Xerox had (maybe still has?) some printers with 5 or 6 color toner capacity. CMYK plus you could order special color toner in stock or fully custom mixes (minimum order sizes apply for full custom) but it was great for companies who had logos which could not be exactly represented by CMYK half toning as the spot color toner could be their exact logo color.
I’m sure the same kind of thing would be possible using Prusa’s documented methods with a little extra work.
Q8 quant is very minimal fall off in terms of KLD against the lab 16 bit. If you have the memory for BF16 KV-cache (which is usually easier to stomach) then the Q8 is very close. But even Q8 quant model with Q8 KV-cache is very close.
Smaller quants for the model start to fall off but more importantly, smaller KV-cache quants fall off much faster so avoid less than Q8 there.
Would love to see actual security focused hardware/software features, like full OP-TEE, fTPM (or a more ideally a real physical TPM), and similar. For example, so that the OTP isn't the only way to store a disk encryption unlock key.
The existing secure boot mechanisms aren't bad, but allowing for more than one public key hash in OTP would be nice, too.
These kinds of things are expected to be on modern embedded SOCs and SOMs now.
You do know those are trivially bypassed with a signal processor, right? If physical access is outside your threat model, that's OK, but it makes (for example) the forced Win11 upgrade for DRM^H^H^H boot integrity enforcement seem ridiculous.
Yeah, fair enough. "Compliance" is probably the phrasing I should've used, rather than "security".
I've been curious for a while about the overall taxonomy of security, especially for embedded platforms. It seems like the only hope is defense in depth, given the power glitching attacks and the like that you can find demonstrated.
Specific to the Raspberry Pi, I believe I even saw a thread at some point where one of their firmware engineers was making the case that secure boot on the Pi 5 was equivalent to a TPM in almost any reasonable threat model, since, in either case, you were out of luck if an attacker had physical access and was willing to put in enough effort.
Normal secure boot does not use the TPM. Secure boot is the proactive process of ensuring only allowed code loads and executes.
The TPM is used for measured boot, the post process to understand what actually was booted and if the right set of things were booted then to allow unlocking of specific items like keys.
Both are important but they are not the same thing.
It's a pretty normal thing to do for small LCD screens. Linux has had SPI framebuffer support via fbtft subsystem (in staging tree now, previously was out of tree) for well over a decade. It works quite well.
Many silicon vendors, when providing said binary blobs to a device OEM or even just documentation or source code for the binary blobs, will make companies agree to a license or other legal terms which prohibits reverse engineering. Often the direct recipient of the binary blobs (the OEM of the device) cannot legally let their employees nor contractors perform the reverse engineering.
Generally, unless a similar license or legal terms are required to be agreed to by the end user, nothing stops the end user from reversing said binary blobs. But before you attempt this, be sure you fully understand every legal document which was presented to you by the device vendor. Click-through EULAs included.
There’s lies, damn lies, and lies that disks tell the operating system. Don’t believe any of them!
If you need to know it’s been persisted to non-volatile storage then you need to own the full stack of every piece of software between the OS and the actual physical memory.
Every managed flash drive is going to have layers and layers of complexity and caching and things you simply can’t easily control or really understand. Don’t trust it unless you know exactly how it works all the way down.
Well said and there are some bitter lessons in the storage industry.
In my last company we need to disable the disk write cache during each reboot, and we also heard a lot industry stories related to underneath firmware implementation from oxide computer podcasts [1]. Yes, to provide truly reliable service, we need to evaluate underneath hardware settings case-by-case.
I frequently see comments which would have made sense in the past (e.g. early 2000th) but kinda aren't fully reflecting reality anymore
it's as if humans have a tendency to make up their mind/world view in their younger years and then tend to kinda stick with it/only change it slowly as long as no big live changing events happen
I give you $100 cash and you give me $100 worth of stock in return. Now you give me $100 cash to buy something from me that cost me $80 to produce. I end up with $100 worth of stock in your company which cost me only $80. No?
NVIDIA gross margins lately are like 75%, so it's more like you give me $100 to buy something from me that cost me $25 to produce, hence I end up with $100 worth of stock in your company and it only cost me $25.
> hence I end up with $100 worth of stock in your company and it only cost me $25.
You also lost out on $75 worth of cash revenue (opportunity cost from selling the same thing to a different customer), so really you just took stock in lieu of cash.
It'd be different if Nvidia (TSMC) had excess production capacity, but afaik they're capped out.
So it's really just whether they'd be selling them to OpenAI and getting equity in return or selling to customers and getting cash in return.
If OpenAI thinks their own stock is valued above fundamentals, it's a no brainer to try and buy Nvidia hardware with stock.
Sure, but OpenAI doesn't have cash. It does have stock.
Even if Nvidia has capped production for now, increased demand still allows them to sell chips at a greater margin. Or, to put another way, presumably Nvidia is charging OpenAI a premium for the privilege of paying with stock.
In that case, you spent $80 to produce an item and exchanged it for $100 worth of their stock.
Now if you check, these companies selling their stock like this tend to have large amounts of debt. If their stock becomes worthless, you just wasted $80 producing an item that their creditors have first dibs on. And liquidating your shares immediately to ensure your gain, would weigh on their stock's value, potentially to the point where their stock would be only $80 worth, and you wouldn't be gaining anything anymore. Your earnings would then tank, alongside them.
> I give you $100 cash and you give me $100 worth of stock in return. Now you give me $100 cash to buy something from me that cost me $80 to produce. I end up with $100 worth of stock in your company which cost me only $80. No?
Sure, but how's that a cheat code? If you normally sell something for $100 that costs $80 to make, and then use that $100 revenue to buy $100 of stock, this is an identical outcome for you.
If they couldn't borrow $100, or get $100 from any other investor, that just puts you in the position of being an investor, and even then the difference between bradfa's version and mine is simply when you became an investor, not that you became one.
Again, this is not a cheat code: if you sell $80 of cost for $100 of stock, the stock you now own can go up or down, and if you overvalued it then down is the more likely direction.
The primary cheat code here would actually seem to be (a) getting preferential access to Nvidia's production through these deals and (b) creating a paper story of increasing OpenAI private valuation.
Aaaannd get to claim the 100 as revenue to show investors that the company is performing better than if I had not made the deal, which also means that demand for the product stays inflated which also means I can keep my margins higher by not needing to discount my product.
Urgently need an IPO so losers can chip in. If the sandcastle plummets before, funds and other AI companies lose a lot, so better bet again and again, even if this is nonsensical.
I have a pair of Freestyle2 keyboards, both are over a decade old. I strongly recommend the V3 tenting kit. You can get a refurb USB Freestyle2 with the V3 kit for $70 direct from Kinesis.
I’m sure the same kind of thing would be possible using Prusa’s documented methods with a little extra work.
reply