Hacker News

A DGX B200 costs like ~$0.5 M and uses around 14 kW.

If you plan to run it straight for 8 years 100% max usage thats around 1 GWhr.

A gigawatt hour is a lot of energy but its not that much compared to the price of the actual machine. In Germany for example with its expensive energy thats about €100k worth, which spread over 8 years is pretty minor compared to the up front half mill.

The real issue with high power consumption is not really the cost of energy but the limited powersupply you can get for a datacenter. A more efficient setup is highly desirable because it means you can fit more in the limited power hookup.

> I have never seen a company use AMD outside of wafer and a couple others mostly in US.

There's a few using them, and even more starting to experiment with them. AMD has long been a source of disappointment around this side of things, so I'm hesitant to feel optimistic we'll finally get some competition. The market really needs viable competition to Nvidia, especially performance/watt.

> I have never seen a company use AMD

Meta is using AMD: https://www.amd.com/en/newsroom/press-releases/2026-2-24-amd...

And OpenAI: https://www.amd.com/en/newsroom/press-releases/2025-10-6-amd...

> I have never seen a company use AMD outside of wafer and a couple others mostly in US.

Worth remembering AMD basically "owns" (not literally) the hardware-side of things in video games consoles for good many years now, with no end in sight.

Typically any company that can’t get Nvidia to fill their orders will have at least some AMD.

> I have never seen a company use AMD outside of wafer and a couple others mostly in US.

Just because you haven't seen it doesn't mean it doesn't exist.

We've serviced over 700 customers on our MI300x.

AMD MI355X uses 1,400W per GPU and NVIDIA B200 uses 1,200W. So AMD uses about 16% more power.

Kimi uses INT4 as its native format, there's no such thing as "better than 4-bit precision" for that model. This is in contrast with GLM for which 16-bit precision is native and 8-bit is in common use.

MI355X can perform FP6 operations with the same speed as their FP4 (unique to AMD) - people should be making MXFP6 quants which would be pretty much lossless, and much closer to FP4 performance than FP8

Doesn't Nvidia with their NVFP4 claim that it's lossless?

I haven't tested enough models Nvidia has converted to NVFP4 besides GLM 5.2 but it seemed fine to me.

My own luck has been hit or miss with it.

First thing I noticed as well

from memory, it is like 96-98% of the accuracy.

Its MXFP4

And to use the heading "Why this matters".

A nice filter is checking for the `.ai` in the end. It is very likely slop if you see that. Slop meaning low-effort/clickbait/shallow/useless/scam etc.

Wafer discontinued their own "Wafer Pass" flagship coding plan within weeks of launch and had to issue prorated refunds. Now they're bragging about squeezing costs down even further via quantization, even though their implementation is clearly lacking.

[1] https://www.ycombinator.com/launches/Q9i-wafer-pass-flat-rat...

And somehow they claimed that it is "lossless".

It's very unclear what's special in Rubin to be optimized for inference? I can see disaggregated bit (with having separate prefill and decoding nodes), but what else?

how do you get 5x faster at inference when inference is memory bandwidth limited? getting 5x the memory bandwidth of a h100 seems physically difficult.

yes it is 213 tok/s single stream (so per user)

Agreed. The writer is pretty loose with their comparisons:

* What does it mean for "performance per dollar" to get faster? Higher, maybe; rise faster than it has in the past, maybe, but just "faster"? Nope.

* The article cites some equipment as being "2x cheaper". I think they mean "half the cost", but if so they should say it.

hi i work at wafer. no the margins are lower averaging at about ~40%. utilization is one of the highest order bits in determining margins here, yes.

A place where the slope/derivative/incremental-performance-per-price changes.

I used to be high-performance like you, then I took an arrow to the knee?

They seem to be serving it at 3x the price while also struggling with maintaining uptime on openrouter; while the vercel router advertizes even bigger speeds but has no clear uptime stats

I guess you really do have to try it at least for some time to actually know

You got it backwards; it's ~200 on single stream so the 2,600 is achieved with ~13 streams.

hi yes it’s not optimized for single stream it’s optimized for total node throughput

There are a limited number of these available in comparison to demand. I think people figured out that LLMs and VLMs can do real work that can replace a lot of humans. And for plenty of jobs, it's good enough to reduce already outsourced staff by 75-90% at a fraction of the cost.

Because lots of people are willing to pay more dollar for smarter token.

Because they are dumping/subsidizing it token processing to try and get companies to fire as many people as possible. So they'll be dependent upon the companies when they have to Jack the rates

this is exactly our thesis at wafer :) thank you for the support

Personally, I can't wait till something like this starts getting to consumer level. https://www.anuragk.com/blog/posts/Taalas.html