Somehow, it seems like I am talking to a ChatGPT . While usually I am against Saas, nobody is forcing you to use it. You can always host your own services locally. With the current cost, it might be even cheaper to buy own hardware+software.That's cashburn-fueled negative margin product trying to amass critical install base before hiking the pricing aka the usual SaaS stuff.
That thing loses MS even more money.Somehow, it seems like I am talking to a ChatGPT
Price of HW (perf / $) and thus cost of service will go down over time.That's cashburn-fueled negative margin product trying to amass critical install base before hiking the pricing aka the usual SaaS stuff.
Where.Price of HW (perf / $)
The algorithms that makes up the AI will get smarter. Less training time needed, higher accuracy. Its what keeping the Pentagon folk awake at night. Despite AI gear bans for China, their AI's will get better and better.Price of HW (perf / $) and thus cost of service will go down over time.
If we won't be careful, I wouldn't be surprised If humanity will become batteries for robots or be just outright destroyed.The algorithms that makes up the AI will get smarter. Less training time needed, higher accuracy. Its what keeping the Pentagon folk awake at night. Despite AI gear bans for China, their AI's will get better and better.
With how bad the job situation is in the world, people would actually line-up to be used as batteries while they sleep, as long as they can have a "normal" life.If we won't be careful, I wouldn't be surprised If humanity will become batteries for robots or be just outright destroyed.
Productivity increase in engineering?Point taken about Microsoft being partial.
But I can tell you that a lot of companies in general are expecting similar results that Microsoft got w.r.t. productivity increase for office workers, engineers, etc in general resulting from Generative AI. Although perhaps a bit lower percentages.
Speaking of engineers, I am one. Much of my current work involves peer review/ verification of other engineers' work. I can't see myself having any generative AI going on on my side of the equation. Neither should there be any at the side of the engineers whose work I am verifying.Have you ever used a calculator? Does a calculator increases your productivity? Not that you can't do calculations with pen and a paper, just that it much easier to use a calculator. Same with AI.
Yes, from e.g. Github Copilot as others also have mentioned.Productivity increase in engineering?
You mean on Zen5 desktop CPUs?With how Strix Point should have 2x more powerful NPU, I think AMD will use even more of die size for this unit.
I meant Strix Point, because as you linked It should have 3x more TOPs, although it won't come only from NPU.You mean on Zen5 desktop CPUs?
Yes, I guess the question is how powerful the NPU on it will be. For comparison, the NPU on Hawk Point is 16 TOPS and Strix Point is rumored to be 48+ TOPS according to this article:
AMD next-gen "Strix Point" series with XDNA2 NPU to offer 3x performance boost for generative AI - VideoCardz.com
AMD Strix Point comes with XDNA2 AI processor AMD is filling the gaps in the next-gen mobile APU series specs. AMD has stated that its upcoming Strix Point APU series will incorporate the latest XDNA2 AI core. This processor is expected to deliver a performance boost of up to three times...videocardz.com
I meant Strix Point, because as you linked It should have 3x more TOPs, although it won't come only from NPU.
Ok, I am blind. It should be only from the NPU.Are you sure it's not from the TPU alone?
This is from the article:
indicating that Hawk Point has 16 NPU TOPS, and 39 total TOPS.
Then there is this from the same article:
View attachment 90145
again indicating 16 TOPS for Hawk Point, and on the same row for Strix Point 48+ TOPS. So my interpretation is that it's from TPU alone?
Productivity increase in engineering?
Hmm, okey, I guess besides engineering which relies a lot on analytical thinking, knowledge, experience, and scrutiny, there is also engineering which relies more on googling and trial-and-error. Like when I for example piece together simple helper scripts for my computer hobby at home (although I don't see this pastime as engineering). Yep, there might be potential for productivity gain with an AI assistant in such lines of work.Yes, from e.g. Github Copilot as others also have mentioned.
I can see that generative AI comes in handy,
– if it doesn't matter if "IP" of uncontrolled origin sips into your work,
– if you are allowed to simply make up stuff now and then.
Not sure if you're agreeing or not with that comment. Anyway, currently to my understanding the Github Copilot AI inferencing is done on severs in the cloud. But I assume when PCs get equipped with NPU, the NPU can offload some of the work from the servers. So Github Copilot has a connection to NPUs in that sense.Edit: Github Copilot has little to do with NPUs though, does it, to somewhat get back to topic.
More off topic, Visual Studio 2022 already have AI built-in even without Copilot. It analyses current project and does help a lot when programming. It can understand your coding style and able to do code suggestion and completion accordingly. After getting used to it, it is bit annoying to work without it.Not sure if you're agreeing or not with that comment. Anyway, currently to my understanding the Github Copilot AI inferencing is done on severs in the cloud. But I assume when PCs get equipped with NPU, the NPU can offload some of the work from the servers. So Github Copilot has a connection to NPUs in that sense.
It doesn't exist.So assuming an NPU will be included on Zen5, where would be best to place it? On the CCD or IOD?
Are you just speculating? Otherwise provide link to source.It doesn't exist.
It's fully Win12 compliant, just not for premium(tm) AI PCs.I have a hard time seeing how AMD would intentionally render Zen5 not fully Win12 compliant by not including an NPU
Are you just speculating? Otherwise provide link to source.
I have a hard time seeing how AMD would intentionally render Zen5 not fully Win12 compliant by not including an NPU. That’s just bad business and a distaster from marketing perspective whether you like it or not.
Does this mean cheap low profile PCIe GPUs with decent AI acceleration for OEMs so they can advertise their desktops as Premium AI PCs?It's fully Win12 compliant, just not for premium(tm) AI PCs.
No, it means absolutely nothing outside of premium (>$999) laptop space.Does this mean cheap low profile PCIe GPUs with decent AI acceleration for OEMs so they can advertise their desktops as Premium AI PCs?