance you screw up something.

Pictures from me wont do you any good I'm afraid since after i checked all solder joints were good i covered much of it in hot glue to protect it from me accidentally nudging a soldered resistor and breaking the solder joint.

I carefully studied this video from buildzoid and used that as a guide

About halfway through should the hardmod related stuff begin.
I'm not sure what resistor value i finally ended up with using though.
I think I first tried 2.7kohm and that worked out to push the voltage where negative performance scaling started from 1.2V to i believe 1.25V.
I decided that wasn't enough so i soldered another resistor of some value in parallell.
If someone would want to know I'm sure it can be calculated as the card senses about 40% of actual power reading, so if im not mistaken that would be reducing the initial resistance from the existing cap to the existing resistor to 40% if it's initial value.
I think i have the resistor values written up somewhere, can see if i can dig them up if you want.

A fair warning though about hardmod to power sense, obviously, any kind of power based protection shutdown or throttling wont work as intended as it sees false power readings.

Last edited by Spacebug; 02-27-2018 at 01:23 AM.
Spacebug is offline  
Sponsored Links
Advertisement
 
post #12 of 20 (permalink) Old 03-09-2018, 10:02 AM - Thread Starter
New to Overclock.net
 
whiteskymage's Avatar
 
Join Date: Oct 2013
Posts: 104
Rep: 0
Quote: Originally Posted by Spacebug View Post
Now, to reiterate, the problem I had with the boost algo lowering clocks when more vcore was used due to higher power consumption was that i could not just increase p7 clocks to compensate.
Sometimes when the game or bench didn't pull as much power from the gpu clocks would increase to unstable frequencies and crash.
Do test this yourself if you want, might be card to card variances in how it behaves and perhaps at what voltage and power consumption the negative performance scaling begin.

This boost problem was lessend a lot for me by hardmod to the current sense circuit, now boost is still active and for me, constantly overboosts from p7 clocks since it senses so low power consumption.
But clocks don't fluctuate much depending on load for me, so it is rather predictable.
I set 1747 p7 clock and get around 1770 when gaming and in benches, +- a few mhz.
So from my experience the boost algorithm seems to be more aggressive and less predictable in its clock changes once you start approaching the higher end of power consumption readings, say 300w and up.

Yeah, gpu prices are insane today, when i bought the card some months after launch i got a demo/open box vega64 for less than intro price of vega56.
Had prices been what they are now I'm not sure i would have taken the soldering iron to it that easily as there is always a chance you screw up something.

Pictures from me wont do you any good I'm afraid since after i checked all solder joints were good i covered much of it in hot glue to protect it from me accidentally nudging a soldered resistor and breaking the solder joint.

I carefully studied this video from buildzoid and used that as a guide
https://youtu.be/vvwE-uwcjNE

About halfway through should the hardmod related stuff begin.
I'm not sure what resistor value i finally ended up with using though.
I think I first tried 2.7kohm and that worked out to push the voltage where negative performance scaling started from 1.2V to i believe 1.25V.
I decided that wasn't enough so i soldered another resistor of some value in parallell.
If someone would want to know I'm sure it can be calculated as the card senses about 40% of actual power reading, so if im not mistaken that would be reducing the initial resistance from the existing cap to the existing resistor to 40% if it's initial value.
I think i have the resistor values written up somewhere, can see if i can dig them up if you want.

A fair warning though about hardmod to power sense, obviously, any kind of power based protection shutdown or throttling wont work as intended as it sees false power readings.
So did you solder the resistors the exact same way he did and it all worked out? He said he is using the same 2.7kohms. I wonder if I can put the backplate if I mod this thing too.
whiteskymage is offline  
post #13 of 20 (permalink) Old 03-09-2018, 06:06 PM
New to Overclock.net
 
Spacebug's Avatar
 
Join Date: Dec 2013
Posts: 66
Rep: 2 (Unique: 2)
Quote: Originally Posted by whiteskymage View Post
So did you solder the resistors the exact same way he did and it all worked out? He said he is using the same 2.7kohms. I wonder if I can put the backplate if I mod this thing too.
Yes, soldered the resistors the same way, but not the same values, had a look at my notes now.
But no, unless you are somewhat skilled in soldering and use small components i guess it will be too bulky for the backplate to fit.

