Sounds like we're about to get a lot more Linux users!
-
Sounds like we're about to get a lot more Linux users!
-
Sounds like we're about to get a lot more Linux users!
-
Sounds like we're about to get a lot more Linux users!
@mttaggart it's the year of Linux on the Desktop (for real this time!)
-
Sounds like we're about to get a lot more Linux users!
@mttaggart oh god sweet Jeebus, Please no. Who of the user base asked for this?
Also, a special fuck you for chasing Apple's looks, including the return of dumb transparency. Why not come up with something yourself, MS? Oh, right, because it might actually be worse.
-
@mttaggart oh god sweet Jeebus, Please no. Who of the user base asked for this?
Also, a special fuck you for chasing Apple's looks, including the return of dumb transparency. Why not come up with something yourself, MS? Oh, right, because it might actually be worse.
@kwramm @mttaggart if MS copies Apple copying MS how far does the rabbit hole go? 🤪
-
Sounds like we're about to get a lot more Linux users!
@mttaggart still waiting for what they’re actually going to do with these NPUs
-
@bmoreinis jesus fucking christ..not compatible with FOSS
how about they just move to another jurisdiction ? fuck the fascists and their tentacles
this person has no moral compass. should not be a dev. they instantly capitulate to, "how can I do the fascism for them?!"
-
@mttaggart still waiting for what they’re actually going to do with these NPUs
@malwareminigun @mttaggart Copilot...
Or do you mean you think they'll essentially use their users as a way to crowdsource compute? (Ok, I would not put that past Microsoft... Worse, I wouldn't put it past the apologist users to just accept it...)
-
Sounds like we're about to get a lot more Linux users!
@mttaggart The NPU requirement is a heck of a hurdle. Quite a lot of people are just going to straight up say no to that. Cost and availability are a huge problem for this (especially availability as shortages start working their way down.)
EDIT: Looking it up I think the NPUs are an even bigger hurdle than I thought? I assumed by now one could maybe buy some PCI-E card or something similar to a GPU, but... I'm not really seeing them? So then the only consumer-available options for NPUs are specialized devices with them built in... In other words, it essentially says no matter what you own, you probably have to buy something else...
I think they have indeed shot themselves in the foot with this one. EDIT: Now I think they cut their foot off with a hacksaw...
-
Sounds like we're about to get a lot more Linux users!
@mttaggart They should call it Windows Modular Extension, or "Windows ME"
-
Sounds like we're about to get a lot more Linux users!
@mttaggart "By the way: if you are using Windows 11 Home, you are missing out on the many advantages of the Pro version. The Windows 11 upgrade is available in the PCWorld software store for a low price of $59 instead of $99.". How about an upgrade for $0, increased uptime, zero blue screens and a faster computer with less resources?
-
@bmoreinis jesus fucking christ..not compatible with FOSS
how about they just move to another jurisdiction ? fuck the fascists and their tentacles
this person has no moral compass. should not be a dev. they instantly capitulate to, "how can I do the fascism for them?!"
@bweller @bmoreinis @mttaggart @catsalad Agreed. Instant capitulation to fascism there.
-
@malwareminigun @mttaggart Copilot...
Or do you mean you think they'll essentially use their users as a way to crowdsource compute? (Ok, I would not put that past Microsoft... Worse, I wouldn't put it past the apologist users to just accept it...)
@nazokiyoubinbou @mttaggart Nothing like that; to the best of my understanding it would be meaningless anyway because the NPUs are slower than the iGPUs on all these platforms. As I understand it, the "NPU" is only about being able to do this stuff at lower power, but it's not clear what features actually use it.
-
@nazokiyoubinbou @mttaggart Nothing like that; to the best of my understanding it would be meaningless anyway because the NPUs are slower than the iGPUs on all these platforms. As I understand it, the "NPU" is only about being able to do this stuff at lower power, but it's not clear what features actually use it.
@malwareminigun @mttaggart My understanding is NPUs are faster and more efficient than GPUs at "ML" tasks (so called "AI.")
As you say though, they're designed for extreme low power. If you scaled a NPU to compete with GPUs I'm pretty sure they'd come out ahead at those specific tasks.
My guess here is, more than anything else, MS just wants to be able to do some stuff locally faster than what can be done via sending to their servers. For example, processing all the screenshots they're taking of people's desktops can be done on the system itself, then they can send all the data to Microsoft to collect after. That data is much smaller than the actual screenshots. (Especially since this could allow doing things like checking for when it actually changes before sending.)
-
@malwareminigun @mttaggart My understanding is NPUs are faster and more efficient than GPUs at "ML" tasks (so called "AI.")
As you say though, they're designed for extreme low power. If you scaled a NPU to compete with GPUs I'm pretty sure they'd come out ahead at those specific tasks.
