J7SC_Orion

Fortunately, Seasonic is offering a single strand 12VHPWR PSU cable - apparently for free if you bought a new one recently (which I did, 2x PX 1300). Even if there's a small cost, I think it is a wise investment. For the record, I haven't had any problem with that hideous 4-into-1 dongle on my RTX 4090, though I was careful when installing it re. bending and twisting as I had read about it before. In addition, I mounted a 120mm fan pointing at the back of the card which also cools that whole power connector plus the cables leading to it.

wanna see s.th. related (no, its not a 12VHPWR...) that will keep you > 'warm and fuzzy' ?

...amazing how quickly things change, or at least perception of things - my 5950X is still good enough and in TSEX, the fastest combo w/4090 at 3DM. That is not to say that I have not thought about, say, a 13600K or 13700K as an intermediate solution, but it is simply too late to get into LG 1700 now. Besides, I hear that the real fireworks will start with Intel's Meteor Lake (LG 1800 ?). Then there are rumours of AMD getting ready to fight back with updated 79xx (3DV?) AM5s. Those exist already, but that doesn't mean that they'll get released to market (yet, or at all). Case in point is SkatterBencher's vid below on the AMD 5990X (4x 5950X in a TR package, non-pro, overclockable). Since AMD's desktop (not their TR) sockets last a lot longer than Intel's shenanigans around 'socket games', if anything, I would consider an AM5... The Ravens' two CH8s (Asus Hero, Dark Hero) were the first non-HEDT systems I build in about 8 years or so - I usually recycle these setups into work-related builds once they have been 'thoroughly tested' (if you know what I mean ), so there are plenty of rumours now about new TRs and also new Intel HEDTs coming. HEDT typically offers up to 7x PCIe 16 which helps in post-fun-now-work setups that carry additional PCIe cards, and also double or more of the memory channels. Then again, X570, LG1700++ certainly offer enough cores and RAM for workstations and light server duty, and the built-in 10G and PCIe 4,5 multiple drives are already near HEDT territory - just get the more reliable enterprise drives to add, ie. U.x drives connected through adapters to the mobo M.2 slots. What many folks don't know is that workstation / enterprise mobos are usually not as robust around some components as top-of-line gamer boards (VRM for example) as they're meant to sit in 130 dB air streams... ...speaking of air streams...air-cooled, fans on 100%, and only a light 3D load to keep temps in check... Anyway, first things first: Water-block for the 4090 ! Finally, let's both be careful with those dongled 12vhpwr connectors - fortunately, Seasonic offers free upgrades (and also for sale) of the single-strand 12vhpwr for my PX1300s

...in that case, a 4090 (or AMD Navi3) is calling to you ...also, they can serve as heaters for the coming winter...no joke, I actually utilized 2x 2080 Ti (combined 760W) to heat a big room when the heat went off during - 10 C last winter...

Fortunately, I had read enough about the new connector to be extra careful when hooking up my 4090 to the PSU via that 4-into-1 dongle. Still, worth keeping a watchful eye on....I just posted my musings on the subject at another forum...so some cut and paste action below: "I am not too thrilled about this new power connector, either. I suppose that if they would have gone for a traditional 4x 8 pin PCIe, the pcbs would have been wider by at least an inch - besides, it is clear that they wanted to move the power delivery as close to the actual die as possible. Still, I noticed before that on my 3090 with 3x 8 pin PCIe and 520 W vbios, each of the PCIe cables (stock ones that come with Seasonic Prime Platinum PX 1300) would get warm - not hot, but warm during heavy GPU loads. At the same time, with a full waterblock, thermal putty and all that, the 3090's VRM didn't exceeded mid 30s C (HWInfo has sensors for that for the 3090 Strix). With these 4 into 1 dongles for the 4090, it is clear that all the PCIe supply cables are bunching together to join into a single connector with a much smaller total area than 4x 8 pin PCIE would have had...so clearly less area for heat dissipation while transferring more energy. Then there is this whole thing about 'do not bend' for 3 cm to 3.5 cm cable length at the joint end. I understand why, but for a consumer product that goes into people's homes, it is worth questioning these release decisions...Hopefully, most PSU manufacturers will make the required new cables w/o dongle available, even for some older models." FYI, I have partially melted some 'traditional' PCIe 8 pin plugs on a GPU before, and that wasn't even during HWBot with HEDT and quad cards pulling more than a combined 3000 W

I really wouldn't know much about 1440p as I game exclusively at 4K. However, 3DM Port Royal is 1440p and I have run that - RTX 4090 is basically double that of a well-running RTX 3090...

