17 POUNDS of Motherboard! A Quick Look At The WRX90E SAGE SE from Asus! 

Remember the times when good (desktop) motherboards could be had for 200 bucks... nowadays you're not even getting a Q-code LED anymore for that money.

@@Blacklands A yeah, my Ivy Bridge Mainbaord had 7 Segment displays, V-Points for measuring voltages, really good VRMs and all that stuff. And my Socket 2011 board had 7 Segment displays, a LSI Raid controller, and all that stuff. The former was 200 bucks, the later 600. When i look at Mainbaords now... yeah i pay 600 bucks for a desktop Mainboard with similar features... and Workstation baords with similar features are also way above 1000.

$10K for the CPU.

They do that for all new forum members, only one time, but you will keep posting...and hoping 😂😂😂

Wendell's the BOSS. Helped me debug why 1/2 the memory in my 64 core system wasn't working. Haven't been on the forum in a long time. My bad.

@@vincei4252 that sounds like an interesting story. Always cool when people help each other, especially people that have such busy schedules.

What xmp what board and what cpu

I've been struggling with EXPO as well. Got a 2x16 5600 kit (HyperX Fury KF556C36BBEAKZ-32) With a Gigabyte PG X670E lightning. Can't get past 4800 without occasional freezes. I even turned off PBO because I thought it was that. I recently updated the BIOS but I haven't got around to play with the freqs yet. The original thing that brought me to comment was that I pulled the model number from the gigabyte QVL and looked through the website with that in hand. Can't find the capacity I want? I look for the next QVL kit with that capacity available to purchase

I haven't had a single PC build where xmp or expo actually worked. Only one would boot and it would bsod constantly. I had to manually key in slightly friendlier timings. My first build was an athlon xp so I've been doing it a while ...

"Pro" from Asus, not exactly the same as Pro from Dell/Lenovo/HP with a support contract

What memory kit are you using and is your system stable?

I am using thr kingston KF560R32RBK8-256 100% stable at 6000.

Bend over and I'll show ya! jk, jk 😂 somebody had to say it

New coffee table confirmed 👍

You mean 170 satoshi's at ai express.

Had a ASUS X99 server board that would crash under load or when gpus were inserted in slots 4-7. Turned out it was missing thermal pads on the VRM's and one of the PLX chips. Manufacturing error. Bought $20 of FujiPoly thermal pads to fix it. Rock solid afterwards and still running.

Yes, it definitely sucks. It'd suck enough to even go with Intel instead, except they don't have any competitors at all. Which is probably the only reason the stupid people at AMD were able to prevail and prevent Zen 3 on TRX40.

Good eye! I love you! This board can do nps3 but idk how to trigger it consistently!!! MADDENING

For 1300 bucks you want them to be like Supermicro...

@@falsevacuum1988 it still has more & better options than most Desktop boards that cost that much now.

Yeah, how are these even connected?? I'm so confused.

I once got a Seasonic X650 PSU DOA. That sucked. The replacement is still going after 7 years though

@@iaial0 Yeah, personally I got one DOA Seasonic PSU (actually blowing the circuit but nothing else, not even another component!) and one that gave up the ghost after a few years. The other... maybe 10 I bought / advised on are a-OK. Even a 10+yo passive 400W gold, first series one. It's powering what I'm writing these lines on, 24/7 for most of the duration too.

NEMIX RAM 8x256GB ECC DDR5 5600CL40 1.1v (20,223$) Just a reminder of what you need to squeeze out the full potential of current HEDT CPUs

You're talking about NVLink not SLI on 2080ti's. NVLink is completely over the bridge except on very specific, Nvidia made systems. You can actually monitor PCIE traffic and see there is no significant uptick in traffic during a render, or other CUDA task, when using the NVLink.

@@KenS1267 No actual WRX80 all have both S.L.I & NvLink support listed in the instruction manuals. Beside I was reading some forums about bunch of indie developers who bought two RTX 4090 when the released. They could not get NVlink to work at all on some CUDA applications that were supposed to use two to four cards.

@@kevinerbs2778 LOL There's no quad NVLink for 4090's. The last quad bridges were for the 2000 series, or maybe it was SLI on the 1000's. There are 4 slot wide bridges because you need bridges that wide to connect the card but they only connect two cards. I have 2x 4090 running NVLink just fine on server boards without any sort of SLI cert right now. CUDA applications flat don't care how many cards you have. That's the whole point of NVLink. They turn more than one card into a single logical CUDA device. You can get the NVidia systems with NVLink through the motherboard and all the A100's and that whole thing looks to CUDA like one GPU despite there being 16 or something.

​@@kevinerbs2778if you don't even know such simple things, i doubt you need either of them

@@toseltreps1101 you act as if i don't kno what each is for when in fact i do know what for? You do know that they're different features right?

Unfortunate. I have not had any problem running an oc’d i9-10980xe, 64 GB ddr4 Corsair Vengeance Pro, with an RTX a4500 and a Hyper m.2 v2 add in bifurcation card with 4 1TB Samsung 970 EVO Plus nvme ssd’s. Also 2 970 EVO’s on the motherboard for OS and Scratch Disk. No AIO, but instead use a Noctua NH-D15S for cpu cooler. System running flawless, no blue screen or freezes, going on 4 years now. Was thinking of upgrading to an Asus W790 Sage soon;, once all the bugs have been worked out, perhaps next year.

Xeon W-3400 (2023). And it loosing to Threadripper 5000 while having problem with 700 Watt power spikes. So in fact no alternative...

Memory training is finding the physical sweet spot between memory, memory socket, CPU socket and CPU. The chips, brand, packaging, etc have a lot of variation. It's like a rehearsal and finding what electrical settings work for stable communication among all the components.

@@Level1Techs so the CPU running though a bunch of values to get as clean of a signal between itself and the memory then. How many parameters is it running through?

@@konic40 what it's trying to do is train the sub timings of the memory modules to sync the modules with the memory controller. memory manufactures primarily focus on the main timings e.g. the ones you see in marketing while leaving the sub timings loose for compatibility across multiple memory controllers since no memory controller is the same between even identical cpu's. it'll change the timing, run a test across all available memory and will continue to do that until it fails, reboot and slightly adjust the timing and retest until it fails. it'll make 3 attempts and automatically go back to the known good timing and reboot for the final time. there is an option to save those settings and bypass the retraining on post but asus leaves it off by default because they're too aggressive with their sub timings. even with the option set though if the system for some reason fails to post due to memory it'll attempt to retrain the memory again.

It's essentially the same thing your cable/DSL modem does at startup. Trying various power and frequency combinations to reach the best speeds

It was hiding his street address on the shipping label. The Newegg label was clearly visible in a later blur shot. The first one was evidently made larger to avoid redoing the blur to handle box movements.

S3 yes, and Intel just issued a firmware/microcode update to correct some performance degredation there

8 channel of 5600MT/s DDR5 RDIMM will offer around 360 GB/s. What's why you want to stack industrial GPU's with HMB memory for AI. Also don't believe Apple marketing, 800GB/s is for GPU not for CPU part of SoC.

@@falsevacuum1988 but that’s exactly why ai enthusiasts buy Apple’s m ultra series. Deep learning is done mostly in its gpu plus neural engine. Can we get increased memory bandwidth if amd offers an ai APU for their threadripper systems? Is the memory bus width determined by the APU in this instance?

@@theworddoner Amd did MI300X and MI300A for AI. And yes, memory bandwidth determined by the APU, on x86 CPU its sucks and this phenomena called "Memory Wall", and Threadripper 7000 doesn't have neuronal engine nor integrated GPU.

@@falsevacuum1988 Sorry if this question seems dumb. But could they not make an AI APU for the threadripper system? On the consumer side AMD is rumoured to make strix halo which is an APU said to have 256 bit bus width. It should have decent memory bandwidth when paired with quad channels. Is it not possible to make a stronger version of that chip for the threadripper system?

@@theworddoner No one knows AMD future plans lol.

Yes it will absolutely throttle. And at a lower temp than you expect. 65c arounds there in my testing

My concern is longevity. Tjmax in the memory chip should be around 95°c. Measuring with hwinfo or bmc shows us the pcb temp, not chip. I also find it disappointing when a workstation throttles.

​@@Level1Techs do you think mounting fans on the ram is sufficient? Also can you share which 64GB dimms you've been able to post on the asus wrx90? I need to build with 8x64GB and am starting to worry

​​@@applicablerobotyou can probably find RAM air coolers, and for once they're useful and not just for the looks

The funniest thing is that, because of this, now EK water blocks for RAM begin to make sense, like 10 years after they first came out!

I believe it's 200W for the total memory, because otherwise it would be 1600W DIMMs only, which would set them on fire I reckon