Overclock.net banner
Cancel
Create thread New Forums

A4000 shunt mod. Is it possible? Can we get 3070Ti performance in 3060Ti Aero footprint?

Jump to Latest
1 - 20 of 34 Posts

Runamok81

· Registered
Joined
·
194 Posts
Discussion starter · #1 · (Edited)
Many of us have modded the A4000 card with the Palit StormX or PNY 3060 coolers to make our own tiny "A4000 Aero" gpus.

The OC'd A4000 performs slightly better than the 3060 Ti, but draws much less power. It's power starved.
Shunt modding is the process of adding a resistor to the card in order to increase it's power draw. It was very popular with the A2000.
It's well known that the A4000 is power-limited 3070Ti card with a smaller PCB. Techspot does a great job breaking down the differences between them here.
Some relevant quotes ...
In short, the RTX A4000 uses the same GA104 silicon as the RTX 3070 Ti, meaning it has 6144 CUDA cores, 192 TAUs and 96 ROPs
Click to expand...
On that note, the A4000 was fed just 770 millivolts under load whereas the RTX 3070 Ti was fed 1040 millivolts, so that's a substantial difference in voltage there, and it's what enables the 3070 Ti to run at much higher frequencies, but it also blows out power consumption, as does the use of GDDR6X memory. Unfortunately, Nvidia doesn't allow you to tinker with the voltage of the A4000, or at least I haven't worked out how to yet.
Click to expand...
Previously, I didn't think shunt modding the the A4000 would do much because it only has a 6-pin PCIe plug, but youtuber Wiltshire Tutorials showed that the 8-pin traces on the A4000 are still active and you CAN solder on an 8-pin and it will work!
Does anyone have any ideas or suggestions on how to shunt mod the A4000? With the 8-pin power plug it may be possible to restore that 3070Ti performance and keep the 3060 Aero size.
I've seen a some whispers on reddit that it is possible, but I can't find any serious evidence that anyone has ever shunt modded the A4000.

Overclock crew, help me out. Do you think it's possible to shunt mod the A4000?
 
#2 · (Edited)
Absolutely possible. Be careful with the VRM, I probably wouldn't go over 900mV-950mV sustained VCore under load if I were doing this. At various points last year I was strongly considering getting an A4000 and modding it but ended up not- I have some thoughts though.

(teardown photos source- Reddit)
(markup by me)
Rectangle Font Art Wood Electric blue


Like I said, mind the VRM if you power mod- make sure Power Stages and inductors are well-cooled and keep an eye on the GPU Core-only power usage (NVDD). It looks like the actual NVDD VRM section is only 5 phases (note that WCCFTech cited it as 6-phase which appears to be incorrect) and assuming this is using 50A 55A Power Stages like the A2000, 3070 FE, 3070Ti FE, and others (AOZ5311NQI or similar), that will be a significant factor in limiting max sustainable clocks. Here's a snippet of the efficiency curves for a AOZ5311NQI with markups for NVDD output based on 5 phases and 1V VCore. Remember that Power Stage ratings are absolute maximums! The most comfortable zone for a VRM is around 1/3 of max capacity but running at 1/2 to 2/3 capacity can be okay if VRM cooling is sufficient- note the exponential curve in heat output as switched current rises and efficiency drops.

NVDD Output Wattage vs VRM efficiency- 5 Phases x 50A 55A
Rectangle Slope Line Font Parallel


The "good" news (for safety) is that stacking 8mΩ shunts on top of the existing 5mΩ should result in a new Board Power Limit of only around 230W which ought to be safe-ish for the VRM as long as it's well-cooled (since NVDD PL is lower than the Board Power) while still giving a nice boost.
Stacking 5mΩ should result in approx 280W PL which is either unsafe or catastrophic for that VRM depending what share of the Board Power goes to NVDD vs GDDR but could be useful if you want to "float" the V/F curve and manage throttling and PL yourself while still having somewhat of an upper limit.
2mΩ stacked would be unwise as that's like 500W- effectively "completely uncapped" territory for GA104, you would have to have a carefully managed V/F curve set at all times as that mod would put the card in a state where it could now very easily damage/destroy a power stage at stock with sudden load applied.

Anyways, hope all this rambling is informative. I think the A4000 is an interesting card and I still vaguely would like to mod one myself someday, maybe when they're available used for cheap in a couple years. Please do post updates if you decide to power mod yours and be nice to the VRM, it will be working quite hard!

edit to add- another thought comes to mind- what is the default V/F curve for that card like? It occurs to me that it may not even be able to call for higher voltage, can you check that? I'd assume it can, but the thing was clearly intended to run at extremely low VCore...
 