I put together a brief summary of the 4090 experience so far > here. For the Ravens' builds, soon there will ensue a game of musical chairs between GPUs and mobos - once I got a full custom water-block for the 4090 Gaming OC. The air cooler can't keep up beyond 500 W (ok though at below that), and hotspot temps just go too high to sustain the clocks it can run at lighter loads on air.

On the pricing, I reiterate that it requires a cost-benefit analysis each person / business has to do on their own because it is not cheap - then again, the pricing here is actually competitive (a bit lower even) than many custom 3090s were at launch before the crazy price bulge due to mining, the pandemic-stay-at-home syndrome, scalpers and other factors. Still, this new class of GPUs is in a performance segment all on its own (until AMD joins, and NVidia follows up with other RTX4ks). For the top games and sims I play, the 4090 fills in what the fast 3090 couldn't re. 4K 120 OLED. I am also intrigued by the particular model I got (Gigabyte Gaming OC). It's relatively low pricing (as opposed to for example the 4090 Strix, $400 more) belies its performance. In any case, there are several indications that this was originally meant as an Aorus model (including some bits of the accompanying documentation). Perhaps Gigabyte read the market conditions and quickly rebranded and repriced, though I don't know for sure. In more general terms, in spite of the valiant efforts to produce air-cooled models by various vendors (3.5 to 4 slot cards, unless AIO), these things need full water custom cooling bocks to get anywhere near their full potential. The 600 W max actually doesn't bother me (as much as it is more like 375 W - 400 W in gaming) since the 3090 Strix already sucked back > 520 W max, and all the custom loops I build are configured to handle a sustained 1200 W (ie. HEDT, 2x GPUs).

I recently added a Gigabyte Gaming OC RTX 4090 and here are some early conclusions, details and pics... 1. "Jekyll & Hyde" - at normal gaming, the 4090 is quiet and bombastically fast while also being more efficient than my 3090 Strix. At full bore, it is in a completely different league of anything else (so far...), and while the behemoth air cooler works well for most circumstances, it needs (and will get) a custom water block for the 'final 100 W' (beyond 500W to 600W). On air, I simply cannot run full-bore for extended periods due to hotspot temps...VRAM temps are great though with this model. 2. As much fun as it is, if you game at 1440p and have a nice 3080 / Ti or 3090 / Ti, you probably won't '''need''' a 4090...but I game on a 48 inch 4K 120 OLED, and this card makes that well worth it. Unlike the 3090 where DX12 was generally better than with Vulkan, the 4090 seems to like Vulkan a lot. This is also the only card I know of which might be CPU-bound at 4K, unless you are running the absolute latest CPU and mobo from AMD or Intel. It will make you want to upgrade everything - the (in)famous slippery slope attached to a new hardware purchase... 3. DLSS3 is jaw-dropping in its apparent visual improvements, even beyond the massive rasterization increase of the RTX 4090. DLSS3 is supposed to be RTX4K only, though there may be some partial driver hacks for other older models out there. Still, the RTX 4090 also has some specific DLSS3 cores and other unique hardware....speaking of hardware: 4. The RTX 4090 is referenced as a 4 nm node, with 76.3 billion transistors, a 608 sq mm die and a 72 MB Cache - the RTX 3090 is on a 8 nm node, with 28.3 billion transistors, a 628 sq mm die and 6 MB of cache. Clearly, heat management re. hotspot is a priority with such concentrated power. 5. The Gigabyte Gaming OC was US$ 1,619 ($20 more than 4090 entry at time of writing) and it seems to be very good quality, apart from it being a good sample (below). With the economy shifting gears, as well as inflation, supply chain and many other serious issues, this is a purchase to be carefully thought about. In my case, I can recycle most equipment into a software-related business, so the old soldiers march on - but still, the cost-benefit of such a purchase has to be taken into account. Of course, the 4090 is just the first salvo in this latest GPU war; AMD's RDNA3 is likely to be competitive with at least the upcoming 4080 (16 GB) and who knows, with its new mChiplet approach and patented infinity-fabric-type connections (appearing as a single GPU), AMD might yet spring a (really) big surprise, may be a bit later. Then there is the upcoming 4090 Ti. With the increasingly tentative economy, inflation nibbling on general budgets and many other geo-political uncertainties, it might even be that even crazier releases by either AMD or NVidia (+ Intel ?) will follow at a quicker pace and competitive pricing as the PC market has already slowed significantly... For my current purposes though, the sheer 4K 120 speed and DLSS3 on top of that made this a no-brainer at the entry-level pricing (and local availability). The actual sample and its additional oc headroom (on air, stock 2x vbios, 600 W max) turned out to be the icing on the cake that both Jekyll & Hyde can enjoy... Big, bigger, behemoth-big... One custom water-block, please... ...even on air, this card never drops below 3 GHz in benches, and with light loads, it can reach up to 3.2 GHz...but with heavy loads and > than 500 W out of a possible 600 W, hotspot temps spoil the party.

