Thank you to Alza for the VEGA 64: www.alza.co.uk/ AHOC Patreon/Shirts/Paypal/Junkyard:cxzoid.blogspot.co.uk/p/suppor... The Twitch: / buildzoid The blog: cxzoid.blogspot.co.uk/ The Facebook: / actuallyhardcoreovercl...
At About 25 minutes you start asking: "Why undervolt?" Really? It is easy: The Chip has a fixed max temperature before throttling down. And this temperature is reached on stock voltage with lower frequencies than with lower voltage, so the Card can Clock itself higher. So you try to get the lowest voltage with the highest Clocks. If the Clocks go down, you have undervolted to much. THAT's the Goal with Vega.
Well, I actually have a question: That thing for the memory voltage in wattman (which isn't a memory voltage) - does it act the way you described only for VEGA cards? Is it the same for other Radeon cards, R9 family for example? From what I've watched (and it isn't much) on undervolting - They basically overclock at stock voltage and then drop down the voltage until the overclock they want is stable. But I think it depends on what performance are you willing to sacrifice for better cooling. And second question - can you kill the memory controller by undervolting in situation when you increase the clock and drop down the voltage simultaneously?
vdochev I'll test on my 480 tomorrow, check back in 18-20 hours for my results. It may work in the same way as my card never drops below 1.000V in any situation (according to software).
vdochev Ok, I checked and yes it works the same way on polaris. Just don't try and set it to 800mV. That instantly crashes. I don't know if it also affects the memory voltage, buildzoid would have to chime in on that one.
To undervolt you leave the frequencies alone and reduce the voltage until it crashes. Same performance, less power. Eg: I run my 480 at 1342Mhz at 1075mV. Side note: at least on 17.8.2 undervolting was completely borked, it worked until the gpu changed P-state and then everything goes back to stock.
Yeah but it also never runs anywhere near boost clock when it undervolts itself either does it? It just goes into throttle mode, undervolting allows it to maintain a higher clock at 1v than it otherwise would when it undervolts itself because it also brings down the clock when it undervolts, it's like overclocking the undervolt if I may say.
I think there's several possible goals for undervolting. One just seems to be that default voltage settings for some cards are just... Wasting power. You can lower power consumption and lose no performance at all. That's the main one I think. More broadly, in a theoretical sense, if you take overclocking to be getting the most performance possible out of a card, then the goal of undervolting and/or downclocking would be to get the most efficiency out of it. (eg, optimise for performance per watt, rather than raw performance, or lowest power draw.) Performance per watt optimisation seems like an interesting goal on some level, but it's probably pretty tricky to actually find the optimal point reliably...
Using a Vega 64 Liquid Cooled card I get some decent performance benefits from undervolting. Gave a quick test to validate it. "Out of the box" my setup was scoring 23,003 on Firestrike Normal, bumping the HBM to 1060 raised the score to 23,629. Keeping P6 at 1667 and lowering P7 from 1752 down to 1687, then reducing the voltage to 1100mv took 50mv off the card, lowered the power consumption (reported by GPU-Z) from 265 ish watts down to 220 ish and allowing the GPU to maintain a higher clock (up to 1630 from 1550) pushing the Firestrike score to 24,224. Noticed even better performance gains in GPGPU applications which push the GPU harder. No longer bumping in to the power limit as hard allowing for increased performance. Sadly I don't have a power meter so I can measure how much of a reduction is actually made.
