jaybee83

btw bro @cylix we also need specs! i just realized ure actually showing the X version. thus far all the reports have only been about the non-X version. wanna know what the difference is?!

ok kool once ure done testing just send it over to me 😛

duuuude! talk about a comeback with a BANG! holy crap how much did u pay for it?! 😄 we want: bench scores, power consumption, overclocking abilities (if any?), temps, driver support (as abysmal as for the 5090?), any quirks, potential watercooling and/or shunting planned? we want ALL the deetz!

"inflation" 😛 /s uuuh id like for you to keep us updated on that. both the CUDIMM and 2 vs. 4 dimm board performance on AM5 interest me a lot 🙂

first thought was of bro @Mr. Fox getting one of those. imagine him skipping the 5090 but getting a 6000 pro instead 😄 dude, talk about getting raised with a rolex spoon 🤣 i am partial to peasant Fossil watches myself, really like their design language 😄 soooo....shunt mod coupled with @Prema Mod then? 😄 btw, i decided to do another vbios comparison, this time with shorter burst loads via benchmark tools. ill focus on the Suprim stock, PNY Epic-X and Gigabyte Master vbioses.

Nnnnnnnnnnnnnnnnnnoooooooooooooooooooooooooooooooo dont sell it! 😞 dont leave the NH55 gang! but if you MUST: good luck with the sale bud 😄

correction on my previous statement: no increased OC, silicon limit remains the same even with the new vBios. card is behaving a bit differently, i.e. stabilities / instabilities at different points / in different benches or games than before on stock vBios. but overall stability point is still the same at 3240-3247 Mhz.

spot on, the differences are miniscule. ofc when doing benches those differences should be larger. and yup, the second one was it. i figured if a vbios allowed for a bit moar powah / higher voltage i could also squeeze a bit more out of it, so best to compare oranges to oranges and set all vbios testing to the same clocks at idle 33-34C temps. naturally, one might argue that slight differences in the overall V/F curve shapes might account for slightly higher / lower clocks depending on the vBios. but since im basically only interested in the highest V/F points on the curve, the range "should" be minimized. granted, if one would be interested in undervolting and going further down the curve, those variabilities would likely be larger. but not interested in UV 😄 all in all, 5090 beggars cant be choosers 😛 being stuck with the 600W wall means well have to find other ways to squeeze a bit more outta these chips haha

Aight boiz, heres the jist of my vBios testing. Testing conditions: Alan Wake 2 Chapter "Summoning" Savegame 4K maxed out with all RT / PT maxed +HDR. Driver: 576.28 (XtremeG Mod) Duration: 60 min. Flashing / Setup Procedure: 1) Flash vBios via NVFlash 5.867.0 2) Uninstall Drivers via DDU 3) First Reboot 4) Blackscreen upon Boot, switch to iGPU 5) Only mouse cursor visible, blindly type in password to login 6) Desktop appears, but no icons - switch to dGPU, all good 7) Install Drivers, adjust settings 8 ) Second Reboot for Nvidia Overlay functionality 9) Set 100% fan speed via MSI Afterburner, check HWInfo for readout 10) Check V/F curve offset vs. stock Suprim vBios, adjust core offset to reach 3240 Mhz at 33-34C (D2D OC profile for my card at stock vBios) 11) Run AW2 as detailed above, record via HWInfo in background. After 60 min. minimize game while still letting it run and do quick screengrab of dGPU section in HWInfo. I attached the screengrabs to this post, together with the PNY Epic-X vBios ive ended up choosing, for anyone interested. How did I choose which vBios to pick on TPU? Welp since most creation dates of the uploaded vbios files are the same, i just went by the most recently uploaded for highest chance to get the latest version for each card. Manually checking if its indeed the performance profile for cards that sport two vBios versions. Quick conclusions: - Asus Astral LC/OC and Gigabyte Xtreme Waterforce were disqualified pretty quickly due to lack of proper fan control (Asus) and very low max rpm (Gigabyte), respectively - Stock Suprim SOC and Zotac Amp Extreme vBioses demonstrated the overall lowest average and peak wattage numbers, resulting in lower core clocks. So they dropped out of the race, as well - Gigabyte Master and Master ICE vBios versions are basically identical, only difference here being the impact of differing ambient temps (+2C on the ICE), thus some impact on the clocks. - In the end it was a neck-and-neck race between the Gigabyte Master and PNY Epic-X vBios. I ended up choosing PNY due to the following: 1) Highest avg. and peak wattage numbers 2) Highest core clocks for avg. and effective 3) Highest max. RPM for the fans (+200 above stock and the other fan functional vBioses) 4) Highest average voltage, even when accounting for differing ambient temps. There u go guys, enjoy! Any questions or comments, lettem rip 🙂 edit: as an afterthought, some additional things i could have included: higher measurement resolution in HWInfo to better get a grip on wattage peaks / shorter, burst-style loads in terms of standard benches & benchmark scores. but hey, only got so much time to go around i guess 😄 MSI RTX 5090 Suprim SOC vBios Testing.zip PNY Epic-X OC 10-Apr-2025 (275445).rom