At first i soldered 1270ohm resistors, reduced the resistance to 384ohm.
Now it took just over 1.3v before the boost algorithm started reduce clocks,
however I wanted more Vcore so I soldered another resistor in parallell (forgot value) to drop the resistance to 221ohm.
That was enough to never notice the negative performance/clock scaling up to 1.4V, at wich point I had the display output randomly dropping out.
I think that was the "normal" amd issue where too much vcore cause the display output to drop, so I decided that was probably way too much voltage anyway and backed off.

Now a few days ago i soldered another resistor in parallell to drop the resistance down to 104ohm
Not cause i needed more power but I wanted to reduce the Vdroop, a side effect of this current sense mod is cause the voltage controller sees less current the LLC won't drop the voltage in load as far as it does by default.
Less Vdroop would mean I could set lower voltage for P7 and still get the same voltage during load, so far so good


Now, a negative but logical issue ive encountered is because the current sense circuit is tricked no power regulation/restrictions will work as intended, and the gpu can freely draw all power it wants.
Ive had the PSU trip ocp on numerous occasions at voltages i could pass benchmarks just fine but not during gaming.
I chose rise of the tomb raider as that always seems to draw more power (total system) then benchmarks, from my experience.

I don't think I've overdrawn the total power rating of the PSU (ax1200i), wall power meter shows a max reading of 1108W, but could be a slow meter.
But more probably, cause it is from my understanding a multirail PSU it probably sense a sudden, short term current spike from the gpu and trips OCP.
That would not happen on stock card cause the power limit would keep overall powerdraw down and i guess even prevent
harsh current spikes?
Anyhow, it is just a balance act to find out how high voltage i can set before PSU OCP trips.
Though i plan to test the multirail theory with 8pin pcie splitters to 2 6pin pcie.
That way i could spread the load from gpu across 4 rails/wires and see if OCP still trips.
That would tell if it is overall system current draw that trips OCP or just the current from the 2 8pin pcie rails now feeding the gpu.
Hoping that will work cause i don't want to fork out for another PSU even if a higher rated singel rail model would be better.
If not then i will just have to back off some on the gpu OC.


One thing is clear though, Vega without power restrictions and with a bit more voltage will pull some serious power.
Around 1100W from the wall is what i would have guessed a more normally overclocked 2x Vega setup would pull, and ive done it with just one card
Admittedly that powerdraw includes an overclocked FX platform, not really known for power efficiency...
Spacebug is offline  
Sponsored Links
Advertisement
 
post #14 of 20 (permalink) Old 03-10-2018, 09:43 AM - Thread Starter
New to Overclock.net
 
whiteskymage's Avatar
 
Join Date: Oct 2013
Posts: 104
Rep: 0
Quote: Originally Posted by Spacebug View Post
Yes, soldered the resistors the same way, but not the same values, had a look at my notes now.
But no, unless you are somewhat skilled in soldering and use small components i guess it will be too bulky for the backplate to fit.

At first i soldered 1270ohm resistors, reduced the resistance to 384ohm.
Now it took just over 1.3v before the boost algorithm started reduce clocks,
however I wanted more Vcore so I soldered another resistor in parallell (forgot value) to drop the resistance to 221ohm.
That was enough to never notice the negative performance/clock scaling up to 1.4V, at wich point I had the display output randomly dropping out.
I think that was the "normal" amd issue where too much vcore cause the display output to drop, so I decided that was probably way too much voltage anyway and backed off.

Now a few days ago i soldered another resistor in parallell to drop the resistance down to 104ohm
Not cause i needed more power but I wanted to reduce the Vdroop, a side effect of this current sense mod is cause the voltage controller sees less current the LLC won't drop the voltage in load as far as it does by default.
Less Vdroop would mean I could set lower voltage for P7 and still get the same voltage during load, so far so good


Now, a negative but logical issue ive encountered is because the current sense circuit is tricked no power regulation/restrictions will work as intended, and the gpu can freely draw all power it wants.
Ive had the PSU trip ocp on numerous occasions at voltages i could pass benchmarks just fine but not during gaming.
I chose rise of the tomb raider as that always seems to draw more power (total system) then benchmarks, from my experience.

