J7SC_Orion

...procrastinating on pre-Christmas home-office clean-up -outside it's cold and the roads a snowy-icy disaster...found something better to do:

...I got mine locally at the equivalent of US$ 1,619, but since then they were sold out all over Canada at various online/ local outlet chains. In fact, there is some minor / scant evidence that at least the initial batch of Gigabyte Gaming OCs could have started life as Aorus Master (due to the old RTX3K inventory at that time). I don't know about the latest batch (supposedly, there's a drop of new shipments of various 4090s hitting the stores in the US at least). ...I would make a quick list with 'most desired' on top of the list and 'definitely avoid' at the bottom of the list. However, 4090s tend to be tightly grouped re. OC and even if you can oc to the ying-yang (my card does over 3210 MHz), the gain in fps is minor - the 4090 stock performance is that good. One thing to remember - especially if you don't go for AIO or custom water-blocks - is that these things are HUGE and don't easily fit into many cases, even bigger cases...just check the available space in your case versus the measurements of your desired cards (per list) online. ...the card will come with a 3x or 4x-into-1 PCIe 8 pin dongle that on the card side has the 12VHPR connector, on the other end the PCIe-s. I ended up using a custom 'single' cable from CableMod, but the dongle I initially used also worked fine, unsightly as it is. Just make sure that there are no horizontal bends and that the connector is pushed in all the way - no GAPs !

...as @Sir Beregond mentioned, I have the Gigabyte-Gaming-OC (since mid-October), and so far, it is great...it came with a 4 year warranty. My card is water-cooled, so @ENTERPRISE's comment on fans won't be an issue. FYI, all told in my home-office / work and play, I have a Gigabyte mobo, and six GPUs of theirs and never had to do an RMA, so I don't really know how that would be like...my previous and only RMA experiences with Asus and MSI (one claim each) were excellent. ...the best PCB (IMO) for the 4090 is the MSI SuprimX / & AIO version...there's also the extremely expensive Asu Strix and the unobtanium Galax Hof, but if you want to go for Asus, I would get the TUF / OC and flash another vbios on it (ie Strix, Galax Hof)....all told, avoid cards that don't have 2x vbios and less than 20+4 VRM.

FYI, L1L2 prefetcher off is not really a 'trick' (as in cheat), but a setting ...but no matter what you call it, it works well on multi-core (in exchange for a small loss in single core, depending on the binning distribution of cores within your chiplets).

...true mGPU chiplets in a GCD (GCXs ?) is something I have been posting about at EHW for about three years or so now (hope springs eternal), the most recent one > here. The key will be the infinity fabric or whatever they're going to call it between GCXs. ...performance gains by future AMD and NVidia (and Intel ??) GPUs should continue to be breathtaking. I am in the software field and as such usually have a lot of equipment handy for work and play...the last few gens have made some pretty big leaps already but with die shrinkage and mGPU, it should be quite a feast. For comparison, two x 2080 Ti in NVLink managed just over 21,000 in Port Royal (and 12,000 in Superposition 8k). The single 6900XT is around 12,000 in Port Royal (7.200 in Superposition 8k) while the 3090 Strix hits about 16,000 in Port Royal (and close to 9,000 in Superposition 8k). Those were the previous years' goodies...now, my 4090 hits almost 30,000 in Port Royal and just under 17,000 in Superposition 8k...quite a change over the last few years, and no reason to expect that this trend will stop (subject to potential delays via the economy). Finally, driver technology (think alphabet soup of RTX, FSR, DSR, DLSS3, Frame Insertion, NV Reflex....) will also have to play a major role with the new hardware over the coming years.

Are you on Windows 11 (or possibly earlier version, depending on settings) ? ...not sure, but might be worth checking that vaunted TPM.

I am having a blast with Unigine's games within their benchmarks. Below is a game 'temptation' via some interesting levers in Superposition. Unigine's Valley is similar - one can float around the forests and mountains > very relaxing. While these are older titles w/o FSR, DLSS3, ray tracing and the like, it is still very entertaining and relaxing at the same time...

...not sure if this was mentioned yet, but you can also check AMD settings in the bios and look for Prefetcher...tuning L1 and L2 prefetch off usually gets you a decent boost with CinebenchR23 multi-score...some games also like it, others not. Just save your bios profile before making that and any other changes (which I'm sure you know anyway).

...all good advice, but AMD definitely did not want to miss out on the Holiday shopping season, IMO

...sadly, nobody will be home as we will be spending our Christmas up in the Whistler area... ...: ...in the old days, 'cold' winter weather would have been perfect for benching, but 4090s' VRAM likes it warm and cozy (25+ C ambient for the runs below). Doesn't matter much as the card already hit 3210 MHz on core when still air-cooled and on stock vbios...but new Galaxy still helps...managed to do several Superposition 8K runs at 3195 MHz (effective) and one at 3210 MHz.