thanks alot for the video dude, was very informative, i really love ur videos and im learning alot, too!! regarding the use of undervolting: u keep stock performance and lower power consumption. i, for my part, am undervolting my fury nitro because im fine with the stock performance of it and i rather have less heat to deal with than having like 5% performance uplift - lets face it. a game thats killing fury isnt going to run awesome on an oc fury. in the meantime im enjoying my silent gpu. for max scores tho the noise and power draw is as irrelevant as it gets :) just recently i put together a customers pc and tuned a rx580 red devil in the same fashion, lower core voltage from 1160 to 1050mV, upped the power limit and lifted that ridiculous temp target of 60°C up to 75°C. now the card is massively more quiet, eats up 20watts less power, has a 100 mhz OC on the memory with the same core clock. i could have clubbed that rx580 to around 1450 or 1500 mhz (depending on lottery), but that doesnt lift it up one performance class while you gotta deal with that ridiculous amount of noise. you dont care whether you play with 70 or 80 fps when 80 fps means you got a frickin hair dryer sitting on your desk. while noise isnt of any concern for a hardcore overclocker ppl using the pc as a "gaming device" imho benefit more from a low noise level operation than a few more fps for alot more noise. now in case of vega its really odd because apparently you get more than turbo mode performance with about stock settings power draw. i really wonder what vega can evolve to, since apparently its capable, but the stock edition is so far from polished its basically a clump of dirt. i wonder why amd was so sloppy on that one - id rather have put together some "mining" vega cards with the not so good chips than putting all chips into rx vega and vega fe production and let the early press benchmarks obliterate the card in power draw measurements because might one of 100 gpus needs that voltage. i'd rather have clocked the stock cards a bit lower and put more effort into finetuning them and make an outstanding reference card based on what the radial cooler design can actually cope with and get positive reviews of "the best stock card design ever". lets face it, the stock cooler is pretty good for what it is, the vrm is insane. the problem is the power draw of that card combined with that cooler and the power and thermal throttlings. amd should just have accepted to either sacrifice a small amount of chips or a bit of performance for a better/more positive appearance on day 1 benchmarks. everyone hates vega except those who love fiddling around with tech. everyone else just reads the day1 reviews and orders a gtx10X0, which probably is worst case scenario for amd and this time its 100% homemade. i guess they just "had" to launch vega someday and they decided to get over with it and someone came up with the idea to give it an untested "power bump" on the last second before shipping. im pretty sure the development vega samples weren't all over the place like the retail rx vega is right now.
i think the point of under volting is to maintain stock performance while comsuming less power. but in this case OC your HBM anyway, its free performance (close enough)
Hey, here in Germany the Asus Strix Vega 64 OC currently goes for 450 Euros (On Amazon.com it sells for f-kin 750$). If I install custom cooling pads for the VRMs and undervolt the card, how much of a power saving can I expect, if I still want to run on Asus' promised stock speeds? On a 50+ Power Target, the card can consume up to 450 Watts, which is definitely to much. Toasters need that much power, if you run them on a low level. However, some people on Reddit claim that their undervolted card runs on Asus' stock speeds with < 280 Watts power consumption. That would suddenly make the card very efficient. I really start to like the Vega 64, but I'd like to not rely to much on silicon lottery. And paying that much money, it's also not cool that one has to additionally tweak the card to gain it's actual potential. Team green's cards can't really be undervolted effectivly, so I think they made sure their cards ship with decent bios settings? I don't understand why AMD can't do that as well.
@@Steve25g Hi Steve, how can you get 925mv? I can set whatever less than 950mv but it keeps 950mv. My card is a Vega 56 Pulse. Also another question, how can i know if i have Samsung memories? Gpu-z doesn't show anything.
Glad you did this undervolting thing even though you want to push the VCore trough the roof as a hardcore overclocker. Especially with the whole memory voltage thing, since even GN didn't touch on that.
Hi, I have a question. How do I know on what BIOS my vega is? I left the switch in its default position and that would be switched to the left when looking at it from where you can read "RADEON". It didnt come with an instruction manual and I couldnt find this on the internet.
What do you mean you don't know where to stop?. If you can set a fix frecuency and start messing with the voltage, wouldn't you hit a point where the card is no longer stable?...
Also can you tell what implication it has when you do the Undervolting & Powertarget raise as you have done but then lower Temp targets down to 75°C max 70°C target ? And at the same time of course raise fan speed if temp limited.
I'm currently running the 64 BIOS on my 56, 1060mv and the default of 1050mv on the "memory" but now I'm going to start playing with that. Haven't seen much performance difference in benchmarks running 1000MHz on the HBM2 yet, but perhaps setting the floor will let it boost a bit more. Memory is very stable as far as I can tell too, and nothing got any warmer as far as I could tell, but perhaps it's throttling via timings? Not sure.
Can you do a similar video with the Fury lineup? I'm really interested in min-maxing my Fury X. I've gotten it down to ~175W by reducing core by only 50 MHz and messing with voltage. I'm having a harder time maxing out performance, specifically what to mess with and what are hard limits I should not touch.
