> So, it's more the former. The testing machine is probably a 8 year old server on an office shelf
I doubt this very much.
The load testing env is the Office 365 load testing environment. I'd be surprised if it is less than 10,000 cores, and the machines are probably good but older; possibly 50 core Xeons instead of 128 core EPYC or something.
It seems likely that the test env was setup to mirror one of the smaller 365 regions a few years ago and the only updates were when machines were replaced.
From the report: "lock contention inside an internal process... The contention issue is worsened on machines with more cores and newer hardware; whereas our load testing tended to focus on our weaker hardware SKUs with fewer cores."
Sure, but the point was about how everyone complains about the original Pascal, and the related dialects to fix its shortcomings, while ignoring that shortly thereafter its creator solved all of them, by designed another programming language actually targeted to large scale software development and systems programming from the start.
Given Niklaus Wirth pursue for minimalist systems programming design since Oberon, I doubt he was a big fan of Ada83, let alone Ada2022.
In the 80s Pascal was a serious competitor to C for systems programming as well as applications. Large amounts of the original Mac Finder were in Pascal for example.
The lack of a good web platform has been mentioned but also there wasn't really any Unix heritage and on Windows the .NET platform and C# killed the advantages Delphi had over Visual Basic 6.
I was a Delphi developer for a bit and most of us moved to Java if we wanted to stay if Microsoft platforms.
> Amazon, Google, Microsoft, and the other big hyperscalers are making their own AI chips to move away from Nvidia. The startups like OpenAI are doing the same
This is true
> so the real demand for Nvidia chips is dropping.
This isn't true.
The hyperscalers are still buying as many Nvidia chips as they can. They are limited by supply, not demand.
Nvidia is carefully making sure they have as many buyers as possible. This is excellent business strategy.
There is a reasonable debate on to what level they should do vendor financing for this. But the general idea of trying to make sure they have as many buyers as possible is very sensible.
If the other front-age article on the Science publication about how human population will have overwhelmed the earth by now isn't enough warning, then I'm old enough to remember how the "Japanese economic miracle" was going to inevitably overwhelm the US.
There is a serious lack of wisdom in lots of these statements.
I like to remind people that multiple mainstream "AI Safety" campaigners have called for first strikes on Chinese data centers:
In a new report titled “Superpowers and AGI,” former Obama-era national security official Jacob Stokes calls for United States to begin preparing readiness drills in response to Chinese AI development — up to and including military strikes on the nation’s data centers, the South China Morning Post reported.
“Data centers can be bombed with conventional munitions,” Stokes writes. (He’s not the first thinker to suggest bombings to stop powerful AI; in 2023, AI critic Eliezer Yudkowsky suggested that countries needed to be ready to “destroy a rogue data center by airstrike.”)
> I'm old enough to remember how the "Japanese economic miracle" was going to inevitably overwhelm the US
I'm also old enough to remember that in the 1980s Japan had become the world's second-largest economy and there were serious concerns in Washington that Japan might eventually surpass the United States economically. The Reagan administration began putting considerable economic leverage and measures to create pressures intended to destabilize Japan's economy. There's historical agreement on this intent. See 1985 Plaza Accord to inflate the Japanese yen. See 1986 US-Japan Semiconductor Agreement to restrict Japan's dominance in DRAM chips.
> In a new report titled “Superpowers and AGI,” former Obama-era national security official Jacob Stokes calls for United States to begin preparing readiness drills in response to Chinese AI development — up to and including military strikes on the nation’s data centers, the South China Morning Post reported.
War Hawks gotta war hawk.
Ever since the fall of the USSR, there has been a faction of people in the USA desperately trying to invent a new "enemy" to scare everyone and drive foreign policy.
That's all well and good, until the theft and spread of knowledge leaves you permanently unemployed.
Because agents can do labor this is more than just "knowledge sharing". It undercuts human labor power and allows for even grosser concentration of capital.
No, they are locking it behind their corporate walls.
Sure, right now they are offering some of it for "free" (give us all your data) or "cheap" (give us all your data and pay some money too) but what happens when all other options are unavailable or difficult to access for most people?
Suddenly it is the corporations that decide what is available, for how much, and what is right and what is wrong.
They have shown time and again they will bait and switch at will, this is the ultimate run on everything - turn everyone who can think for themselves into zombies and then force them to use technology for everything and charge for it either directly or via endless adds or some mix of the two.
Trying to centralize all information and "intelligence" in one or two providers, charge for "intelligence" "like a utility", and divest citizens of their own skills and recourse to resources is not "expanding access to the commons" lol.
There are several papers about "deskilling" and cognitive debt as a result of heavy LLM reliance. Ask your agent to put together a reading list. (If you ask your agent to read the papers for you, there you go, you are an existence proof of the deskilling).
I’m not sure that Hardware has the same moat as Service/Platform.
If (it’s a big if), NVDA get leapfrogged in hardware capability it will be trivial for the next generation of deployments to stop using them entirely (Google and Microsoft are already developing custom inference chips, as are others)
I don't think this is true. AMD's hardware, on paper at least, offers the same or better performance than Nvidia. Eg the Mi355x vs B300 is close. The former released before Nvidia's GPU.
The same story goes back for years and years in consumer hardware too. AMD GPUs often had more raw compute capability, but fell short in actual game performance (or other applications).
Nvidia's moat is, at least partially, the software to make the hardware useful.
Nvidia is king of circular financing. peak accrual earnings include non-cash items and demand Nvidia is funding itself. On a cash (real actual reality) basis the multiple is 33 to 50
I doubt this very much.
The load testing env is the Office 365 load testing environment. I'd be surprised if it is less than 10,000 cores, and the machines are probably good but older; possibly 50 core Xeons instead of 128 core EPYC or something.
It seems likely that the test env was setup to mirror one of the smaller 365 regions a few years ago and the only updates were when machines were replaced.
reply