  • Rep+
Reactions: pf100, Awsan, speed_demon and 4 others
Save
Reply Quote Rep+ Rep+
Discussion starter · #3 · (Edited)
Not rambling at all. Very informative, and great guidance. I'm software engineer by trade, not electrical. So all if this is incredible useful for me.

I'll need to admit here that I am new to shunt modding. I need to take some time to understand the relationship between the power phases and efficiency curve. And - more importantly - how that guides piggyback shunt selection. Reading up, on it.

With regards to the three shunt options discussed, normal 3070 Ti power limit is 290W. So, best case would be the A4000 PCB and it's VRM handling that load. But that seems optimistic. Your suggestion of stacking 8mΩ shunts to raise power limit to ~230W seems like the safe least risky move. We'd just need to hope the VRM can cope? Is that the wildcard here? On a pessimistic note, this techspot article indicates that the VRM was only designed to handle 140W. But, not sure how they drew that conclusion.

You had a question about the A4000 default V/F curve? I presume you mean the stock V/F curve present in Afterburner? It's below. The curve calls for voltages way over the normal 700mV - 825mV operating range for the A4000.
Colorfulness Slope Rectangle Parallel Font

Above: The A4000 stock V/F curve starts at 700mV and 1100Mhz and ends at 1250mv and 1900Mhz

A question... In order to decrease the "measured" voltage and up the board power limit, which shunt resistors get a piggy back? Do we know? Or do we need to follow GamersNexus lead here, Do we just need to use a voltmeter to identify and modify the one(s) associated with the PCIe (12v-in 6+8) cables, right? No need to touch the one for NVDD or PCIe slot 12V, correct?

Presuming we identify the right shunts to modify, would a 8mΩ shunt like this one from mouser get it done?
 
#4 ·
Hey, glad to help, I love this stuff.

The shunt modding process for Ampere usually involves stacking all of the shunt resistors. Anecdotally I've heard success modding only the 12V input shunt(s) but I've also heard of cards having issues without all being done, something with the power balancing the card does between multiple 12V inputs. Also I'm pretty sure modding NVDD is important since that's to do with the GPU core which has a separate Power Limit from the Total Board Power.

Those Panasonic shunts look okay, pretty much the same as what I've used but I have learned the hard way that the kind like that with solder pads only on one side can be more difficult to solder well. What's your SMD soldering level, and will you be using iron or hot air? The kind with solder tabs fully exposed on both ends (like the stock resistors) may be easier to work with especially with an iron. The Panasonic ERJ are also very thin which is good for heatsink clearance but makes them more fragile, and I clumsily bent some while soldering on one of my cards.

That V/F Curve looks normal, the highest the vBIOS will actually call for on GA104 is 1100mV which is more than you'll need.

Regarding VRM capacity, I agree with the statement on Techspot with a note. It's not so much designed for only 140W in the sense that there's no safety margin above that, but rather in the sense that unlike gaming boards there's no built-in performance & stability headroom above the spec. There's always some headroom in VRM designs since they run more efficiently at lower loads- it's just a question of how much overhead, which partially depends on VRM cooling (bad cooling will eliminate safety overhead always, good cooling can't overcome absolute limits and exponential efficiency loss).

That said, meeting 3070Ti power levels with that VRM is unrealistic. I think 230W Board Limit would be ok as long as you set the overclock curve such that it's not actually sustaining that output for any length of time, and as long as the VRM is properly cooled. Going over 5W heat dissipation per Power Stage starts to make thermals a bit scary, going by the specs that point is around 175W output for the GPU core.

(Note: I did a mistake in my prior post- the common Power Stage for GA104 cards is 55A, not 50A. The markups on the datasheet graphs are correct tho as those were calculated directly from the chart data)

For the sake of Science I did a quick undervolt on my 3070Ti, here's what fully-enabled GA104 is like in TimeSpy at 875mV
It may not be fully stable (the clocks are stretchin') and VRAM power consumption is higher with OC GDDR6X so there's that, same core as A4000 tho.

Font Technology Screenshot Slope Number


and the V/F Curve for that in Afterburner
Rectangle Slope Font Parallel Pattern


I reckon aiming for 900mV on your A4000 should result in around 200W TBP and 175W on the core which ought to bring performance in line with RTX 3070, but at lower power consumption and with more memory! Just be careful with things like FurMark which will use insane power even at < 1V, but 230W Power Limit shouldn't immediately detonate the VRM if reached by accident.
 
  • Rep+
  • Helpful
Reactions: Runamok81, Paradigm Shifter and snootaiscool
Save
Reply Quote Rep+ Rep+
Discussion starter · #5 ·
Thanks again for the advice, analysis and info. Yeap, 8mΩ it is.

