"Nvidia is so far ahead that all the 4090s are nerfed to half speed" • Hacker News Board

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#91

Post by dredmorbius » Mon, Dec 16, 2024, 9:34 PM UTC

unethical_ban wrote: post #40

It's interesting how we the people (broadly) accept this practice in software and even some hardware, but not in other areas. Note how frustrated people are when you hear about "unlocking" sensors and services available on cars. If a product is made, and the cost to provide that product is the same one way or the other but you cripple it to create segmentation, then that is greed. Period. Objectively. And if you're o…

Aside from possible technical explanations (e.g., the binning of products based on defects which permit sub-optimal performance as creato describes: https://news.ycombinator.com/item?id=42435397>), there's market segmentation.

French polymath (economist, engineer, bureaucrat) Jules Dupuit famously described this concerning railway carriage accomodations and the parlous state of third-class carriages:

It is not because of the several thousand francs which they would have to spend to cover the third class wagons or to upholster the benches. ... [I]t would happily sacrifice this [expense] for the sake of its popularity.

Its goal is to stop the traveler who can pay for the second class trip from going third class. It hurts the poor not because it wants them to personally suffer, but to scare the rich.

https://www.inc.com/bill-murphy-jr/why-does-air-travel-suck-...>

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#92

Post by Retric » Mon, Dec 16, 2024, 9:36 PM UTC

akoboldfrying wrote: post #89

Xymist wrote: post #36

Earlier quoted context omitted.

They just shouldn't do that. If they can afford to sell the Ferrari for $20k, they should do so. To everyone who wants one, for whatever reason.

I disagree, and this line of thinking is positively dangerous. Just because Ferrari might be capable of making that car for $20k, I don't have a fundamental right to demand it from them any more than I have a fundamental right to demand that you make me a sandwich right now for $5. > they can afford Before using the word "they" in a prescriptive sentence, think about whether you could substitute "I" and you would sti…

The goal isn’t to directly force them, but to create a market competitive enough that the only way to compete is to sell the best product they can with a minimal markup.

I have no issue selling into a competitive market, that’s just how things work for individuals. It’s only at the scale of countries and giant companies that the ability for anti competitive behavior really shows up.

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#93

Post by modeless » Mon, Dec 16, 2024, 9:38 PM UTC

akoboldfrying wrote: post #89

Xymist wrote: post #36

Earlier quoted context omitted.

They just shouldn't do that. If they can afford to sell the Ferrari for $20k, they should do so. To everyone who wants one, for whatever reason.

I disagree, and this line of thinking is positively dangerous. Just because Ferrari might be capable of making that car for $20k, I don't have a fundamental right to demand it from them any more than I have a fundamental right to demand that you make me a sandwich right now for $5. > they can afford Before using the word "they" in a prescriptive sentence, think about whether you could substitute "I" and you would sti…

There's no fundamental right. But wishing for competition is certainly reasonable! We should all be rooting for competition to improve the efficiency of our markets.

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#94

Post by tgsovlerkhgsel » Mon, Dec 16, 2024, 9:40 PM UTC

cosmic_cheese wrote: post #45

Between EVGA getting out of the Nvidia card business, Nvidia continuing to be problematic under Linux (even if that’s improving), all the nonsense with the new power connector, and the company’s general sliminess, I’m increasingly leaning towards an AMD (or potentially Intel) card for my next tower upgrade. AMD and Intel might only be competing in the entry-to-midrange market sector but my needs aren’t likely to exce…

AMD has inexplicably decided not to invest in software. Just like car manufacturers don't realize that a shitty infotainment system can keep people from buying the $100k car, AMD doesn't seem to realize that people aren't buying a GPU for ML if their ML framework doesn't run on it...

And this goes down to consumer drivers too. I've sworn to myself that I'm not buying AMD for my next laptop, after endless instability issues with the graphics driver. I don't care how great and cheap and performant and whatever it is when I'm afraid to open Google Maps because it might kernel panic my machine.

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#95

Post by akira2501 » Mon, Dec 16, 2024, 9:41 PM UTC

mensetmanusman wrote: post #70

akira2501 wrote: post #65

Earlier quoted context omitted.

Interesting that it's entirely failed to do that so far. Nvidia is helping power the next generation of big brother government programs.

Aren’t we like 0.000001% in the journey?

That is the largest goalpost move I've ever seen, but sure, at that order of magnitude anything is plausible.

