An interesting development, but it seems to be focused exclusively on parallel compute (enterprise dGPUs use cases):

The Austin, Texas-based AI chip startup says it’s developing an optical processing unit (OPU) that in theory is capable of delivering 470 petaFLOPS of FP4 / INT4 compute — about 10x that of Nvidia’s newly unveiled Rubin GPUs — while using roughly the same amount of power.

From my limited understanding for CPUs (which are arguably far more complex and less “predictable”), Moore’s Law is definitely dead.

If you look at single-thread CPU performance, gains from say ~2013 (Haswell/Ivy Bridge) are relatively modest compared to modern ~2025 era top end CPUs (9800X3D). Just compare a late 486, say the i486DX2 from 1994 to a P3/Tualatin from ~2001, there is no comparison at all.

Fascinating stuff. It seems like the dead internet theory is coming to life.

Good point, TSMC is not just the pick axe seller in the gold rush, they are a generalized “best in class” tools seller.

To some degree, I don’t think it matter for them what they are baffing, they’ll always have demand as long as they are the clear leader.

Good stuff! )

I love Sweeney’s holier than thou attitude. MFer, you don’t give a shit about anything other than self-enrichment and social standing among other oligarchs.

It is worth noting that the benchmarked Snapdragon X2 Plus ran on a reference platform, while testers used commercially available products for the other chips. This is a key caveat, as results can vary widely depending on chip binning, cooling, power limits, SSD speed, memory latency, and installed apps.

What this means is that performance with real devices will likely be even worse.

With the X Elite we also had benchmark results on “reference platforms” that were never hit on real devices.

I thought this was the case already.

I am a Fractal Design person when it comes to cases, but these look really nice.

It’s too bad it will be impossible to build an new PC in the next 24-36 months.

Another propriety tech…

HyperNormalisation has its flaws (too big of a focus on a single “theory of everything”) and an argument can be made that it emphasizes style over substance, but it’s a great experience. Even if I disagree with some of the structural arguments of HyperNormalisation, it does raise a lot of good points and in a very engaging way.

I would also recommend Bitter Lake from the same director which IMO is a more intense experience.

Thank you!

I did check the site, but I missed it.

I wish there was a text transcript of this, I will check the video out after the holidays though.

I have my doubts about their benchmarks with respect to real world use cases. I have a 5800X that scores about 2,200 on GB6 ST. So does the 4,000 score mean it will be roughly x2 performant for constant throughput single thread use cases?

I use a Debian stable ARM distro called DietPi; it’s perfect for aow touch DIY NAS/media server/Pi-Hole.

You get all the stability and predictability of Debian with a nice set of ClI tools and configuration utilities from DietPi team.

I’ve been using them for 8 years (been donating too), other than an issue with a major Debian upgrade, I’ve had no issues.

Hope they get to 10K.

I am on Samsung/Android, but I will not be paying for American platforms after this (my A73 works really well considering considering that I have had nearly 3 years). To be honest, I have zero desire to upgrade.

I hope Jolla can develop a sustainable consumer business that’s fundamentally different from the US/Chinese model.

I hope they are able to get 25K pre-orders and expand their geographic sales coverage after that.

Removed by mod

Don’t take shitposts so seriously!

A tankie is a supporter of authoritarian state capitalism and the genocidal imperialist policies of such regime that follow such a model. Nothing more, nothing less. )

You don’t need think hard about it.

At any rate whatever your trying to say is not really in context of my initial shitpost.

I don’t know, I am guessing not.