• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Microsoft to Unveil Custom AI Chips to Fight NVIDIA's Monopoly

easy part of code could be written by AI and developers work less.
As a developer, no, I would not trust or accept code an AI put out. I've seen it and it's borderline useless.
 
I feel big companies will push for custom silicon.

It is sad really, we have better technologies yet we still waste more than 40h a week working instead of living. AI is another missed opportunity, it can really speed up a lot of things, e.g. easy part of code could be written by AI and developers work less.
Work always fills to expand the available time, AI will just help you deliver more in those 40h. Productivity increased constantly over the past decades. The computer does a lot of the work an office worker would have done by hand in the past. Are you working less than that office worker from 30-40 years ago, or just filling your 40h with different tasks?

Microsoft hopes to match or beat the performance of NVIDIA's offerings and reduce the cost of AI infrastructure
I have no doubt that it would be cheaper for MS. Nvidia prices their offering at one order of magnitude over cost, while MS plans on building these chips for themselves. But I'm really wondering what are the odds they get even close to matching Nvidia's performance, not even saying to beat it, from their first generation of the AI chip. MS can probably optimize better for their particular types of workload, which squeezes some efficiency from the chip but that's still a lot of years of experience to catch up with.
 
I have no doubt that it would be cheaper for MS. Nvidia prices their offering at one order of magnitude over cost, while MS plans on building these chips for themselves. But I'm really wondering what are the odds they get even close to matching Nvidia's performance, not even saying to beat it, from their first generation of the AI chip. MS can probably optimize better for their particular types of workload, which squeezes some efficiency from the chip but that's still a lot of years of experience to catch up with.
Dude, general purpose gpu that are lightly tuned for AI is not great. The crux is that currently that's the best that there is. A focused chip design like Tesla's D1 is equivalent at 1/6th the cost, not to mention the reduction in footprint and power consumption. MS could totally match what Tesla has done.
 
This has to be the most uninformed post I have ever seen in my life.

Literally moved to live on their campus because I hate commuting.

24/7/365.

But what do I know.

A review of ~12 months:

Paid premium for video gaming IP. Argued before the FTC that industry consolidation was required due to the strength of its main competitor. Has an inferior console when compared to its competitor both in technical specifications and the fact that racism/other similar statements are extremely prevalent in Xbox Live communications.

Cancelled HoloLens. Share buyback was priority.

Removed the ability to uninstall its browser, pushed updates which changed privacy and notification defaults for its users. The public facing justification is an outright lie; “our default web browser is an essential component of our operating system and can't be uninstalled.”

Users must edit the registry or use tertiary applications to remove this bloatware, but once complete, the system runs just fine.

Delivered possibly the worst search engine ever in the past, and has given the market 0 reason besides force to use it the past year. I don’t know if you see the trend here but creating compelling products is not one. Borderline fraud when comparing initial company claims to real life results. Similar to Apple Maps initial rollout, except they iterated much more aggressively, and ultimately created a tool that was better than the current market offerings. Once stable, it was substantially faster than Googles previous offerings and infinitely more convenient than any legacy web or gps hardware products.

Currently arguing with the FTC about Googles “monopoly on advertising”, and is “trapped in a vicious cycle”. (It’s called make something better.)

Historically, has entered and abandoned the music hardware, music marketplace, mobile phone, ~mobile phone marketplace, social media, and wearable markets. Trillions left behind. All while claiming to have a superior product. All failures.

It has made ~60 acquisitions in the past 5 years and has a quarter million mostly world class employees including some living legends. The problem is the company’s decisions are made, essentially, by the collective will of investment managers, not engineers. Very similar to Boeing, in the bad sort of way.

You give me that engineering power and I will rip a hole in space time itself.

It’s okay that Microsoft is a value destroying company. They own two of the most critical pieces of software used worldwide. But if one is not skeptical that some <initial hardware rollout> from Microsoft is going to be superior to an entrenched incumbent, who is actively iterating, then that’s just silly imo.
 
Last edited:
Work always fills to expand the available time, AI will just help you deliver more in those 40h. Productivity increased constantly over the past decades. The computer does a lot of the work an office worker would have done by hand in the past. Are you working less than that office worker from 30-40 years ago, or just filling your 40h with different tasks?
Indeed, in less developed countries, 40h is the norm or worse in even less developed countries, because both the leaders and the bosses don't understand basic capitalism let alone economics.

In more civilised countries workers work less than 35h with multiple weeks of vacation.

Maybe we need another Russian Revolution to improve working conditions, but who knows the Americans might surprise us.
 
Ai is unfortunately not going away, because it is the main engine powering ads and ad delivery mediums (social media, streaming, VR, etc). This is the core reason Ai exists, any other justification for Ai existence is so irrelevant that it is borderline misinformation.
One can only hope that GPU's become obsolete for for Ai much like what happened to them when Bitcoin ASICS came to market and GPU Bitcoin farming died.

This is probably what will end up happening, if I was to bet.

GPUs are not the greatest fit for this, and this is something, unlike mining, that is going to have a pretty massive market for it. Plus having dedicated hardware makes it easier to control...
 
Google already has been doing things for years now. They were the first.
Yea, but they move at a glaciers pace and aren't bringing it into production until 2025 whereas Tesla has had their D1 in service for months and is on the ramp up phase. There's a difference between talking about doing something and actually doing it. In the meantime they keep buying gpus.

Ex. this just out later in the day. This is called Tesla speed. Dojo was announced in 2018, completed in 2021, into production in 2022, and now ramping. It's been handling auto labeling and vision training since 2022. Now they're building a central location for it...

At Tesla’s headquarters in Austin, Texas, a new bunkerlike structure is under construction that could one day help move the company beyond electric vehicle manufacturing.


When it’s completed, the building will house part of a new supercomputer, known as Dojo, that Tesla is assembling to help run the artificial intelligence software behind the self-driving capabilities in its vehicles, according to two people familiar with the project. Eventually, Tesla could use Dojo to sell cloud services to other companies, the way Amazon Web Services does...

 
Last edited:
Back
Top