NattyKathy said:
The shunt modding process for Ampere usually involves stacking all of the shunt resistors. Anecdotally I've heard success modding only the 12V input shunt(s) but I've also heard of cards having issues without all being done, something with the power balancing the card does between multiple 12V inputs.
Click to expand...
It was a different card, but the GamersNexus video here said you shouldn't stack the shunt resistor measuring the PCIe slot voltage as you may end up pulling more than 75W and fry the slot. So, do we have to weigh "balancing the voltage" against frying the slot? The video mentioned that the slots could be fine pulling 100W. I think we might exceed that with stacking a shunt there? I'm going to keep studying some other shunt mods for more info here.

NattyKathy said:
What's your SMD soldering level, and will you be using iron or hot air? The kind with solder tabs fully exposed on both ends (like the stock resistors) may be easier to work with especially with an iron. The Panasonic ERJ are also very thin which is good for heatsink clearance but makes them more fragile, and I clumsily bent some while soldering on one of my cards.
Click to expand...
Novice on the soldering level. I'd recruit a friend for help, practice practice practice, or ship it out to have it done. Looks like these 8mΩ Eaton resistors would be more forgiving for soldering novice? Stacked shunt height could be an issue, but - so far - all the shunts look to be on the backside of card? For backplate clearance we have wiggle room, could run no backplate, or increase standoff or reduce backplate thickness because we've got dxf of a backplate and sendcutsend.com can make a varying thickness and material.

NattyKathy said:
I reckon aiming for 900mV on your A4000 should result in around 200W TBP and 175W on the core which ought to bring performance in line with RTX 3070, but at lower power consumption and with more memory!
Click to expand...
This is the goal! Not to mention shorter and quieter (notcua fan mod) than any other 3070 available. Just have to work up the nerves to put it under the soldering iron.
 
#6 ·
Have you settled yet on which shunts to replace with 8mΩ resistors? As a 3070 Ti owner & former Asus Phoenix owner, the idea of an 3070 Ti Mini that boasts double the VRAM & with No G6X, all within roughly the same form-factor as my old GTX 1050 is sick as hell.
 
Save
Reply Quote Rep+ Rep+
#7 ·
Hello folks, I had a customer reach out on this project. We pushed this to the front of the dev pile and ended up with the following:

Rectangle Font Parallel Screenshot Number


As you can see, we got some work out of this thing, she rips!

Here's some photos of the card as benched before the power mods. Average temperature of 48C in the bench log are with the fan set to 90% and these runs were set on a standard windows install on one of our office computers, same windows environment I imagine most of our customers would use. No special bench windows install here:

Circuit component Font Electric blue Auto part Fashion accessory

Circuit component Hardware programmer Passive circuit component Electronic component Font

Circuit component Rectangle Electronic component Computer hardware Font



Board power was measured using internal sensors.

TBP corrected for resistor measurement values on this bench run was:

Section 1: 206w peak with an average of 185w
Section 2: 192w peak with an average of 187w
Section 3: 199w peak with an average of 188w

Measured core voltage hasn't changed from stock, under load core voltage hovers between .750 and .756.

Next step will be to play with core voltage and see if that has any real impact on performance, although I don't know if this is really warranted since we have hit essentially stock 3070ti bench numbers here.

You think we should sell this alongside our a2000 mod?

Any other benches you all want to see?

-John
 
Discussion starter · #9 · (Edited)
@jkgventures - You've essentially got a recipe for the only ITX card that can hit 3070Ti speeds. The SFF community will flip out over this. Can I be customer number two? I will happily send my A4000 card your way! PM?

My A4000 has has got the same heatsink, but with a noctua fan mod. Build log has the details =

Ready to take it to the next level!
 
#10 ·
jkgventures said:
Hello folks, I had a customer reach out on this project. We pushed this to the front of the dev pile and ended up with the following:

View attachment 2600215

As you can see, we got some work out of this thing, she rips!

Here's some photos of the card as benched before the power mods. Average temperature of 48C in the bench log are with the fan set to 90% and these runs were set on a standard windows install on one of our office computers, same windows environment I imagine most of our customers would use. No special bench windows install here:

View attachment 2600216
View attachment 2600217
View attachment 2600218


Board power was measured using internal sensors.

TBP corrected for resistor measurement values on this bench run was:

Section 1: 206w peak with an average of 185w
Section 2: 192w peak with an average of 187w
Section 3: 199w peak with an average of 188w

Measured core voltage hasn't changed from stock, under load core voltage hovers between .750 and .756.

Next step will be to play with core voltage and see if that has any real impact on performance, although I don't know if this is really warranted since we have hit essentially stock 3070ti bench numbers here.

You think we should sell this alongside our a2000 mod?

Any other benches you all want to see?

-John
Click to expand...
Hi John, I'm the YouTuber from the video that was previously linked about the 8-pin PCI-E connector mod. I'm VERY interested in the shunt modding that you've done on the RTX A4000. I've been trying to figure out how to do this myself for quite some time now. If you're going to be accepting commission work for the mod, I would love to send you my card so I can do a video on your work! Hope to hear back from you soon! Thanks.
 