I'm not sure how much extra max you can get a Fury (he had some). I wasn't getting anything great (maybe 1120 with +6 or +12 mV, which I know is basically nothing on the voltage, I just wanted to see if that small bump in voltage actually allowed anything more, and it does). I didn't want to kill this one since I use it for school, every day use and whatnot. My last card died at -55c from daily use with phase change. The solder joints cracked from thermal fatigue cycles, and honestly my backup is a 2007 Fire-GL card, and it doesn't exactly have the DisplayPorts to run my monitors. So I wasn't wanting to use that thing again (also sounds like a jet... unknown power draw, much better space heater than a Fury, a LOT more heat (at stock), and you need to wear hearing protection if you overclocked it). So you know why the Fury hasn't been overclocked properly. If I go to 1130 Mhz, you get massive artifacting. I've left it at stock (1050). I'm sure I could hit 1200 easy, but too soon since that last card, and I don't want to use an ATI flamethrower if it breaks.
very informative Video. glad someone found out what the freakin "memory" voltage is doing. another thing in whatman that seem broken to me, are the manual min states for GPU and mem clock.
Hi @Actually Hardcore Overclocking do you have any idea why RX 480 Memory clock is stuck at max speed even while in the desktop using 144hz refresh rate ? If I use 120hz refresh rate the memory clock goes down normally and not stuck at max speed.
Hey i know this is a year late, but I found 3DMark for some reason to not like my undervolt settings. But 3DMark superposition, and several 4k gaming benchmarks it runs well and sustains itself during gaming.
Hello. I see that you unmounted the backplate? Any reason for this? Is the card running cooler without it? My back plate is quite hot. You have a reference vega 56 or 64 (with turbine)? Thank you.
I know i know... its not your thing to do guides ;) but hey, asking is free... Can you explain a bit more specific how you can cool the VRM on vega with morpheus II cooler ? Is it not okay to just use the thermal pads with the small heatsinks? whats the disadvantage of this?` Further do you think if raijintek will put out an Update oif the cooler for RX Vega?
I undervolted P6 and P7 to what I perceived to be the stock P5 voltage (which I believe is around 900mV) and then worked my P7 clock up per the usual procedure while iterating the voltage in the 900-1100mV range. Found there wasn't much benefit to going above 1000mV in terms of clock speed, but I may be doing this all wrong.
TBH it's just named differently. On Radeon if you'd treat P6 as base clock and P7 as boost clock you'd see that it works basically like nvidia's boost. The "overclocking" term nvidia uses isn't correct IMO. Unless you actually opened afterburner (or whatever) and applied higher clocks, the cardss running at factory settings and therefore it's not an overclock IMO.
that isnt true though. if you open afterburner and set +50mhz on nvidia and it still runs slower and lower clocks, because it thermal or power throttles its not an overclock :P
Really? Because for me and driver 17.9.1 and udnerwater my gpu clocks have actually gone UP and on their own. Used to be like 1580 or 1590 or wtv and would go down quite a bit on the stock cooler. On my vega 64 underwater with 17.9.1 my idle clocks have gone from like 500mhz to 26mhz, my temps have gone from 84°C to 40°C on the core and 91°C to 44°C on the HBM, but my core clock now peaks at around 1640mhz without me touching a thing (in balanced mode too btw), and core voltage has changed fom 1250 to 1200 under custom if i check it out. So far though on water i am very happy with the card, it still a bit buggy (hbm will sometimes drop to 500mhz and cause some lag) but otherwise its been perfectly fine. I have not tried to undervolt and overlock yet but seems like all i might touch is the HBM since my core is already where it should be. And playing Siege (only game i have that peaks core clocks, even firestrike runs at lower clocks not sure why) it will average out around 1600mhz+ on the core and 945mhz on the HBM.
I got a Vega 64 Liquid with Samsung HBM2, its default is set at 950mV in wattman, i notice that the Air cards are set at 1050mV at default and even some liquids, i know their are two versions of HBM2 with SK hynix as well, also i get slight artifacting at 1075Mhz on the memory, no crashes though does increasing voltage on memory help on artifacting?
You are super funny and professional, can't believe you are 1 age less than me, you sound muuuuch older. Gj anyway, love your videos. Would you recommend going for custom heatsink vega 56 gpu? Some good one, such as rajinteks morpheus (,maybe with better looking) considering that custom manifactured gpu won't arrive soon and they will be overpriced as hell?