My guess here is, more than anything else, MS just wants to be able to do some stuff locally faster than what can be done via sending to their servers. For example, processing all the screenshots they're taking of people's desktops can be done on the system itself, then they can send all the data to Microsoft to collect after. That data is much smaller than the actual screenshots. (Especially since this could allow doing things like checking for when it actually changes before sending.)
@nazokiyoubinbou Sure, I'm not talking about hypothetical hardware, I'm talking about what's actually shipping. As far as I understood Recall did not use the NPU but what that thing even is has changed a lot
-
@nazokiyoubinbou Sure, I'm not talking about hypothetical hardware, I'm talking about what's actually shipping. As far as I understood Recall did not use the NPU but what that thing even is has changed a lot
@malwareminigun Ok, if we talk about what is typical in most of these devices, then yeah, it's pretty much just to accelerate simple things locally and they probably still intend most of the real processing to be handled on their own servers.
This mostly just makes it a bit cheaper for them to collect their users' private data.
I do think they likely beat out iGPUs though. Plus they're better positioned (eg not going to scorch the system and give away the fact that they're doing all that to users who only look at temps or something.) This also frees up the GPU so it won't be visible in games and stuff if it keeps doing it. The intent is probably to keep them constantly working on collecting and processing data so it's just a background constant people simply accept long term.
-
@malwareminigun Ok, if we talk about what is typical in most of these devices, then yeah, it's pretty much just to accelerate simple things locally and they probably still intend most of the real processing to be handled on their own servers.
This mostly just makes it a bit cheaper for them to collect their users' private data.
I do think they likely beat out iGPUs though. Plus they're better positioned (eg not going to scorch the system and give away the fact that they're doing all that to users who only look at temps or something.) This also frees up the GPU so it won't be visible in games and stuff if it keeps doing it. The intent is probably to keep them constantly working on collecting and processing data so it's just a background constant people simply accept long term.
@nazokiyoubinbou https://chipsandcheese.com/p/intel-meteor-lakes-npu
> Meteor Lake’s iGPU has no problem beating the NPU, as Intel’s slide suggests. My “All GPU” result of 1.38 iterations per second aligns very closely with Intel’s 20 iterations over 14.5 seconds, or 1.379 iterations per second. The iGPU has a noticeable 62% performance advantage over the NPU, even while chugging a full-fat FP32 model. If I level the playing field by handing Intel’s INT8 model to the iGPU, its lead extends to a massive 261%. Finally, a discrete GPU like AMD’s RX 6900 XT is 6.7 times faster even when using FP32.
-
@nazokiyoubinbou https://chipsandcheese.com/p/intel-meteor-lakes-npu
> Meteor Lake’s iGPU has no problem beating the NPU, as Intel’s slide suggests. My “All GPU” result of 1.38 iterations per second aligns very closely with Intel’s 20 iterations over 14.5 seconds, or 1.379 iterations per second. The iGPU has a noticeable 62% performance advantage over the NPU, even while chugging a full-fat FP32 model. If I level the playing field by handing Intel’s INT8 model to the iGPU, its lead extends to a massive 261%. Finally, a discrete GPU like AMD’s RX 6900 XT is 6.7 times faster even when using FP32.
@malwareminigun I thought we weren't comparing discreet GPUs for the obvious reason mentioned above? A 6700XT is not even on the same scale as a NPU... A proper comparison would be like running it on a modern Ryzen chip's built-in GPU.
I'll accept the point on meteor lake, though I'm not sure if they're at the same level even then
Honestly, I suppose it's probably, more than anything else, to free up the GPU so it won't impact gaming or force them to pause the data collection while the GPU is in heavy use. I kind of added edits to that effect, but I'll go all out and say outright that it may be that more than anything else. This way even if you're in a game there is essentially hardware dedicated to doing this stuff in the background that barely impacts anything else in the system.
-
Sounds like we're about to get a lot more Linux users!
@mttaggart hopefully, I'm not gonna run on my hardware, that's for sure
-
@malwareminigun I thought we weren't comparing discreet GPUs for the obvious reason mentioned above? A 6700XT is not even on the same scale as a NPU... A proper comparison would be like running it on a modern Ryzen chip's built-in GPU.
I'll accept the point on meteor lake, though I'm not sure if they're at the same level even then
Honestly, I suppose it's probably, more than anything else, to free up the GPU so it won't impact gaming or force them to pause the data collection while the GPU is in heavy use. I kind of added edits to that effect, but I'll go all out and say outright that it may be that more than anything else. This way even if you're in a game there is essentially hardware dedicated to doing this stuff in the background that barely impacts anything else in the system.
@nazokiyoubinbou I'm citing this for the IGP numbers being substantially higher. That a DGP is another factor of three or more is expected but not the main point here.