J7SC_Orion

...

...and their main GPU factory is just a block from the Yellow Brick Road, apparently.

I found out the hard way that ~5200 W total combined for everything is the limit for our place; 4x sub-ambient Classies w/ EVBot + Intel HEDT on LN2 pot (all on 4x PSUs) was the first time I triggered the master fuse switch since we lived here...however, that does not include the 240V dryer circuit in the laundry room I have been eyeing as of late . What is more, an older system (TR, 2x 20280 Ti) in another room pulls well over 1,100 W when on full song - 780W of which is just for the NVLink 2080 Ti GPUs combined. From that perspective, the RTX 4090 and its 660W+ is an 'Energy Saver' @iamjanco ...since 4090s don't do NVL SLI (at least not officially), you may want to trade your 2x Asus 4090 for one of these...rumour has it up to 1320W w/the XOC vbios (note the 2x 12VHPWR connectors)...

I recently opened a thread re. > a glimpse at future desktop GPUs by briefly summarizing the latest enterprise-level models by the major GPU producers. Some of those enterprise models had a factory power rating of 560 W. I just read a funny (...if it wasn't happening to you) post at another forum about a chap with a RTX 4090 doing some OCCT GPU stress testing and tripping the master circuit breaker, with an ensuing 'comment' or three by his wife about the whole house going dark... In any case, below you find my 4090 doing s.th. similar with OCCT at 661.5 W in GPUz - the lights stayed on -, and that wasn't even on full power budget (115% out of 121%) for that vbios. Even in regular gaming per the other parts of the pic, you can easily get to over 500W...then again, the 4090s are super-efficient and at s.th. like 380 W, you will get way more complex graphics fps than with the same 380 W used for prior gens by AMD and NVidia. Still, with environmental and climate issues rightfully getting a lot of airtime (pun intended), what is your take on reasonable max power consumption for a single graphics card for home use ? Moar (like some Galax models with 2x 12VHPWR and > 1200W) ? Less ?

While I suspected per earlier post that it could be directly from EVGA, I don't think they're doing it for the money. Further, yielding funds for a charitable purpose is a good thing (never mind the publicity it generates), but it won't give you the true price for a 'collector's model'. Also, a couple of the 20 or so engineering samples made it to reviewers such as GN - the PCB looked quite odd...by far the longest of any 4090, with the power intake having to travel a long distance to the main bits across an empty plain. FYI, for around US$ 5,000, you can buy the 4090 Ti's studious cousin already a.k.a. the RTX 6000 Ada

@Bastiaan_NL, @Avacado & Co... new rules EEC info anyone ? I finished updating the build and ran a few benchies at 3DM ... ...apart from my Port Royal at 30,337 (not shown or subbed as it was bugged), I also didn't sub some other results, including those which would have gotten me into the top 5 HoF / single card for any hardware combo (i.e. Graphics Score).... While I am all for extra checks and balances applied to everyone, I already stated that the motivation behind the loudest voices seem to be their own HWBot points-spread interest...further, there are artefacts checkers out there (as shown before) they could have incorporated into SystemInfo for everyone and all hardware. Instead, the ECC requirement at HWBot just hits latest gen of NVidia cards but not others. This introduces a skew, for example in favour of the upcoming AMD 7900XTX. While I expect the 4090 to continue to dominate, I just don't like uneven weights applied in what is supposed to be same rules / level playing field for everyone comp. In any ace, I haven't subbed HWBot for years, so it is more of a philosophical thing. Here is the impact of EEC (note the 23 GB instead of 24 GB tags)...obviously, it is slower by design as it also involves extra cycles. That said, as I expected, NVidia released a new driver that helps memory speed for regular and EEC (though nowhere near enough for the iron-ball-and-chain handicap). Also, a special GALAX HoF vbios appeared 'just yesterday' - unusual in that it was written specifically for the 1x 12VHPWR cards, instead of the 2x 12VHPWR feeds the Galax 4090 HoF OCL uses ...the rest of the 1x 12VHPWR is pure XOC, and it runs up to 667 W (some folks have seen 680+ W), basically half of what the 2x 12VHPWR is capped at...I haven't loaded it yet but look forward to it, along with the new faster driver

...interestingly enough, EVGA is also hq'ed in Brea, California (see eBay location tag), so it 'could be' them (or not). From what I recall, there were ~ 20 or so engineering samples produced, complete w/ new 'Nex Gen...' stickers.

...my reasonably priced TT core P8 case after massive dremeling and with additional loop updates for the work + play build. Trying to use some 5K original pics for a smallish dimension via ancient graphics software - I'll get the hang of it, eventually. 3x tempered glass sides and top are obviously not mounted to avoid reflections...

...sorry to hear about the 'little accident'. I used to drop LG 2011 HEDT CPUs onto the socket and bent pins with great regularity...got quite good at bending them back (I put some tips into a thread at EHW can't locate right now, like using a carpet knife as a back-stop when bending pins back with a needle). Anyway, the plastic piece you mentioned might be a game changer re. insulation issues. ...might be worth checking with EVGA re. socket fixing cost.

...

I have very fond memories of Old Montreal during reading breaks at university...there was than one issue though of police pursuit, but all is well that ended well...

...no joke, no embellishment ...back in the day, I put up a 'run' from Vancouver to Whistler in a modded C4 'Vette at 48 min flat at just past 5 am...in those days, I worked for the feds as an economist, and I knew from a buddy who worked for the RCMP computer department that between 4.30 am and 5.30 am, the RCMP shift change was happening, and they were all at the Tim Hortons in Squamish...