Setting p states to the same thing from what I can tell is an safety feature as manually setting pstates to the same on my nano bios would throttle to the next non similar pstates.
Pretty much what I ended up with undervolting my 56, same power for around 10% extra performance. But its weird in gaming cause when it downclocks during loading screens or cutscenes, it drops to states P3/P4 which are at 1050/1100 mV, so it just increases temps during low power workloads. Also setting the 'memory' voltage @ 800 mV just breaks everything, but 850 mV worked fine for around 1380 Mhz, 60 Mhz over balanced for 20 watts less.
I would be very curious to see the wall power if you put a gtx 1080 on the same testbench, How much difference there really is. Cause you know everything one hear is "Vega pulls 300-400W, its so bad". But if one tinker a bit with it, what is the real difference then.
I'd under volt to reach a specific goal, like a sound level or fan speed. I'd be interested to see what Vega does when limited to 100W to estimate what a laptop or console performance would be like.
Wish someone would actually make a good breakdown of what the hell is up with Vega oc/uv. My vega 64 seems to have a good uv but I just don't understand what the hell it's doing.
As for the point, it kind of depends on the person. Personally I make a noise level my target and try to get a stable system within that noise level while still trying to keep my system within temperatures I'm comfortable with. For someone with a (high quality) low wattage power supply they might aim for the most performance possible within a power budget. Some might just have a temperature goal. For instance, there *are* passive after market GPU coolers out there... You can just about get away with running a 480 at stock clock with one.
What do you think about the hardwareluxx undervolting result? They seem to be getting wayyy better result than everyone else and is being heralded by some pf the amd 'supporter' as the norm for how every rx vega card could perform.
Honestly with what they say they did in the article I'm not seeing how they could have such a massive drop in power draw. Like here I'm setting the card to 1V and it's pulling the same as stock. Then again they probably use the 220W BIOS soo that would make their undervolt results look better.
With the Fiji cards you can get significant drops in power draw, while retaining (or improving) performance by a combination of OC, undervolt and power choking them. I have found that in many scenarios they won't power throttle even with very strict power limits. I would agree with bzoid that they are probably power chocking the card, thus maximizing the performance per watt.
I would say a successful undervolt is getting at or over stock performance at less than stock power draw. So you have to pick how much performance you want/need and then you push down the power draw as far as you can. I've got a 1070 that overclocks pretty average, but I have it running at 1900 instead of the 2150 it can get pretty easily so I can push down the power draw and make the card cooler/quieter. My case isn't the greatest for airflow and my cpu (2x e5-2670) can't really push tons of frames for most games anyway, so I don't lose any real world performance with those gpu settings, but I get a much quieter and cooler pc (and needing to run the ac a little less).
You should graph performance numbers and power usage, then choose an undervolt that gives you the most pleasant experience. Limiting frame rate does help control thermals too.
Finding the maximum performance is nice, finding 90% of the stock performance at 50% of the stock power or more performance than stock performance at the same power might even be more interesting, depending on how much you want to spend and how much you enjoy noise. ;) I am curious how Buildzoid destroyed that other card, I missed that one.
Peter Jansen watch his previous video, basically he had a voltage mod that for some reason was causing delays for the vsense pin on the voltage controller which resulted in massive voltage spikes (over 2v!) for a few microseconds which in the end killed the card.
I'd love to see a power efficiency challenge - Use any desktop parts you like, but total system power draw (after accounting for power supply efficiency) has to be 300 watts or less. See what the best performance is you can get from that. The more challenging targets would be 150 watts + passive cooling only. And the extreme version would be best performance for power draw of under 100 watts, but you're not allowed to use any kind of cooling at all - not even a heatsink. (if you're wondering about the point of that last one, go have a look at what a 486 SX or earlier system looks like - I challenge you to find a heatsink or any cooling fan besides the power supply fan.) Basically, what is the most performance you can squeeze out of the least amount of power and heat production (without resorting to specific low-power parts such as laptop components, anyway.) I think this would be every bit as interesting as any overclocking challenge would be...
