Etern4l

Sorry, it's called "Elevate", there is also Lumosity.

Without splitting hairs, it's a 3090 Ti from the gamer's perspective, but with less RAM, and I would expect some performance delta on larger workloads (actually utilising those 16GB of RAM at least), as 4090M only has half the memory bandwidth of the 3090Ti/4090.

Sorry, not really, unless you count "brain training" apps like Peak.

The TS score is stock 3090 Ti FE. Goes higher with OC. I guess that's impressive or disappointing, depending on how you want to look at it.

Yeah, 2R /32GB DDR5 (and now 48GB!) modules just run slower, and 4 modules are yet another level of challenge.

Just took a quick look: Single socket: EPYC Dual socket: half EPYC, half Xeon Quad socket: Xeon That's just the front page, didn't delve in to see if it's possible to configure a single-socket Xeon. https://www.dell.com/en-uk/dt/servers/poweredge-rack-servers.htm#tab0=0&accordion0

The ribbon one is pretty cool, but it's the slightly cheaper option: https://www.corsair.com/uk/en/Categories/Products/Accessories-|-Parts/PC-Components/Power-Supplies/600W-PCIe-5-0-12VHPWR-Type-4-PSU-Power-Cable/p/CP-8920284 Mine has individually sleeved cables which gives it that extra flexibility: https://www.corsair.com/uk/en/Categories/Products/Accessories-|-Parts/PC-Components/Power-Supplies/Premium-Individually-Sleeved-Type-4-12VHPWR-Cable/p/CP-8920331 Bends effortlessly at 90 degrees such that the cable requires about 2-3 cm clearance from the card. To be fair, that depends on the card, the socket is fairly deeply seated on the 3090 Ti FE. In summary, yes - looks like I might gain maybe 5mm of clearance without any need for cable bend. Something to consider if I need to stick a 4090 in. The question though is about the reliability of the L-adapter itself, and I am a big fan of the KISS principle.

While that may be true, I would like to think of myself as "minimally biased" (obviously, there tends to be some degree of ownership bias, unless the owner is deeply unhappy). I evaluated AMD 5000 series very closely and fairly at the time of the last personal purchase decision, and wasted no time to point out the flaws on the Intel side (below average likes on those posts lol). Here, I've been mostly asking for more information :) The picture clarifies things. That 12VHPWR cable looks much thicker than the Corsair one, which makes sense given the intended use.

I quote: "Also, note that Intel is still expected to remain the data center market leader by a wide margin in 2023. Hence, if the company could stem and reverse the slide moving forward, it could lift buying sentiments further and put pressure on AMD's more expensive valuation." I didn't know that, to me thst would be dependent on the use case for the servers. Do you need FPGAs to accelerate databases and web servers? My question was more concrete though: what's the size of the FPGA market relative to say the total global DCAI spend. My guess is it's tiny, a few %. It thought bitcoin miners are ASICs rather than FPGAs? At least that's what they are called. Respectfully, that 12VHPWR dongle looks like abother useless gimmick pushed by the influencers. Would take more vertical space than the native Corsair cable.

Right, clearly there are a number of dimensions which you would need to score individually. A single aggregate number for all aspects of intelligence would be meaningless, because it would depend on relative weights for each of the dimensions.

Are they eating it (the CPUs) or just approaching the table, hoping some stuff is left for them once Intel leaves the table full? Genuine question that can't be answered by chest-thumping marketing materials alone. As for the significance of Xilinx, how big is the FPGA market in relation to data centre CPUs for example?

Clear, no "bling-bling".

2 sticks of Kingston 32GB would run fine at 5600, or a bit higher, on the desktop, but the default latencies would likely be 40-40-40-40 and forgot the last one, 80-sh, you can look it up on Kingston website. Try that and maybe higher voltages, althoughmy experience with Kingston DDR5 suggests this won't help here. Does it revert straight away or after a long period of training? Quick reversion would indicate some lower-level issue with the BIOS. BTWII recall watching a video where a claim was made that RGB variants of modules tend to perform not as well. Have you guys had similar experiences? I don't care about RGB, but sometimes it's the only variant available. @Mr. Fox @Papusan @tps3443 @Rage Set

FYI Kingston Fury Beast 5600 DDR5 modules (32GB at least, and others too I think) have 3 profiles: 5600CL40. 5200CL40 and 4800CL38. This is very handy due to all the issues with 4 2R DIMM support, and AFAIK nobody on the planet has managed to run 128GB at 5600MHz with 12/13 gen. AMD seems to struggle even more, another reason I made the right choice by going with Intel. I'm currently running mine at 5000CL30-38 super stable apparently. The performance improvment from this + the new mobo + the delid is insane. +15-20% roughly in practice, feels like 14th gen. Now I'm looking at these new 48GB modules from Corsair. They come in 5600CL40 and 5200CL38 variants (2x48GB kit, 4x48GB is only available as 5200). I know I will only be able to run them at 5200 or 5000, however, I'm assuming that those 5600 modules (just $10 or so more expensive) are better binned and might be able to run at lower latencies. Also MSI demoed 192GB running at 5200, actually using those 5600 modules. I guess I should do things the hard way as usual, and get the 5600 modules, then manually set the basic latencies and voltage, and hope the board does a good job with the rest. I'm not tweaking 30 RAM timings by hand lol Good return policy will be essential.