well id say were spot on then 😄 ive dialed in new OC settings due to the changed V/F curve, seems like i can now do +20Mhz on the core offset, pretty nice. although not sure if this is vbios / driver / backplate cooling related or maybe a combination of all three. not 100% sure yet on the stability for D2D but should be pretty close. gonna do a session of Deliver Us Mars later to finish it up.

here you go: This is with the PNY Epic-X vBios flashed with stock clocks, no voltage added. At stock and 32C, the V/F curve for the PNY is 45 Mhz higher than that of the Suprim. This is just regular OS install that im running day to day, so no extensive tuning. Win11 24H2 Build 3476 RTX 5090 Suprim SOC (PNY Epic-X vBios stock).3dmark-result

now thats a pretty nifty tool 🙂

sure thing, can post later once im free from work.

aaaaah now that makes sense 😄 yeah some were speculating where the stock wattage should be to let the card properly stretch its legs and guesstimates were exactly in that 700-750W range. i have so much thermal headroom, even on air! those couple houndred extra watts would be no problem whatsoever... well well, 5090 still in your possession, shocking 😉

i know he was already testing a "KP-ish" card at PNY, but is he actually still with them? i thought he said hed be moving back home this year stateside to be with his family, so working for a taiwan-based manufacturer wouldnt be in the cards anymore / having sufficient opportunities in the US. still, could be that PNY took some of the KP input and implemented it into their vBios 🤔

aaand done! well i guess im going with the PNY Epic-X vBios from here on out, who knew 😄 in the end it was a neck-n-neck with the Gigabyte Master vBios. the PNY provides higher avg. voltage, higher max. rpm and also higher avg./max. wattages. all that results in the highest avg. / effective avg. core clocks. thx to @johnksss for bringing that card to my attention, otherwise i wouldnt have thought of looking at it 🙂 ill provide more details tomorrow, now its bedtime!

huh? hows ur "stock wattage" at 700W? curve editor doesnt help if 600 is the wall im hitting no matter the voltage or clocks that im running. if youre talking about miniscule spikes then sure, those would manifest in a few extra points in short burst benches, where that is naturally important. but over longer load periods that doesnt help much and just equalizes into basically nothing. but thats what im testing for, the longer loads. i was initially hoping that that extra leeway in powerspikes would help out in the overall average, but looks like thats not the case. ok back to stock vbios now, recording another baseline. then i can get an overview on everything and decide which vbios ill go with for D2D. in any case, one thing is for sure: we absolutely need higher power limits, thats our main brick wall here.

welp stock boost doesnt help me much if my max oc stays the same due to max attainable voltage and silicon lottery / stability 😄 sure leaving things at stock u might notice smth but not when all is maxed out. im mainly interested in having more leeway in terms of power limits, more flexiblity for spikes above 600W, since the 5090 is hopelessly power starved. but no luck thus far, they all behave the same pretty much. and yes ofc, im actually not surprised that the LC vbios versions behave so differently, not complaining about that 😄 just mentioned it cuz thats what stood out 🙂

damn so overall trust in sales is destroyed there huh?