Thanks Tyler. The risk of modding, I guess. Kuralthys I would like him to aim for 50% power and 90% performance. I know a guy who did this for his Fury(X) card, it can be done if you are a bit lucky. That answers his question: "what is the goal of undervolting and how far should I go?". It all is about finding the sweet spot, the point which lies far above that what can be reached with the lower card (580) but at a much low wattage than what the card uses @ stock. It is about using the fact that the voltage has to rise much stronger (about squared) in function to the frequency. AMD tends to clock the GPU high to be competitive in DX11 with the Nvidia equivalent cards.
Hum, that could actually be a interesting overclocking/undervolting competition... Who can get the best performance out of a particular power budget. Any components and settings permitted to stay within the power budget. On low end hardware that would involve overclocking and increasing the power limit. On high end hardware you'd have to undervolt to stay within the limit. =)
The point of undervolting is to see how low you can get the power consumption while maintaining (Or slightly increasing) stock performance. For instance with my reference 7970 I was able to lower the voltage to 1030mV from 1156mV (while OCing ram to 1800MHz) got me 140w with a 10% boost of over stock performance! That places my early 2012 7970 at similar performance to a 780 Ti with 970 power usage!
I am awaiting on my new Vega 56 from Sapphire. My PSU is a Corsair Vengeance 650M, rated at 25A (54A in total) on each of the 12v rails. Can I use single cable with 8+8 pin connection or should I go with 2 separate 8 pin ones on different rails? Also how different your suggestion will be if I undervolt and overclock and/or flash Vega 64 Bios with relevant power limits?
Andreas Kavafis 64 bois on vega 56 is great. Just lower the voltage on core, increasing the memory bandwidth until it crashes/stable, nd up the fan curve. With my vega 56 card (64 bois) The core sits around 1550-80mhz (P6 1000 , P7 1100) Memory is at 1025 mhz , (1050v) 25% power + Uses around 250-280 watts Temps stay around 70-76, within full load.
Thanks for the Video, I under volt because I run my many Gpu's 24/7 at 100% load. I do what I can to save money on power. Sure if I had one GPU and played one game then to hell power I'd want my vega running at 1000 watts! For now I want to see my vega's run at 130watts with 1100 memory core clock of 1000CC ive only seen 189watts from my vega thus far.
well for undervolting ide say the reason is to try and get a preformance you want while minimising power used as on my r7 i OCed teh CPU and also under volted it a bit reducing power draw and keeping it very cool and the lower you can have the voltage more or less the better with no downsides
I guess undervolting for a gaming card would fall somewhere between stock performance @ lower power usage to higher performance @ stock power usage. You could go farther down in the summer for example, but at that point you're probably better off just using an fps limiter.
thing is... i'll be running furmark for like an hour and is stable as a rock (asrock?)... and then go and play whatever game and i get memory artifacts... driver restarts... hard freezes. The 3d mark benchmarks take so goddamn long to run. I wish i could just start up a benchmark... keep it running... and while it is running, incrementally vary the card parameters. I have a vega 56. I've managed to get it to... about 1550 MHz actual (as reported while running a benchmark). However. While playing a game, the gpu is not running full blast. It ramps up and down wildly. I suspect the clocks the user sets are just *suggestions* to the card. Corollary to that is that the card just goes as high as it can keeping the combination of *temperature* and *power consumption* under some limit. The kind of GPU load i suspect plays a huge, really huge, role to its performance *and stability* when overclocking. To cut a long story short, if you can keep the sensitive components of the gpu at a comfortable temperature and provide it with enough current, you can raise its power consumption ridiculously high (as evident from a vega 64 reaching 700W or something ridiculous like that... don't remember the channel, could be here, could be GN could be JZ)
to get mor stable undervolt try setting voltage like this / 1091 not 1090 ive found recent drivers if set to odd # it gets more stable and runs cooler for some reason on my vega 64 . mine the voltage spikes that cause crash is not there when set to odd numbers
The point of undervolting/overclocking is to run slightly cooler and less noise with slightly better performance (stable clock rate not jumping up and down). It is not to save power! P.S. I did not need to touch power target. But I will experiement some more with that.
My 580 and 480 behave the same way with regards to Wattman's "memory voltage" settings. I keep it at 875mV on both and input higher clocks and core voltages if I'm playing a game and then set it lower when underclocking the core for mining at night.