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#96

Post by bri3d » Mon, Dec 16, 2024, 9:45 PM UTC

sabareesh wrote: post #28

MPSimmons wrote: post #23

Earlier quoted context omitted.

e-fuses ( https://en.wikipedia.org/wiki/EFuse ) are typically etched into the silicon as an exactly-once operation, meant to irrevocably set a configuration. Some devices, for instance, have an e-fuse that makes it impossible to change cryptographic trust signatures after the e-fuse has been blown.

That's intriguing; one might assume that a key feature of eFuse would be the ability to reset easily. But I guess it could be implemented without it

There are two kinds of thing that are both called eFuse: current-limiting devices which are almost always resettable, and one-time programmed bits which are intentionally one-time programmable (some are implemented as physical fuses which are blown by overcurrent, others are implemented using various other types of write-once nonvolatile memory, and some bad ones are implemented by making normal flash pretend to be write-once using firmware-level protections).

Professionally I usually see OTP referred to as "fuses," "OTP," "straps," or "chicken bits," with the specific word "eFuse" reserved for the current-limiting device. But in popular media the trend seems the opposite.

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#97

Post by TiredOfLife » Mon, Dec 16, 2024, 9:45 PM UTC

cosmic_cheese wrote: post #45

Between EVGA getting out of the Nvidia card business, Nvidia continuing to be problematic under Linux (even if that’s improving), all the nonsense with the new power connector, and the company’s general sliminess, I’m increasingly leaning towards an AMD (or potentially Intel) card for my next tower upgrade. AMD and Intel might only be competing in the entry-to-midrange market sector but my needs aren’t likely to exce…

EVGA also significantly reduced warranty on their PSUs. Changed PSU components without changing model number.

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#98

Post by hamandcheese » Mon, Dec 16, 2024, 9:46 PM UTC

zeusk wrote: post #56

itsthecourier wrote: post #33

Earlier quoted context omitted.

I want to add nvidia sales are 90% data center and 10% gaming, and the author being part of the 10% who wasn't abandoned is complaining they got a product at half the speed, way lower price, instead of half the price same specs as the datacenter client man.

Or the 90% are charged absurd markup because clearly they can deliver the hardware for 10% use-case for 1000$ and still make money on top but they would rather charge the data centers 50k for the same product

There are companies lined up around the block to hand over $$$$$ for the only competitive GPUs on the market. Is the markup actually absurd?

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#99

Post by YetAnotherNick » Mon, Dec 16, 2024, 9:47 PM UTC

david-gpu wrote: post #46

YetAnotherNick wrote: post #30

Earlier quoted context omitted.

NVidia has software moat for specialized applications but not for AI, which is responsible for most of their sales now. Almost everyone in AI uses pytorch/jax/triton/flash attention and not CUDA directly. And if Google can support pytorch for their TPU and Apple for their M1 GPU, surely others could.

> NVidia has software moat for specialized applications but not for AI, which is responsible for most of their sales now. Almost everyone in AI uses pytorch/jax/triton/flash attention and not CUDA directly And what does pytorch et al. use under the hood? cuBLAS and cuDNN, proprietary libraries written by NVidia. That is where most of the heavy lifting is done. If you think that replicating the functionality and perfo…

Did you read the last part? Pytorch uses drivers, and drivers exists for Google's TPU and Apple's M1 GPU as well and both works pretty well. I have tested both and it reaches similar MFU as Nvidia.

Re: "Nvidia is so far ahead that all the 4090s are nerfed to half speed"

#100

Post by gary_0 » Mon, Dec 16, 2024, 9:47 PM UTC

michaelt wrote: post #69

chrsw wrote: post #10

Earlier quoted context omitted.

Intel didn't have a software stack moat.

In 2000 Intel had a huge software moat: Microsoft Windows, and the large install base of x86-only software. Rich webapps hadn't been invented. Smartphones? If you're lucky your flip phone might have a colour screen. If you've got money to burn, you can insert a PCMCIA card into your Compaq iPAQ and try out this new "802.11b" thing. Java was... being Java. Almost all the software out there - especially if it had a GUI…

So many devs are too young to remember a time before you would expect to just download some open source and compile it for x86/amd64/arm/emscripten/etc and be good to go. In the old days, if you didn't want to write that library code yourself, chances are all your AltaVista search would turn up was a guy selling a header file and a DLL and OCX[0] for $25. If you were lucky!

A vast amount of code was only intended to compile and run on a single OS and architecture (circa 2000, that was usually x86 Win32; Unix was dying and Wintel had taken over the world). If some code needed to be ported to another platform, it was as good as a from-scratch re-write.

[0] in case you wanted to use the thing in Visual Basic, which you very well might.