I have a Vega 56 getting delivered today. I am doing my prep work, and getting everything I need to overclock this sucker. If my card comes in with Samsung memory I already know which bios I am going to flash to, but if it doesn't I am not sure which Vega 56 bios I should use, I am currently torn between two options. Stock Bios Socket Power: 180W \ TDC Power: 300A Option 1 - Socket Power: 260W \ TDC Power: 300A Option 2 - Socket Power: 222W \ TDC Power: 403A Which would be more important when getting a good stable overclock. Socket Power or TDC Power?
We'll need more information before we can make any recommendations on whether flashing is a good idea for your particular card. However, based on the given information, that table seems quite strange because over 400 amps load current is going to be a bit much. Usually, you're hardware limited to never use more than 300 amps on an RX Vega. 300 amps * 1.2 volts = 360 watts
Thank you for the insight. I just got my new Sapphire Vega 56 Pulse installed and it has Samsung HBM, so I will be able to go with the Vega 64 bios! I am still making sure the card is stable right now before I go down that road. I keep hearing left and right bios switch, which was should you flash with the card installed in the PC? I ran a FireStrike on my heavily OCed 970 last night and got the scores 3DMark : 10918 and Graphics 13546. With this new Vega 56 bone stock I got 3DMark : 15928 and Graphics : 22076
@JonasBeckman may have the necessary information to guide you with that particular model. Usually, as long as the card comes with two separate bios, it should be easy enough to flash back if you have an issue. On reference model cards, the writable bios is with the switch towards the bracket, and the second, write protected bios is towards the fan. However, as that is a non reference model, you'll need to be very careful when deciding if and what bios to flash. Also note that although the bios download pages on TPU do usually list some parameters, non reference models can have practically any part not match to a bios from another model. There's also some unverified information here, where people have successfully flashed with both a XFX nano 64 and a Sapphire Nitro+ 64 bios https://www.reddit.com/r/Amd/comments/96vkn4/sapphire_vega_56_pulse_with_64_bios/ I suggest doing as much research as possible before you consider flashing a non reference RX Vega. Also, your Fire Strike scores look perfectly fine.
Well so far I am not stable stock, AMD Wattman keeps crashing at random, even when just web-browsing. So I need to figure out what is going on :/ Hopefully I didn't get a dud card...
Well I think I found the problem, with this new card my CPU OC seems to have become unstable, I went from 4.6 back to 4.5 and all crashing seems to have stopped.
Vega 56 Bone Stock/Balanced - 3DMark : 15928 and Graphics : 22076 Vega 56 w\ Vega 64 Bios (Linked Below) minus 100hz on the CPU for stability Balanced - 3DMark : 16605 and Graphics : 23668 Now to start tweaking... https://www.techpowerup.com/vgabios/198055/gigabyte-rxvega64-8176-171026
Vega 56 w\ 64 Bios 1777 P7 @ 1120, 1000 on mem, real in game core speed of 1710-1720, 3DMark : 17340 and Graphics : 25663, 65C max core temp. I am going to run at this for awhile and see how stable it is, it was stable through Fire Strike, GPU Core Burner Furry Donut 20 mins, and Superposition. Once I verify stability I will start to push memory from 1000 to hopefully 1100 and then call it good. The fan is a little loud, I could run it a little warmer, but if I play with the fan curve it seems to kill my zero fan speed at idle, something I will have to play with.
I have got it near final, at about 1692 max in game/benchmark, 1025 on memory (not 100% stable at 1050), but I am having some weird issues in some games. The boost just seems to stay low in some games some times. I watch my sensors with HWiNFO, and I will be at %60 GPU usage and around 1200 core, 90 watts power, 50c, and I just wont be capping my frame-rate where I want it. Its incredibly frustrating...
If the same thing happens at default settings, it's the game or CPU. Otherwise, try making slight changes to your p6/p7 voltages.
Well a bit perhaps and the one thing with the Pulse here is that it's using the Nano PCB style for Vega which primarily means that it has less phases from a 12+1 setup to 7+1 thus the recommendation to flash using other Nano PCB GPU's like the Red Dragon. As long as it is Samsung memory of course since with Hynix things are a bit different and there's no guarantee that the flash will work correctly since you aren't just writing the altered voltage into the bios doing this. (And overclock results aren't going to be super great though it can still hit 900 - 950 Mhz but testing is important to be sure there's no artifacts or stability issues and then thermals with the HBM preferring a 72 degrees temperature or lower before it hits the first throttle step and drops timings.) EDIT: And yeah 18.12.2 and on with Adrenaline 2019 and AMD still hasn't quite fixed the OverDrive 7 API and changes (VII on OverDrive 8 is doing fine now though.) so you get this zero fan nonsense problem and the curve being a stepping instead of a gradual change which can be a problem with volume and noise and also performance if you use a lower setting until AMD (Finally!) fixes this little problem whenever that will happen. (Zero fan is improved a bit but not perfect so it's changed a little bit at least but needs more work still.) EDIT: Ah and I see the GPU is working using the Windforce bios, well as long as it's stable and doesn't exhibit any unusual problems or other behaviors it's all good. (People have flashed these using the stock Vega 64 bios too so it's not like it's a requirement to use a similar card design for the bios flash.)