If I can add a comment here, from what I can tell, the temperature target is not acting the same as on nvidia cards. I have 70 target (not limit) and it stays at 65C in today's rather nippy London weather. I like colder room temps fortunately. But, digression aside, it just ramps up the fan faster. The temperature limit is the one that throttles the gpu clock. Should mention I'm doing this on a vega FE using a combination of afterburner and watttool because wattman is still shit. It'll be removed completely anyway with next drivers. Anyway, I love your videos, by far some of the most educational ones I've seen and please do not get morose because of the FE death. It's experience gained, this only further minimises your chances of buggering up other more expensive cards. Cheers! Edit: i get 1725Mhz@1150mv and hbm is 1100Mhz but stock voltage is 900, not 950 or higher., Never touched it but it's rock solid from my initial testing. Let me know if you want more info on it :)
I guess that the point of undervolting is to a) lower the power consumption while maintaining the same (or close to same) performance, or b) to get higher performance while maintaining the same (or close to same) power consumption.
Could you make a video about flashing to Liquid cooled BIOS and editing registry for higher power draw? Is 1750MHz possible (power consumption BE DAMNED!) ? I got a 750W Power supply, I am not really sure how far is "safe" to push Vega so that I don't f**k this PSU over. Surely it should handle around 600W of system power consumption well, above that, efficiency curve drops.
I think there is some caviat with the VCore. From my testing, there is no difference what so ever between setting it to 1000mV and 800mV (the lowest wattman allows). No stability issue nor lowered power comsumption. Dunno why
Jiří Tuháček I guess when the card detects it gets too certain voltage it gets unstable and override what you set, you should see the difference if you have power draw clamp or wall power monitor where the point the card decide to switch voltage back to 1.2v stock
It would be awesome if you could do this again now that the drivers + methods are better. With water cooling too. I have two of these on air and I can say one thing for sure: hashcat will push these to over 500W each and I will get free heat.
For me undervolting is to maintain stable temps and core frequencies aka boosting performance for the same or less power consumption. The standard settings are there for better binning but most cards dont need 1200mv and will even run worse and hotter unless you really lose the lottery and get a bad card.
undervolting for me is trying to get hardware running on stock frequencies with as low voltage as possible - until it crashes ;) sometimes even getting some overclock and undervolt together. my old i7 3770k did this pretty well.
I dont really get what the point is. The amd wattman controls voltage really good for you just set power limit up to max the set desired clocks. the voltage is dynamic and does a really good job at it. BTW 1300mhz core 1100HBM card is really power efficient and is still a good gaming card. Core clocks dont even do that much for some games because optimization is just that bad. HBM does seem to make a good bit of difference on most games.
I think the idea is to get the most performance with the least power consumption. to try to tune it to be the most efficient it can be. like tuning a carburetor. most HP best mileage.
@buildzoid About your theory about the so called "Memory Voltage" . Can you 100% be sure you are sensing the GPU potential difference and not the memory potential difference with the volt meter ? Just asking !
The point of undervolting is only present with AMD cards - it is to get stock performance or better at lower power consumption. It's a simple form of min-maxing for daily use. Less power consumption for basically the same performance. When it is available, why not do it.
I highly doubt it's an AMD only thing. The whole cause and point of undervolting is poor binning, and conservative voltages. When you have 100 chips, and 10 of them need, say, 1.3 volts to run stable at a given clock speed, 10 need only 1.1, and 80 need 1.2, what voltage do you specify in the bios? Well, if you say 1.1, 90% of cards will fail to run reliably. 1.2, 90% will run reliably, but you have to either not sell 10% of the worst chips, or live with cards being considered unreliable. If you set the default to 1.3 however, they'll all work, but many of them would be able to use less power by undervolting. This same result would be expected with any situation where the binning of parts that are considered 'the same' varies widely, and you have no way to auto-detect ideal voltages for each individual chip. In principle it would happen with AMD, and Nvidia GPU's and even processors. Why it's treated as an AMD specific thing would seem to be because AMD is especially conservative with it's defaults and sets them way higher than most cards actually need them to be.
KuraIthys I agree, it's not that nVIdia wouldn't benefit, but due to AMD's very conservative high voltages and power states, one can see why their cards would benefit the most.