I don't think I've overdrawn the total power rating of the PSU (ax1200i), wall power meter shows a max reading of 1108W, but could be a slow meter.
But more probably, cause it is from my understanding a multirail PSU it probably sense a sudden, short term current spike from the gpu and trips OCP.
That would not happen on stock card cause the power limit would keep overall powerdraw down and i guess even prevent
harsh current spikes?
Anyhow, it is just a balance act to find out how high voltage i can set before PSU OCP trips.
Though i plan to test the multirail theory with 8pin pcie splitters to 2 6pin pcie.
That way i could spread the load from gpu across 4 rails/wires and see if OCP still trips.
That would tell if it is overall system current draw that trips OCP or just the current from the 2 8pin pcie rails now feeding the gpu.
Hoping that will work cause i don't want to fork out for another PSU even if a higher rated singel rail model would be better.
If not then i will just have to back off some on the gpu OC.


One thing is clear though, Vega without power restrictions and with a bit more voltage will pull some serious power.
Around 1100W from the wall is what i would have guessed a more normally overclocked 2x Vega setup would pull, and ive done it with just one card
Admittedly that powerdraw includes an overclocked FX platform, not really known for power efficiency...
Well I don't have 2 Vegas, I got 1 Vega and a 750W PSU. I was thinking to just do whatever Buildzoid did and lift it up by 20%. Then, I will manage to hopefully maintain clockspeeds when gaming constantly, and the boost can go crazy without crashing the driver. I know the card will take 450W power, but a 750W PSU should be able to handle it. I know that the watercooling system will handle it, so that's not a concern.

I want to see if I can at least get it to be stable on benchmarks and games at 1750 to 1760 MHz. Anyway, could I see your card?
whiteskymage is offline  
post #15 of 20 (permalink) Old 03-11-2018, 04:36 PM
New to Overclock.net
 
Spacebug's Avatar
 
Join Date: Dec 2013
Posts: 66
Rep: 2 (Unique: 2)
I tried to take some pictures of the back of the pcb but can't get a good viewing angle.
The card is in a pretty cramped HTPC case with watercooling tubing running to/from gpu,cpu,vrm and chipset on mobo, with the most tubing clutter right at the back of the gpu pcb.
And I don't much feel like breaking the loop just to get some pictures...

If the whole setup sounds odd its probably cause it is
The rig started out as a HTPC for the livingroom/home theatre, and gradually turned into a gaming rig with the help of more powerful components, watercooling and an external rad.
Powerdraw isn't really an issue or priority when I got a projector and more than a few amplifiers drawing power/spewing heat.
An extra kW here or there won't make much difference except just more heat in the livingroom, which depending on the seasonal climate might or might not be a problem...


But really, the soldered resistors looks just about the same as it does in buildzoids video, except my soldering work looks messier and has multiple resistors soldered in parallell across the legs of the first resistor I soldered.
A cleaner job would be to solder just one resistor of appropriate value, but with my poor soldering skills I think I'll probably screw something up if I de-solder the first ones and solder back another one, easier to just pile on resistors in parallell soldered to the legs of the first resistors.

...And I have a big fat variable resistor on the back of the PCB as well that works as voltmod for the HBM.
...And a pile of extra caps soldered to the HBM VRM, which probably didn't help anything, but again, probably screw something up if I try to de-solder it.
...And a lot of hot glue fixating/supporting/covering various solder joints and the VR for the HBM volt mod.
Looks pretty ugly but I really could not care less since I can't see the card in the case...

So, yeah, I would much rather just refer to buildzoids video and carefully examine it, use that as a guide for the soldering as I did.



However, perhaps I have found a way to solve the negative clock scaling with increased voltage with powerplay modification.
I poked around some more with the softpowerplay editor and found that the field "usAVFSOffset" located in "Gfx Clock 7" or the Pstate 7 clock section perhaps can help, atleast it changed clocks during load slightly.