#11 ·
Hey guys, I'm glad to see there is some interest in this!

I'll be adding the various options for this mod to our website soon, but in the meantime feel free to get in touch either here or through our website www.jkgventures.tech

Once I get everything ironed out I'll post some more details here and make a new post here and on reddit.

Thanks!
 
#12 ·
Wow! Great work @jkgventures that's extremely clean. Excited to see a pro jump into this discussion, the A4000 are super neat cards that deserve some special modding attention!
 
  • Rep+
Reactions: snootaiscool and Wiltshire
Save
Reply Quote Rep+ Rep+
#14 ·
As @NattyKathy said, this card is driven by 5 AOZ5311NQ. I've only added shunts to the ones attached to the 8pin power rail. The pcie one is still left as is because I wanted to test performance of just power shunts. I'm using the aftermarket cooler you can get off amazon for the 3060ti but you may be able to get better cooling on the VRM with the older ones seen a couple of months ago on taobao. I'm running a pretty aggressive fan curve just to make sure things are cool and added a delta fan at that.




Automotive tire Motor vehicle Computer cooling Automotive wheel system Gas
Font Line Gas Art Metal
Computer keyboard Computer Motor vehicle Computer monitor Gas



Computer cooling Electric fan Computer hardware Gas Technology
 
#17 ·
Hi, today i moded another card. I make some tests. And haves some questions. All tests done with +260 core in afterburner. Added resistors is 2512 0.008R 8 mOhm R008 1%
1)6pin connector resistor add give me higher freq and showing power limit 120w in osd. But after 2 mins freq back to stock and stuck around 1750
2)6pin+NVDD resistor add give the same perfomance, except no boost to higher freq at start
3)All 3 resistors add give freq around 1900mhz and 110w in osd. Works stable and real power consuption is around 180w
Here the question. Why i dont have 230w real power after shunt add? and why osd and HWinfo shows only ~110w not 140w. in gpuz i see perflimit as power limit throttling?
 
#18 ·
Hi all - I joined up to post in here - I am a fellow "A4000 + Palit heatsink + 8-pin soldered" convert. I was previously running the MSI 3060 Ti Aero ITX OC but found the 8GB VRAM limiting for AI applications and also I would have to throttle the power limit down to avoid the fan getting too loud/it running too hot.

I just wanted to say fantastic work to all involved with the shunt mod! I had been wondering if it was possible after seeing it done on the A2000 with good results. Glad to see that it is. :)

I'm also looking for a bit of assistance. I have found the card I've bought, while it runs "fine", it appears to be downclocking itself. Even when running <50°C (122°F) I am getting core / mem of 1425 / 6500 - which seems low.

I have manually tweaked the card in afterburner to +200/500 core/mem and that gives me ~1500-1600 core (fluctuates) and a solid 7000 memory.

Does anyone know if it's possible the BIOS was modded on this (likely ex-mining) card? Does anyone have a guaranteed stock BIOS I could try flashing on?

Thanks for any help anyone can provide. This card is so close to perfect for everything I currently do and I'm glad to see others enjoying this little beast!
 
#19 ·
sanakess said:
Hi, today i moded another card. I make some tests. And haves some questions. All tests done with +260 core in afterburner. Added resistors is 2512 0.008R 8 mOhm R008 1%
1)6pin connector resistor add give me higher freq and showing power limit 120w in osd. But after 2 mins freq back to stock and stuck around 1750
2)6pin+NVDD resistor add give the same perfomance, except no boost to higher freq at start
3)All 3 resistors add give freq around 1900mhz and 110w in osd. Works stable and real power consuption is around 180w
Here the question. Why i dont have 230w real power after shunt add? and why osd and HWinfo shows only ~110w not 140w. in gpuz i see perflimit as power limit throttling?
Click to expand...
This should be your answer: [Official] NVIDIA RTX 3080 Owner's Club Volt-modding the NVVDD/MSVDD rail is needed to exceed ~180W, although given how paltry the VRM on the A4000, that sounds like a horrible idea personally. It's your call though.
 
Save
Reply Quote Rep+ Rep+
#20 ·
Well guys, we upped our score to 14123 which soundly matches a stock 3070ti.

we have officially listed this as a service we offer on our website now, extensive testing has shown these will work well in the long term with the SFF cooler and the custom brackets we developed to provide adequate clamping pressure for the VRM thermal pads.

A4000 Power Mod with SFF Cooler – JK&G Ventures

jkgventures.tech jkgventures.tech
 
1 - 20 of 34 Posts
About this Discussion
33 Replies
10 Participants
Last post:
sanakess
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling