Flavlus

Yeah gave up on it. Too much town to get a board that supports it. And even then , a case for it to fit both and so on. Gets real expensive. Went 64gb ram to fo the big ram stuff on cpu even if slower than gpu.

Any thoughts on the rest of the build?

any reason why?

$Pricing is in NZD. Will mainly be using this machine for AI / Deep Learning work so need the Ram and Vram. (maybe a little gaming on the side) Looking at AM5 mainly due to high core counts and the potential to upgrade later to Ryzen 9000 or whatever will be the last of AM5. In saying this will this motherboard last or should I upgrade? Options include: Gigabyte X670 GAMING X AX V2 $480 MSI PRO B650-P WIFI ATX $399 Gigabyte B650 GAMING X AX $380 Unsure for future upgrades if the B boards will be fine. if so which one? What should I look for here? PCPartPicker Part List: https://nz.pcpartpicker.com/list/vc92jH CPU: AMD Ryzen 9 7900 3.6 GHz 12-Core Processor ($659.00 @ PB Technologies) CPU Cooler: Thermalright Peerless Assassin 120 SE ARGB 66.17 CFM CPU Cooler ($75.00 @ Computer Lounge) Motherboard: MSI PRO B650-P WIFI ATX AM5 Motherboard ($399.00 @ Computer Lounge) Memory: G.Skill Flare X5 64 GB (2 x 32 GB) DDR5-6000 CL30 Memory ($368.00) Video Card: MSI GeForce RTX 3090 SUPRIM X 24G ($1000.00) Power Supply: Deepcool PX-G 1200 W 80+ Gold Certified Fully Modular ATX Power Supply ($336.18 @ PB Technologies) Total: $2837.18 Budget is around $3000 NZD (I have a case for around $150 to add but its not on part picker) THANKS IN ADVANCE

Yeah ive tried. Its seems to pick certain things it like and like hallucinates the objects are where they are. Its just too small for this use case. I think 16 works .. Will have to play around with res and maybe try some smarts with tiling. Yeah just trying to find a board that does bifurcation is ideal. So many dont and its real annoying. Any easy way of telling? Im mainly looking to see if they have 2 reinforced (metal ) x16 slots. Looking at manuals is also a way but some are written terrible and done say how it can split at all. not even 4x. To reiterate, i don't care for speed and more care about the 48gb vram. Im fine even if it had 3070 speed but with the double VRAM. so trying to get that memory pooling to be a thing. So from my understanding, I need to get 2 of the same ish card ( same height) then the correct spacing NVLINK connector. All on a board that supports bifurcation x8x8. RIGHT? (or sell a kidney and buy an a100)

I was YOLOv8 training on image size 800 batch size 16. I tried 8 but it messes up the model. 16 seems to be the sweetspot If I could use 16 and split it between both gpus. Speed is not an issue really. A100 takes like 10 mins to run so even if this takes 30 mins but I can run locally .. all goods. It's mainly for testing purposes. Will deploy to an h100 or something if I need to do big workloads.

Ok so if I have a set of images I'm training in with even a small batch size I'm currently using about 38gb vram through Google Collab and similiar platforms. Would 2 x 3090 let me run the training or will I have out of memory errors? Ram I'll see on pricing. I think I wanna start out 32gb and see to add another 32gb. I'm mainly gpu training and such so never had high ram usage.. maybe 15gb to 20gb usage. Especially if I go ddr5 .. more $

With the exchange rate they seem about double the cost of US

Regarding memory pooling. Nvidia says its a NVLink benefit. The newer nvlink. The ones pre rtx didn't do this. So much contradicting info

Thanks for the response. From my understanding 3090 can do memory pooling. Through nvlink connection. But maybe not .. I'm confused now. Regarding motherboards seems like the US gets them much cheaper. Ones mentioned here in nz are $600+ nzd. So like $400 usd Not really that cheap. They are close to twice the price of a r5 7600 to put things in perspective.

Hello, I'm doing a new build based around a 3090, but I will have 2 in the future using NVLINK to utilize memory pooling for Machine and Deep learning. The issue is I cannot find any decently priced Motherboard for Intel 13/14th gen or on the AMD AM% that supports Bifuraction x8x8 allowing the 2 pcie slots to run x8 each to get some decent performance. This seemed to not be an issue on AM4 and so many motherboards even cheaper ones were supporting it. Do I need to go with older intel or AM4 or is there something newer that supports it? Thanks in advance

Thanks for the responses, I found a good option for a r9 3950x on an X570 Aorus Master. Its should be able to do x8 x8 with bifurcation according to the documentation. its a 16 core 32 thread but an older gen. Im wondering is this would be a good option or should I spend a little more and go r9 7900 (non x) which put me on a DDR5 platform and potentially more upgradability in the future. This would cost more now and be 4 cores, 8 threads down on the 3950x. Or am I over thinking this and something like a 3600 or 5500 is enough? Thanks

Hello, I'm looking for some advice on building an ML / LLM / AI machine. I'm looking at starting with a RTX 3090 because of the 24GB of VRAM and the ability to go NVLINK to 2 x 3090 in the future for that sweet 48GB with Memory pooling. The 3090 seems to be the best bang for the buck since it's around 1/3rd the price of a 4090 here. (used) Im needing as much VRAM as I can get my hands on as I'm working with images as well with object detection (this is very VRAM heavy) In terms of the rest of the build that will be happy with a 3090 or even maybe 2 in the future what should I look for? Do I need 16 pcie lanes per card? Any suggestions. Keen to find something used and maybe upgrade later but dont know where to start really. Seems like I can find some banger deals on R5 3600 right now (everyone upgrading). Do people recommend more cores or faster cores? DDR4 or DDR5? Thanks to all in advance

LINUS NECK FLARE : minute 17:02 fo Mac Studio M2 Ultra review

Yeah i heard the 5930k might overclock a little higher but also might not Considering my stock boost clock was 3.6 and now im at 4.5 thats allready a massive jump. Better than my i7-2600k that couldnt hold more than 4.2ghz for any reason. I think i lost the lottery on that one. Then my 4770k Was not stable above 4ghz untill i delid it then it was ok 4.2 and its still an old DDR3 system. Meh. Im selling that and moving on