...I leave my car to build up a protective layer of dirt...especially useful as we're nearing the winter, besides, it looks like one of those new 'flat' wraps. Also, we finally got some colder temps and rain, and I took the long way home from the store - looks like some folks went straight to Christmas decoration this weekend, while others are still doing the Halloween thing.

...the behemoth air-cooler on my Gigabyte 4090 Gaming OC is good for normal apps & gaming, but it needs water-cooling for extended benching sessions ! ...continuously bumping into max Hotspot temps before my known max clocks at 1.1V (> 3100 MHz). I did not use the 'full power' option in the NV tab yet for benching and let it downclock instead since Hotspot temps are the issue. VRAM at +1404 seems to be the most 'efficient'. Once w-cooled, I'll try 'full power'. 5950X is ok for Port Royal, but DDR5 and IPC improvements would help. 4090 water-block first though for the existing loop - it's a must. @Storm-Chaser ...did a 3DM Speedway (below) for the first time today, is that what you were referring to ?

...as mentioned before, I was going to wait a bit, but then it kind of fell into my lap. Still looking to find out more about AMD's RDNA 3x, but I definitely wanted a fast card with the full NV DLSS 3 ('full' because there appear to be some hacked versions out there for RTX2K/3K, but w/o the extra specific cores / hardware in the 4090). That 48 inch OLED can go up to 4K/120, so one thing led to another...

...first order of the day is to get a water-block - and so far, only an overpriced EK block is out there for my card. My earlier EK 'adventures' nixes that one, for now...hoping for Phanteks, Alphacool or even Bykski. The card is a great clocker though on air (over 3100 MHz already), but air-cooling this monster works for gaming, not for benching...76.3 billion transistors in 4 nm node concentrate heat Still working on finding VRAM max (most efficient), doing it in 100 MHz steps.

The Raven A GPUs adopted a cuckoo baby... ...going to be a tight fit until waterblocked...fyi, the horizontal green lines are the same length...

...I was going to wait for more info on RDNA3X and 4090 Ti, but the best-laid plans and all that. US$ 1,618 is not chump change, but this is a well-reviewed card (up-to 600W full-fat) that should finally allow me to hit the limits on the 48 inch 4K 120 Oled...DLSS3 on what I play was another factor - and then it turned out that a place near my home seemed to have the largest stock still available, sooo...

...I wouldn't call the X570 platform 'old', but since it is listed now by you folks, I'll sneak in the Asus Crosshair 8 Dark Hero. Superb platform for 5950X and 4-stick Samsung-B DDR4 4000 that hasn't missed a beat. It was/is the first Asus board to introduce the 'Dynamic OC' for simultaneous Ryzen top multi-core and single-core overclocks. Chipset is fanless as well.

...love the 990FX Sabertooth ! Speaking of ye ol' Abit (re. the A*R32-MVP Deluxe), somewhere in storage downstairs is this 'survivor' (black + white pic on the left to hide the dirt; missing one OCZ stick)

...I am a bit of a motherboard collector (not to say pack rat). Working in the computer-related field sometimes blurs the lines between work and play, but most of my motherboard choices were / are geared towards quality components that could serve for both OCing and later on as reliable dev servers. Back in the day, I also ran between 2 - 4 GPUs, so that also affected what I would buy for work or and/play. With the latest-gen mobos hitting the market, what are your three top-fav motherboards you still own, and/or use ? For me, it is the Asus Maximus V Extreme (Z77), the Gigabyte SOC Force (Z170) and the MSI X399 Creation....looking forward to see what you folks scurried away into the 'keepers' box

...no issues with drivers for my 6900XT, either.

Thick'n sticky Gelid GC Extrema doesn't squeeze out easily, either.

...6 + 8 pin PCIe, and, ahem, glue ? Still, could become a nice midrange card if they can get the drivers fully developed (which I think they will)

...I've run up to 5x D5 in series, but that was for a dual HEDT / mobo system with a total of 8x GPUs back in the day. In my applications, 2x D5s in series give you the best balance (and I never run less than 2x D5s for other technical reasons). 3x D5s start to get a bit tricky to bleed and synchronize but are doable. It is probably better to have separate loops with 2x D5s per, unless there massive head pressure is needed...

I updated performance collage of my 5950X - CH8 Dark Hero with two new HWInfo effective clock runs from today. On the left, L1L2 prefetch off, on the right L1L2 prefetch default. Using the Dark Hero's DynamicOC for daily now.

1200W and 48 GB or bust