...I snuck by at very high speed when you were at the doughnut place ...c'est la vie

...that's what most of my builds end up like. I don't mind some of the more expensive cases as a starting point for custom work - things like custom-made tempered glass, stainless steel posts and USB/C + / power cables can be expensive on their own. Still, my trusty Dremel tools (I got several ) and I like on-off solutions as the outcome....the TT Core P8 for example with dual side-by-side ATX mobos and water-cooled components. I used to stuff eATX server boards w/ water-cooled CPUs and 4x GPUs into Antec 300 ATX cases (still have three of those for firewalls etc). While they are 'flexi-flyers' re. their construction and metals, they were easy to mod because of it

...you mean...?

...given the 'heavy metal' (literally) of the old server case I posted earlier and also this one, I am thinking that it would make a great skeleton for a custom build ...apart from HD Dremel wheels, the only real cost would be a custom tempered glass front, or may be thick acrylic...I bet I could load that thing up with 2x ATX and 2x ITX mobos for a 'quad' build project

Nice collection & trip down memory lane ! ...I recently dug around in the companies' store room and came up with this old monster server case (w/ ABitIC7 and Zalman flower cooler inside). The server case itself is about 25 years old and the top cover is off as it contains a separate compartment for a 2nd PSU. We have four of these server cases that ran our software business back in the day and each weighs a ton but can support upwards of 400 pounds of weight on top w/o flexing as it is really thick metal gauge. ...too bad about the 'institutional beige-yellow' colour scheme of those years. After a much-needed cleaning, I might get it powder-coated for a future build project. Also, I have to remember to be careful as in those days, the sharp edges of the interior would cut your hand instantly...

...and another 'tech' video involving some of the accelerators in the OP I watched last night. It is time-stamped re. losses just from wiring re. memory and for large data centers and AI operators this adds up real fast. The rest of the video is also worth watching (as is the channel in general) re. just how much energy gets wasted, how they're starting to hit the 'memory wall' and the drive to get to processor-in-memory chips to cut efficiency losses and get past the memory wall.

...I am talking about efficiency - when you have two cards that both run at 400 W but one produces much higher results with that 400W (score, fps), it is more efficient in that it has less heat loss. Dumping the same 400 W into s.th does NOT automatically create the same external heat output. As a somewhat distant but akin example, when cars where first turbocharged en masse, there was a lot of energy loss due to heat loss (less of the fuel transferred into actual work, more into waste heat). Then they stared coating the headers etc, and finally moved into what is called 'hot-V' in a V6 or V8 But @Sir Beregond and @pioneerisloud , I was really just trying to help the OP re. his question on power connections for the 4090, and also compared it directly to my 3090 (up to 520W) in the same loop w/everything else the same...

...obviously, no need for coals to Newcastle as 'Watt' is used to quantify the rate of energy transfer. If you read my previous post carefully, I wrote the card is running cooler at the same wattage...for whatever reason, the 4090s are more efficient at the same wattage than the previous gen, and less warm metal in a case probably also helps with that efficiency. Overall, the 4090 gen seems to have fewer heat losses > higher efficiency.

...I thought @Sir Beregond was talking about heat in his case when he wrote '...sometimes this 3080 Ti with its 350W really is dumping out way too much heat'. On that, my point was that at the same wattage, my 4090 runs cooler than my 3090 in an identical cooling and case setup, including the area around the mobo, PCH etc. Not only is it a different node (4N) but the cards' pcb and water-block are MUCH smaller and more concentrated than the 3090 (and 6900XT) blocks - less overall warm metal in your case to heat things up. It's kind of funny if you strip a 4090 from its giant stock air-coolers and are left with just the pcb and water-block.

Oddly enough, the 4090 runs about the same temp to slightly cooler than the 3090 in exactly the same (extensive) loop. The 4090s do like higher VRAM temps than I am running though, but I can always turn the thermostat for central heat up if I want to bench

...some more OP-related head-spinning future tech for GPUs...somewhat speculative of course, but interesting nonetheless...

Per @iamjanco 's comment above, I got my Gigabyte Gaming OC RTX 4090 dual-bios model over a month ago...a local outlet of a national chain had about 30 or so for the equivalent unit price of US$ 1,619 but since those early few days, everything on the 4090 front (unlike the 4080s) seems to be sold out. In some tasks, the 4090s are almost twice as powerful as a well-running 3090.... Re. power consumption, yes they can get to 600W (more w/ transient spikes though they seem better-controlled in this gen), depending on the model. Typically though, they run very efficiently with far less power than RTX2K and RTX3K...for most games such as Cyberpunk '77 and FS 2020, I use about 380W-400W with a mild oc, and only hard benching will get me to 600W per a few adjustments in MSI Afterburner. The 4-into-1 PCIe 8 pin-to-1 x 12VHPWR that comes with the cards in their boxes do work, though great care has to be taken that they are not twisted and also FULLY inserted (no gap). Apart from the aforementioned 4-into-1 dongle, I also got a single cable from Cablemod (pic below) that uses 4x PCIe 8 (=6+2) pin connections at the PSU end. In addition, Seasonic was kind enough to send another 12VHPWR cable for free (for a recently purchased Seasonic PX1300W unit) and that cable 'only' has 2x PCIe 8 pin connections at the PSU end but is also rated for 600W (stamped right on it; wire gauge also play a big role in all this). I haven't tried that one yet but have seen other folks report that it works fine at 600W. The 12VHPWR cables are actually 12 + 4 pins (the latter 4 are sense pins / wires re. PSU capabilities). I do think you would have issues reaching full power with just using 2x 6 pin PCIe - or even powering up at all. In any case, Cablemod has a nice configurator > here - perhaps your model is still part of their database even if you don't buy from them, re. additional info.

...why not go all out for you next build ?