I set all memory setting to minimum AMD allows, ran it for a while, decide to turn memory voltage back to auto, crashed within minutes. Vega FE water flashed MI25, latest pro drivers.
matej nemec my HIS 56 has the EKWB Vega block on it on a custom loop, loop never goes over 45Degrees Cel. Hence the Vega64 WC bios, which I thought gave a higher stock OC, with out touching anything in Wattman.
I went from 85°C to 40°C on the core, my core clocks sit higher even running stock everything. I used to peak and stay stable at around 1450-1500 on siege now its more like 1600-1630, and it makes the most difference for HBM timings.
bluej511 im intrigued on buying a water block for my vega 56 (64 bois) card lol but jeez...theres so many things to buy, from water block for the card, to all the other assessities..price just climbs up lol
Probably a really basic question, but you've got your card PCI-E power cords plugged in exactly how Seasonic specifically tells you not to...So, why does Seasonic say that...and why is it seemingly ok(?) to ignore it?
So technically 1 of these daisy chain cables should be able to handle 360W. Without any mods to lift VEGA's power limit I'm not gonna exceed that and plus it's on an open air test bench soo it's like the cable is starved for cooling. If I'm working with GPUs pulling more than 360W continous I go and use 2 separate 8pin cables per GPU but even then I've gotten away with using daisy chained cables. Now if you were in an environment with low airflow under long term heavy load the cables can end up heating up to unsafe levels in this configuration.
Ok so I did a little bit more research thanks to your reply. Best info I could find is that my Seasonic PS (850W Focus Plus Gold) has 18 AWG PCI-E cables (not 100% on this, could only find specs for 750W which are 18 AWG, but assuming the smaller gauge to be safe).....and they should have an ampacity of 10 amps? For 12V that gives a carrying capacity of about 120W per strand? 6+2 PCI-E cables have 4 pairs for a total carry capacity of 480W? So as long as my GPU isn't pulling more than that I should be safe to daisychain? Also, I'm one of those "undervolt people" :p
To add to what Buildzoid said, the heat is because of resistance and higher resistance means less stable voltage, which does actually affect your OC. Jay had this issue in a video he did a couple of months back, but I can't remember which one it was. As far as power draw is concerned, 6-pin cables have 2 anodes with a 3rd optional one that's usually disabled and 3 cathodes, and 8-pin cables (or 6+2-pin, same thing) have 3 anodes (the disabled one enabled) and 5 cathodes (2 are sense lines for sensing that the connector has been plugged in), so at 10A that's 360W for the cable. With a +50% offset, you're looking at up to 360W pulled through it on a V64 so if you're undervolting it'll probably be fine, but I wouldn't do it otherwise, and if you want to maximise your OC then you'll want to go for 2 cables to lower that resistance.
Actually what Jay found is that Nvidia's power management stuff is a little overly sensitive. I tested with my VEGA on +200% power limit and it doesn't make difference. Though just 1 cable gets pretty hot pretty quick.
I think the term 'undervolting' has been co-opted to mean tuning Vega for efficiency. As illogical as it sounds, you managed to point out the stock BIOS suck and simple tuning can improve performance quite a lot and because a voltage value is reduced to achieve this it's gained the 'undervolting' nickname. Thanks for highlighting the pitfalls of tuning Vega though.
undervolt helps less heat which helps for less noise on the stock shitty blower. 👍. also gets same or better performance than stock. maybe less heat will help hbm be stable at higher clock as both core and hbm share same cooler. i have had had the card fail stress test if you go too far down with voltage. basically decide the same way as with overclocking. lower volts, run test, see if performance got worse or it crashed... if you don't have better cooling then you can gain maybe 10% performance easy with vega 64 in my experience without needing to crank fan speeds up... which if you care about noise is a good thing 👍
For those that are here to understand something about the undervolting, bicause of the current state of the market : Undervolting is to find the maximum performance/power level at the desired level of fan noise. If someone is looking to optimize a Vega card that is under water, they are to set a maximum power draw target and make the most of out of it.
The goal of undervolting is to achieve stock performance with lower power consumption. So if you want to achieve the best undervolt you don't go for the lowest you can get because that would compromise performance. Simple as that. Of course everybody can decide what performance is acceptable for them, so basically every time you don't go for the highest possible overclock because you want to limit power consumption could be considered some kind of an undervolting.