Is anyone familiar with Corsair Vengeance DDR5 modules? I'm wondering whether they come wth multiple XMP profiles preconfigured, e.g.whether the 5600 version also has profiles 2 and 3 for 5200 amd 4800 MHz. Corsair just lists a single XMP profile for these.

Alas, it's not that simple, quote: "The test administrator should look way beyond global IQ scores."

Visual, spatial, conceptual, linguistic, etc. I'm sure some aspects of it correlate with certain tests, but it's a fairly indirect approach.

The problem is that those "general intelligence" tests focus almost exclusively on problem solving and perhaps recall, whereas the two very important (probably more important overall) aspects of human intelligence, namely executive function and creativity are harder to test I guess, and thus not included in those metrics. Usually not much on language processing either, which - as we know - is quite important.

1.43V... What was the power draw in CB23? 400W lol? BTW w has t do you do with ring voltage?

Strange name. Would be more accurate to call this a Micro Society (micro=1 in a million). Perhaps it's just short for The Mega-Genius Society! Any famous members?

Yeah, 1 day would capture any curing behaviour (very pronounced with the Honeywell stuff), and1 week+ would be useful to assess pump-out resistance, although this is hugely dependent on heatsink fit, contact area size, heat flux, workload pattern etc.

The new ACE Z690, made in 2022 rather than 2021 (kind of along the lines of what you guys were saying about the Apex) is too good to be true. 1. The socket is different, the pins specifically look somewhat more slivery - hope that will prove more resilient and stable long term 2. Heatsink fit using the ILM ranges from very good on the original IHS to perfect on the copper IHS - no contact frame needed at all, and with just hand-tightened heatsink screws. Day and night vs the old setup with the contact frame and basically foolproof. 3. Memory stability is hugely improved: running 5000MHz with 28-36-36-36. Can't believe it, was never able to even train at CL28. With this it seems like it either doesn't train or if it does it works with no errors, in starrk contrast to the old board 4. CPU runs at lower voltages - currently testing 53/43 which achieves 40500 in CB23 (in Linux, which does about 500 points better than Windows), at 1.16V/235W, with temps in the eighties. Pretty good for an AIO. Dlidded CPU is absolutely fine on PTM7950+Apex, and definitely part of the early success story. The experiment continues..

Agreed if you are building a workstation with a $20,000+ budget. Which is what you would need (OK, maybe a bit less) to get in the vicinity of the raw performance provided by 13th gen + gaming GPUs rig if you are limited to using overpriced 2 slot compute/workstation GPUs. It' s a bit of a hack, I know, but a pretty cool one, if you learn to navigate things like the idiotic LGA1700 socket, and to manage power to avoid a meltdown. Host I/O bandwidth is often of little concern in compute tasks, so there is no appreciable penalty due to the TB connection. Notice how miners (used to lol) use 1x raisers to connect their GPUs.

Data analysis/ML stuff, quite memory intensive. TB ports provide the option to connect additional GPUs, 10GbE (if the board lacks this), or fast external storage - very handy and a major gap in the AMD offering from my perspective. Xeon boards also lack TB support from what I have seen so far which is a shame.

All makes sense. I incorrectly assumed Apex was the equivalent of Godlike. It's a XOC 2 DIMM board, a competitor of EVGA Dark, which is why I never took a closer look. I need 4DIMMs, E-ATX, at least 2 PCIe slots, two TB ports and best reliability/build quality/QC. ASrock Taichi aside, the main options are MSI ACE as well as Asus Hero and ProArt. Asus Hero failed the smell test due to the burning cap issue, while ProArt is weakly built, and had lesser memory OC potential. The ACE also did consistently well in benchmarks against those two Asus boards. MSI board specs are a bit more detailed/technical as well, which I like, for example they state things like TB controller model etc. For some reason though, Asus probably is valued by a lot of people, since availability tends to be worse. I suspect BIOS / Windows software or perhaps more aggressive launch pricing could be the reason (at the cost to QC possibly). Whatever, I didn't even review Asus options this time around since I saw @Mr. Fox's complaints about Linux compatibility of an Asus board and want none of that.