Have some RTX 5090s, hmu if you have experience with local LLMs

[realboyo]

Some of you have the ideas but no hardware, I have the hardware but no ideas. dm to discuss

 

[Renophobeadoobe]

se

Some of you have the ideas but no hardware, I have the hardware but no ideas. dm to discuss

View attachment 5451108

send me one nigga

 

[sh1ny_flakes431]

NIgga how much u sell them

 

[realboyo]

NIgga how much u sell them

I don’t sell them, but if you want to flip 4090s and 5090s remove the coolers and ship the PCBs to Hong Kong (free port with no VAT, GST, or import tariffs). After that smuggle them to the mainland and they’ll pay you 5k+ usd even for entry level 5090s

 

[realboyo]

se

send me one nigga

Show tits

 

[6”3neartyronelite]

I don’t sell them, but if you want to flip 4090s and 5090s remove the coolers and ship the PCBs to Hong Kong (free port with no VAT, GST, or import tariffs). After that smuggle them to the mainland and they’ll pay you 5k+ usd even for entry level 5090s

shii this lowkey the method how long would this take to do usually

 

[cargo70]

Some of you have the ideas but no hardware, I have the hardware but no ideas. dm to discuss

View attachment 5451108

How much is that all worth, bro?

 

[erome]

Can u give me one

 

[realboyo]

shii this lowkey the method how long would this take to do usually

5-10 minutes to strip each cooler, min 3 days to ship to hk and then you get wired. You can also not remove the cooler it just costs a lot more to ship cuz the cooler/heatsink/baseplate is 10x bigger and heavier than the PCB itself don't recommend

Hardest part is sourcing 5090s (since everyone and their mom wnats that 32GB GDDR7), & figuring out the logistics to get it smuggled over the border. If your carrier realizes you do not have the ability to source reliably for long-term business they will rape you (they take upon the brunt of the risk, given the illegality of the situation) so don't do it if you're only able to send over one every 2 months.

I don't do ts myself (bought ages ago to hold as an investment vehicle rather than to resell immediately) so I'm not experienced but the reason I have minimal info is from a local I sell memory products to who has shipped hundreds of 4090s + 5090s in the past year, if you have any specific questions I can try to ask him as long the question isn't retarded.

If it happens to be the case where you're unable to do it on a large enough scale for whatever reason, you can still try to find 5090s for <=4.2k USD or 4090s for <= 2.1k USD on local marketplaces and then resell on the same platforms (make sure to ask if you can benchmark on site before buying, don't get scammed). It'll only be a few hundred in profit per unit which is substantially less but wtv. Even then, good prices on individual gpus are really rare now tbh, the best time to have been doing this was no later than a few months ago, so parting out second hand high end rigs is a lot more viable, it just takes more work. You'd have a massive buffer given that the price of memory is not going down for years & smth like the astral will keep appreciating until like 5.5-6k USD per unit by eoy, unless Iran nukes every datacenter on the planet so it would take insane skill to fuck up somehow

 

[realboyo]

How much is that all worth, bro?

<350,000 CAD, obtained it all for ~200,000 7-8 months ago except for rog matrix & msi lightning which was recent

 

[cargo70]

<350,000 CAD, obtained it all for ~200,000 7-8 months ago except for rog matrix & msi lightning which was recent

Any plan with it?

 

[realboyo]

Thoughts on palantir level 3D facial morphometry. Facial analysis apps have been around but I could probably run a heavy server-side pipeline with a big enough GPU cluster, and do geometric fusion (photogrammetry) across hundreds-thousands of frames of the same person (with different angles and lighting).

It would have an accuracy of sub 0.1mmbetween each point (ig vertice since it's 3d not 2d) rather than relying on MediaPipe which is single-frame and 2D with very limited points. It's the reason why faceiqlabs is so shit they use MediaPipe take one 2D photo and analyze it with 5.0mm spatial error

With 5090s you could reasonably do an analysis of 200,000+ vertices plotted at once, by combining a shit ton of frames taken via a video or multiple photos. Via this you can move your face around over multiple frames for literally thousands of times more data, and it'd be 1 trillion times more accurate than the current apps that use MediaPipe and analyze a singular stationary photo (afaik the current apps just do one front one side and analyze separately).

Because MediaPipe doesn't have that many points It interpolates large distances between the points/landmarks (e.g., cheekbones, jaw contours, temple region). It cannot measure micro-features like nostril symmetry, lip volume (or any volume accurately for that matter), or subq fat distribution.

By using like 100,000-200,000 vertices on your face from multiple frames instead of 500 points on a single frame, it would provide a continuous surface topology instead of sparse dots where shits so inaccurate u have to manually landmark.

This would be able to compute volumetric measurements like volume of the chin projection in cm^3 given it would it have sub1/10mm accuracy on the contours of the face rather than just 2D pixel distance that are 5mm apart. Morphs would be so fucking sick holy

However, because processing a few hundred frames at 4K will take like 3 mins using 5090 rigs which eat up a lot of power (~750W), just one scan will cost 1-2 cents in electricity alone.

Ad networks (AdSense, Unity Ads) pay an average CPM of $0.0015 to $0.004 per ad watched

So it'd be like 3-6 ads watched per scan to break even

Would y'all watch 2 minutes of ads for israeli military grade facial analysis

 

[realboyo]

Any plan with it?

Want to do some cool projects before I sell them next year hence this post, they're already opened so I might as well otherwise it'd be a waste