"usAVFSOffset" has default value 0
I tried various values ranging from 1 to 99 and got progressively higher maintained core clock during load/benches.
clocks didnt scale 1-1 for me but I think I got some 15-20MHz higher clocks when set to 99, with lower values giving lower increase of maintained load core clock.
Since I already had maxed stable clocks for the core voltage I supplied these clocks were not stable and eventually crashed...

Now, since my card already has hardmod for power and now "reads" just ~1/5 of its real current draw I have no use in manipulating with clock control as the card basically flatlines the clocks on load, varying a few MHz up or down.
But perhaps the "usAVFSOffset" could work to counter the clock drops due to the boost algorithms power reading variable.

I can't test it on my card since it basically has no powerlimit or accurate power reading but had I not hardmodded the card I would have tried to just mod the softpowerplay table with increased power and current limits to hopefully avoid powerbased throttling.
And then play around with values on P7 "usAVFSOffset" varying from 0 to 99 and see if the negative clock scaling (with increased Vcore) decreases, perhaps an appropriate value here could get the card to remain close to set P7 clocks during load?

Worth a try i think...
I tried to set values above 99 as well but that resulted for me in crashes when the driver loaded during startup, so i guess values above 99 is a no go.
Spacebug is offline  
post #16 of 20 (permalink) Old 04-22-2018, 02:01 PM - Thread Starter
New to Overclock.net
 
whiteskymage's Avatar
 
Join Date: Oct 2013
Posts: 104
Rep: 0
Quote: Originally Posted by Spacebug View Post
I tried to take some pictures of the back of the pcb but can't get a good viewing angle.
The card is in a pretty cramped HTPC case with watercooling tubing running to/from gpu,cpu,vrm and chipset on mobo, with the most tubing clutter right at the back of the gpu pcb.
And I don't much feel like breaking the loop just to get some pictures...

If the whole setup sounds odd its probably cause it is
The rig started out as a HTPC for the livingroom/home theatre, and gradually turned into a gaming rig with the help of more powerful components, watercooling and an external rad.
Powerdraw isn't really an issue or priority when I got a projector and more than a few amplifiers drawing power/spewing heat.
An extra kW here or there won't make much difference except just more heat in the livingroom, which depending on the seasonal climate might or might not be a problem...

But really, the soldered resistors looks just about the same as it does in buildzoids video, except my soldering work looks messier and has multiple resistors soldered in parallell across the legs of the first resistor I soldered.
A cleaner job would be to solder just one resistor of appropriate value, but with my poor soldering skills I think I'll probably screw something up if I de-solder the first ones and solder back another one, easier to just pile on resistors in parallell soldered to the legs of the first resistors.

...And I have a big fat variable resistor on the back of the PCB as well that works as voltmod for the HBM.
...And a pile of extra caps soldered to the HBM VRM, which probably didn't help anything, but again, probably screw something up if I try to de-solder it.
...And a lot of hot glue fixating/supporting/covering various solder joints and the VR for the HBM volt mod.
Looks pretty ugly but I really could not care less since I can't see the card in the case...

So, yeah, I would much rather just refer to buildzoids video and carefully examine it, use that as a guide for the soldering as I did.
You volt modded the HBM too? Does this allow you any additional overclocking on the HBM side? I manage to go 1100MHz. I am not sure if buildzoid said, increasing the power draw to 425W, if u do an HBM, can you allow to draw 30W for HBM ?
whiteskymage is offline  
post #17 of 20 (permalink) Old 04-24-2018, 12:39 PM
New to Overclock.net
 
Spacebug's Avatar
 
Join Date: Dec 2013
Posts: 66
Rep: 2 (Unique: 2)
Quote: Originally Posted by whiteskymage View Post
You volt modded the HBM too? Does this allow you any additional overclocking on the HBM side? I manage to go 1100MHz. I am not sure if buildzoid said, increasing the power draw to 425W, if u do an HBM, can you allow to draw 30W for HBM ?
Yes, on stock 1.35V HBM I believe I got around 1100Mhz stable without visible artifacting, much higher would give visible artifacting in various games or benchmarks.
With 1.42V I can run 1210Mhz without visible artifacting.

