www.bodybuildingmealplan.com/leg-press-vs...

But it's still crazy to me that one single machine can basically do everything in the lower body.

It doesn't fully replace free weight exercises of course, they're more intensive, involve stabilization muscles, and have bigger gains for some things: I found an article comparing a squats and leg press and it's interesting how it apparently produces a lot more growth hormones and a bit more T?

I should get into the game again

I'd say that's still an achievement. Because for how long did CEF have an arm build? Windows on ARM is not a new thing.

Google barely made an ARM build of Chrome last year, even though Chromium Edge has been ARM native since a month after its launch in 2020

So a laptop without the 200Gbps nic and only 64GB of RAM could maybe start at $2.5K?

Or it could all just go down the drain since Nvidia said they're gonna make GPU chiplets for Intel CPUs.

Nvidia's N1X SoC next year could probably also push stuff forward... If its priced right

It'll be based on the GB10 which is found in expensive DGX Spark systems right now ($4K USD for 4TB? $3.5K for 1TB? But all configs have 128GB RAM), but an additional cost is probably that they have 200Gbps NIC

It's surprisingly moving forward. Qualcomm's hype build up and use of influencers along with Microsoft forcing ARM64 on consumers with Surfaces I guess worked out because even like Discord has an arm build now (though it's still not the first thing you see on the website) and EAC works on ARM now

The odd part is that Gaussian Blur is a whole separate subtest and that is actually where the 8 Elite has the biggest lead so they did optimize dot product functions a bit maybe? Unless the driver always does a hack for higher performance in doing these only.

So the GPU has a hard time elsewhere?

But going through Wikipedia about this makes me feel like wow. One class covered this, but I don't understand it at all.

The first step involves computing a gaussian filter which itself involves doing a convolution: the integral of a dot product of two functions?

And then other steps linked in the article.

Their document on the tests just say it uses the Canny edge detector algorithm.

According to Wikipedia it seems to not be all that complicated? This is like Calculus 3 level of concepts. Except for convolution which itself builds on existing concepts?
en.wikipedia.org/wiki/Canny_e...

I don't know what algorithm Geekbench uses for those two subtests but it would be interesting to find out why that in particular is worse.

In Geekbench GPU I can see the 8 Elite is about 30% faster than a Steam Deck which is okay.

The subtests show interesting differences. The 8 Elite only matches the Steam Deck's APU in the Horizon Detection component and its worse in Edge Detection.

browser.geekbench.com/v6/compute/c...

TSR was a similar story, but better. I only tested these in the menu but TSR Epic was like 42fps, TSR medium was like 56fps and lower was 60fps.

It seems like TSR just makes use of compute shaders without any machine learning