"Nvidia made me do it"

...also, don't forget to bring skis (lot's of snow already) and bags and bags of spekulatius Christmas cookies

...on ECC, I know folks are subbing scores with artifacts (because they actually posted that info...). But to do manual culling retroactively is stupid, IMO - because what about someone who is just very good re. efficiency and prep ? Which one is which ? As I mentioned before, to make it fair, there should be a quick artifact checker incorporated into SystemInfo right before the actual bench starts... ...also: I don't need to buy a GPU LN2 pot because...I'm a pack-rat

Connecting somewhat to the original post and repair / labs segments and white PCBs, this fellow has some outstanding equipment and solder skills (he's currently also working on DerBauer's 4090 Strix)

...I know you have a Z790 Apex (per your earlier posts) which is why I 'hinted' - btw, apparently the 4090 Strix is now also available as a white edition. As to mobo upgrades on my side, I'm trying to hang on until CES '23 to see about AM5 X3D news, though the Z790 Apex / 13900 KS is on my shortlist in case AMD doesn't deliver a decent AM5 update.

fyi...that white Galax HoF PCB would go very well with a white Asus Z690 Apex (hint). Just make sure you're well versed in dongleing in white .

...does that case fit an Asus TUF 4090 ?

...well, those were different days. That said, the used 'Vette was less than half of what a new Honda Civic was selling for ...it would have cost the dealer way too much to bring back to regulations conformity (using their own service department prices). There were other methods one could take (and I did so) to make it (mostly) legal. Best of all, I drove for about eight years and sold it for a good chunk more than I originally paid for it. ...with the electric vehicle gen upon us, we might be able to use just a few of our computer modding tricks to 'make it our own'...

...yes, the 'outside' / air-cooler does look completely over-the-top and is the polar opposite of the 'inside'...but then, that super-clean PCB is what most people see when they buy this card and mount a LN2 pot on it . They might also offer a water-block for it (again). I am still wondering though (as many have over the past few Galax GPU generations) how Galax gets away with the custom things like 2x 12VHPWR given NVidia's super-tight design and bios control.

@Bastiaan_NL @bonami2 & Co ...I was cleaning out some boxes in my home-office and found these old print photographs (sorry for the pic quality...). It's from a trip from Vancouver to SanFran with my GF back in the day when I had finished rebuilding this used, abused & auto-crossed Z51/52 '95 Vette the dealer was selling for real cheap as it didn't make it through emissions (headers, no cats, dual 3 inch pipes, Flowmasters, modded port fuel injection etc ...). After rebuilding and modding it, I managed to sneak it through certification and then we took that trip down the Pacific Coast Highway...

...my fav 4090 PCB is the one of the Galax RTX 4090 Hof OCL below...2x 12VHPWR on top for ~ 1320 W. Anyway, my RTX 4090 w/Bykski block and 1x 12VHPWR CableMod cable has thermal putty at the PCB's power intake connecting to the block and runs nice and cool.

the not a 4090 is the first shot in the video, plus the second segment at the YouTube bottom bar

...that 'wild' RTX 4090 Galax vbios is s.th. else. While heat is not an issue, it does truly behave like a XOC vbios...among other things, it doesn't really downclock. As posted elsewhere at EHW.net, at 115% out of 121% PL, it was already pulling > 661 W in OCCT GPU test. I am 'too chicken' to see where the full 121% would land me in terms of watts... ...and some more minor visual updates on the Raven twins'...next stop: New mobo for one of them, perhaps early next year when I have actually figured out what to get (Z790 or AM5). If only there would be more 'trusted' leaks about the alleged 7950X3D at this stage...

When I watched it yesterday, I couldn't help but notice there was a certain feeling of sadness and/or resignation in the room (for obvious reasons if you think about it). When 'TIN' Tsemenko left in '19, it was already a big blow, IMO...but now it seems that they are all just drifting towards an uncertain future, waiting for the other shoe to drop...Kingpin looking at an eBike business could be one direction he might take, or he could indeed join another GPU vendor though that is considered unlikely. A quick sidenote on the 'not-a-4090' engineering samples. While the PCB around the power intake in cards with the regular placement of the 12VHPWR connector on top can be a heat issue, it is also relatively easy to solve. The EVGA not-a-4090 ES having the 12VHPWR connector on the side looks like an elegant solution at first, but since their PCB was thus inexplicable long (by far the longest of any 'next gen'/4090 I have seen) introduces other problems. For example, the power had to travel furthest distance to the core and VRAM. Also, the PCB would be very prone to flexing. Finally, with that length and the side-mounted power connector, this card would have been near impossible to fit in even a big case, especially with the 'do-not-bend' 4-into-1 dongles (and even regular 12VHPWR single cable strands).

Yup, it is kind of amazing how much it shrinks once you take the air cooler off a 4090, though it is appreciated per > this size comparison....I couldn't even have mounted it in the twins' case. Also, in case you haven't noticed, Pook is eyeing the Asus 4090 TUF