You will be fine, Use HWinfo to monitor the VRm temps. They can get really hot really fast, you can have the core at 55c and vrm at 100c so be carefull. If you have a waterblock I dont think it will be a problem, also you have a water block for the nano from this guys: https://www.aliexpress.com/item/Byk...LSE-Radeon-RX-Vega56-8G-HBM2/32908348198.html If you what to upgrade to it. looks nice
Yeah there's that too, Nano PCB and everything is closer together. Water cooling really helps these GPU's and it probably works really well for this configuration what with all the components being nearer one another and what with the heat these VRM's can put out even if they themselves have a pretty high temperature tolerance. (Near 150 degrees Celsius I believe was what the ones on the AMD Fury could take though around 100 is closer to their regular operating temperature.) Sapphire's Cooler isn't bad even if it's not the full Nitro+ treatment (Those sure disappeared fast.) but it's not going to compete with a good water kit. I am also not 100% on what components AMD have on the PCB and what Sapphire might have changed on their custom Pulse design of it if anything, voltage caps on the stock model from memory has a thermal limit at 95 degrees Celsius so that's a bit lower than on the full size Vega from how I remember a video breakdown going through the stock Vega Nano model though there also footage from the Pulse and Red Dragon PCB. This would be the one for Vega 56 Pulse, unfortunately there never was a Vega 64 variant for this smaller PCB design although the Samsung memory models and a overclock can get pretty close to it if the GPU doesn't hit any throttling for thermals, voltage and such. https://www.reddit.com/r/Amd/comments/89uzar/vega_nano_pcbvrm_analysis/ https://i.redd.it/ez3jnmsqazp01.jpg Did quite a bit of reading on Vega and the later Nano PCB last year but it's quickly forgotten, learning the software and the drivers has helped a bit too but AMD is making some changes now with their later drivers from mid last year I think when the auto parameter for Wattman was improved on how much voltage it drew from the GPU and what it needed to now with Adrenaline 2019 and larger changes to most of the settings here including even unlocking a few extra settings like the rest of the performance states. EDIT: But for the question here in this topic I suppose as long as testing is stable and nothing unexpected occurs then the flash was successful and everything is working fine allowing a increase from 1.25v to 1.35v on memory and probably pushing the HBM2 clocks closer to 1100 Mhz or thereabouts depending on where it starts showing either memory errors, graphical glitches and artifacts or crashes if it's right on the edge of what it can take. Should make for a nice performance bump, one of the bigger performance blockers for Vega thus the higher than average gains from overclocking it higher if the memory can take it. Older data I had for that roughly saw a 2 to 4 percent gain from core clock speed increases and then 6 - 8 percent for memory so it makes a difference though this was from testing when Vega 56 and Vega 64 came out so it's probably changed a bit in how newer drivers handle overclocking. (And this would be from game tests, benchmarks can be more sensitive pushing the core and memory to it's limits and also good for stability testing.) EDIT: Though I am still an amateur at this, so much to learn and here's Vega20 and the Mi50's AMD made into the VII's and what these change, will be interesting to see what they turn Adrenaline 2019 into in newer drivers too as they fix up these flaws a bit. Hopefully.
I think it is the game engine and maybe the CPU, it only happens in very specific locations and is repeatable, happens both at balanced and with the OC. I re-OCed my I7-4790K up to 4.7ghz stable and it seemed to bump it up just a little. I plan on replacing that old I7 once the new Ryzen procs come out. After a bit of testing I ended up switching to the Sapphire Nitro Vega 64 bios. The Windforce bios seemed to work well, but I lost Zero Fan completely as an option in Wattman. The Nitro bios has less sensor monitoring available though, as I can't see any VRM temps. I tried the XFX Double 64 bios (64 nano pcb), that would boot into windows, but it wouldn't register as anything in GPU-Z and the drivers would not see it at all. Thanks, I will keep that in mind, I am getting a expandable Swiftech H360 X3 unit in today. I may pick that block up when I replace my mobo\proc later this year. I am pretty impressed with the Nano cooler so far, I can't see my VRM temps with this Sapphire Nitro bios, but they did look good in everything under the Gigabyte bios except the Kombuster Fuzzy Torus stress test, the other Fuzzy tests were just fine. The temps on the core never exceed 75c at around 1800 rpm and the HBM temps are only ever 2-3 c above the core temps, hotspot temps max at around 95c. So far I have only got my HBM up to 1025, 1050 proved to be unstable. Is there a specific setting or voltage to help stabilize the memory at 1100?
For memory performance unfortunately it's going to vary for every card from what I know, there's no real setting for HBM voltage either and it's locked to a singular value with Wattman having what is often considered the voltage floor and possibly shared p3 state for what the voltage can drop too at the most so it doesn't affect the actual memory voltage at all which is 1.25v for Vega 56 by default and 1.35v for Vega 64 and that's it. (EDIT: Well the voltage for the full load state that is. There's a mid state and a idle state too for memory.) For the speed well 1000 to 1050 is still good and a nice bump from the default 800 Mhz value but binning and luck of the draw really for how high it can go, thermals factor in but I only know of the 72 degree Celsius throttle point where it starts loosening up the memory timings and then I think it's 95 degrees where it shuts down to avoid any potential hardware damage. Hynix seem to hit around 900 to maybe 950 and Samsung well 1 Ghz up to 1100 with a few really good chips hitting 1150 or around that though 1020 to 1050 comes up frequently too when searching for other users and their results so it's just as varied as the GPU chip itself from what I can tell. (1700 at default voltage to 1700 with undervolt to not even hitting much above 1600 without having to add additional voltage if possible.)
If timings at 72c+ are changed I would love to change the fan to keep it under 70c, I appear to have the headroom with the cooler to do this, but if I change the fan curve at all in Wattman I lose Zero Fan completely. I am curious if I can change the target temp or change the fan values in powerplay tables to adjust for that.