hm dont like what i saw with the xtreme waterforce vbios. while the significantly lower rpm might free up a bit of wattage for the core and vram, temps shot up by 10C, in turn lowering clocks and voltages across the board. thats not really helpful... currently testing the PNY Epic-X vbios, all three fans functional and maxing out at 3150-3400 rpm, pretty nice. thats two fans at 3150 and one at 3400. iirc the epic-x sports two larger and one central smaller fan, so that makes sense. preliminary conclusions thus far: barely any difference between the vbios versions when it comes to avg. and max. wattages and resulting performance. basically all inside error / ambient temp. fluctuations. the only ones that stand out are the ones where the fans cannot be fully controlled, i.e. 2 out of three functional (Astral LC/OC) or very low max rpm (Xtreme Waterforce). after the PNY comes the Zotac Amp Extreme, then back to the stock Surpim to rerun the baseline with the updated drivers (576.28). so far not really sure what peeps are raving about with regards to Gigabyte vbios versions. granted, coming from a base model card, flashing the Master or Suprim vbios would give you a nice bump in stock performance due to a higher shifted V/F curve. but when overclocking to the same max. frequency and running into the power limit, differences are basically non-existent. btw can confirm that max. and avg. core voltages are unaffected by vbios, only dependant on type of load, temperature and available power budget.

done with the two gigabyte master vbioses, now on the Xtreme Waterforce. first thing i noticed: HWInfo only shows 1 fan vs. standard 3 but all three fans can be controlled as one and are spinning. however, max. rpm is only 1800 vs. 3000 at stock. lets see how that pans out under load. it "should" be fine since so far on automatic fan curve ive only seen up to 1600-1750 rpm max. under max. load. most interesting aspect will be power usage ofc. on to further testing!

sure thing, lemme grab two of the pics i initially took to illustrate, pretty straight forward actually: with the card installed in standard horizontal orientation, the power socket is on the underside of the PCB. at stock, there is a thermal pad placed on the pcb backside directly on top of the connector, with the pad connecting to the metal backplate (red circles). ive removed the pad and used two 30g TG putty pro batches to fill out the gaps over the core, vRAM and the power connector, all tracing a coherent line and connecting them (blue circles). then i just slapped Phobya Ultra 3mm thermal pad sheets onto the rest of the PCB backside where there was still enough space (indigo circles). sorry for the amateurish illustration, just a quick n dirty thing to provide some visuals 😄 noice, ill keep an eye out for that one then! finished Astral OC, same issue as with the LC namely that only 2 fans can be controlled. although in this case the max. speed is faster than stock for both at around 3200-3400 rpm. also just finished the Gigabyte Master, first success here with full control for all three fans 🙂 now off to the Master ICE. haha yep i recently had to chuckle when i realized that my desk was filled with Thermal Grizzly stuff 😄 its not that im a fanboy, i always compare a product type from all vendors, but TG usually ends up at the top of the wishlist due to a combo of pricing, performance and availability. Roman sure knows his stuff! and while he might be financially well-off, i dont envy him. hes basically a workaholic, working 7 days a week and has taken maybe 1 vacation in the past 5 years or so. hes also had a near miss with burnout couple of years ago, one of the reasons he dialed back a bit and left Caseking.

are u using a fan on the connector? ive also added putty on the connector backside and the cable is exposed to the 140mm fan im using to cool the RAM, so those two factor surely helped with the 4090 and will help with the 5090 now 🙂 btw, were back in business! just reflashed and Astral OC vbios and successfully installed drivers. Plan is to test the following: Asus Astral LC (already tested, gonna skip now due to limited fan support on Suprim) Asus Astral OC Gigabyte Master GIgabyte Master ICE (likely no diff to Master, but why not) Gigabyte Xtreme Waterforce PNY Epic-X OC Zotac Amp Extreme Infinity all the other vbios files currently available on TPU arent worth the hassle imo since theyre from reference boards / basic models. i didnt include any MSI vbios since i figured that the Suprim will likely have the best anyways across the MSI stack. will post results once done 🙂

speaking of vRAM temps: adding TG putty pro to the pcb beneath the backplate over the core and vRAM spots and filling out the rest of the PCB with large sheets of Phobya Ultra 3mm pads i got a reduction of 4C on the vRAM and about 1C on the core. not too shabby 🙂 im including checks for max fan rpms on the Suprim while testing different vBios versions. stock max is 3000 rpm and i must say: even at 100% the noise is not annoying at all, im impressed.

OMG! this is so stupid! LOL 😄 so i flashed back to the stock performance vBios, rebooted and tried to install the driver: again no dice! like....wtf? then i remembered something: sometimes nvidia drivers are behaving funky if theyre located too deep down from the root folder, basically if the path is too long they error out. so i copied the driver folder to C : and bam.....driver installs 😄 aight, back to the starting point then! at least this would be an easy workaround 😛