The voltmod is limited to max 1.44V as a safety precaution, and that isn't enough to allow much higher clocks, above 1210 I would get occasional artifacting and at 1230-1235Mhz the screen is covered in small blocky artifacting, "the usual Video memory craps out" kind of artifacting.
I'm toying with the idea to loosen up the limitations a bit to allow perhaps 1.5V and see if that helps to clock higher.
It can also be a timing thing where timings just are too tight to run higher clocks, don't know.
Or that the HBM or memory controller just don't like running higher clocks.
I will probably at some point test if even higher voltage helps clock some, but for now I'm running fine with 1210Mhz and 1.42V.

I don't know if the HBM modules or VRM rail is limited on power, I've not seen any indication of throtteling or downclocking on the HBM so I have no idea how much power it pulls overvolted or if it presents a problem.
Spacebug is offline  
post #18 of 20 (permalink) Old 04-27-2018, 01:59 PM
New to Overclock.net
 
Join Date: Dec 2005
Posts: 64
Rep: 2 (Unique: 2)
Quote: Originally Posted by Spacebug View Post
Yes, on stock 1.35V HBM I believe I got around 1100Mhz stable without visible artifacting, much higher would give visible artifacting in various games or benchmarks.
With 1.42V I can run 1210Mhz without visible artifacting.

The voltmod is limited to max 1.44V as a safety precaution, and that isn't enough to allow much higher clocks, above 1210 I would get occasional artifacting and at 1230-1235Mhz the screen is covered in small blocky artifacting, "the usual Video memory craps out" kind of artifacting.
I'm toying with the idea to loosen up the limitations a bit to allow perhaps 1.5V and see if that helps to clock higher.
It can also be a timing thing where timings just are too tight to run higher clocks, don't know.
Or that the HBM or memory controller just don't like running higher clocks.
I will probably at some point test if even higher voltage helps clock some, but for now I'm running fine with 1210Mhz and 1.42V.

I don't know if the HBM modules or VRM rail is limited on power, I've not seen any indication of throtteling or downclocking on the HBM so I have no idea how much power it pulls overvolted or if it presents a problem.

Im mining with Vegas. Its another community as you know. There people say that to increase HBM to 1200 and above youll need to set SOC 1200 using PT. No hardmod needed.
RatusNatus is offline  
post #19 of 20 (permalink) Old 04-27-2018, 05:07 PM
New to Overclock.net
 
Spacebug's Avatar
 
Join Date: Dec 2013
Posts: 66
Rep: 2 (Unique: 2)
Quote: Originally Posted by RatusNatus View Post
Im mining with Vegas. Its another community as you know. There people say that to increase HBM to 1200 and above youll need to set SOC 1200 using PT. No hardmod needed.
The SOC clock needs to be equal to or higher than hbm clock yes, on gaming drivers before october 2017 (i think) the soc clock was fixed at max 1107mhz if not edited in PPlay table.
However with gaming drivers after october 2017 soc clock dynamically change depending on hbm clock.
If hbm is set to over 1200mhz soc clock jumps to 1300 and so on.
So soc clock is a non issue atleast on modern gaming drivers.

But to get stability on my vega64 without visible artifacting in games or benches over around 1130mhz or so hbm clock it needs higher hbm voltage than 1.35v.
The increase in stable, non artifacting hbm clocks with over 1.35v ive found to almost follow the same trend as the change from 1.2V to 1.35 like the vega56 with 64 bios flash does.
So at least my card seems to scale with hbm voltage over 1.35v.
Spacebug is offline  
post #20 of 20 (permalink) Old 04-19-2019, 04:38 AM
New to Overclock.net
 
xtremefunky's Avatar
 
Join Date: May 2016
Posts: 21
Rep: 0
Hello Guys,

I saw this thread and i really like how You talk about to modify the Vega.

Can someone of You please explaine how to hardmod the Vega?
I didnt find any information about.
I want to mod the power reading and if its possible the voltage control.
Where can I get these info?

Thank You very much!
xtremefunky is offline  
Reply

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off