Rikimaru
Veteran
LGA MB pins break much more easily.Am I the only one excited about AM5 and LGA? The thought of no more bent pins gives me a lil chubby.
- Status
Bondrewd
Veteran
Unfortunately for you, the IHS is pain and the actual CPU will be extremely brittle.
LGA MB pins are well-protected during packaging and transport. CPU pins are usually not and the defect ratio seems to be higher.
Bondrewd
Veteran
That's not where stuff goes sour.
Most bent LGA pins are the installing idiot dropping the CPU sideways onto the naked pin grid.
Rikimaru
Veteran
I've never dropped a CPU but I think it's pretty easy to drop LGA CPU on MB. They are very thin and do not have a lot of area for a good grip.
DavidGraham
Veteran
Couldn't be further from the truth, NPUs in mobile phones are used extensively, from camera and video applications to voice commands/sound recognition, assistance agents (Google, Siri ..etc), GPS apps, face recognition, on the fly emjoi, smart typing, text prediction, battery optimizations according to user habits, touch recognition, language translation, speech to text .. hell, the entirety of iOS as about now relies heavily on machine learning algorithms that are accelerated locally on the device.
Bondrewd
Veteran
Yeah and usually it's fine but when it hits the pins; catastrophe.
Nearly everything here isn't ondevice or isn't touching the weird non-standard ML block what-so-ever.
Everyone baked their processing into ISP codecs so ML garbage goes off even with camera.
That's bullshit and you know it.
Please stop trying to give me HotChips 2018 flashbacks; it was silly back then and is silly now.
DavidGraham
Veteran
The only bullshit here are your statements, mobile companies are increasingly investing into ML blocks, Apple is unquestionably emphasizing the heavy involvements of their NPUs.
Have an educated read:
https://arstechnica.com/gadgets/202...s-machine-learning-across-ios-and-soon-macos/
https://www.samsung.com/global/galaxy/what-is/npu/
Bondrewd
Veteran
Yea for marketing since CPU/GPU gains of note are no more.
Please no more HC2018 rethread okay?
It was silly as is back then.
Lmao I'd do that too if my CPU design teams deserted.
I seldom agree with most of what DavidGraham posts, but he’s right on this one.
iOS has been utilising the NPU on the A12 Bionic and newer (the A11 Bionic’s NPU was too underspecc’d) for photography and video applications.
As alluded to, executing ML on the CPU is a fallback option for iOS devices with an A11 Bionic or older chipset, and Intel-based Mac computers.
source: my time at Apple and discussions with my SW engineering buddies.
But anyway, back to AMD discussion.
AMD has just secured conditional approval from Chinese regulators to acquire Xilinx ...
https://www.reuters.com/technology/china-conditionally-approves-amds-35-bln-deal-xilinx-2022-01-27/
https://www.reuters.com/technology/china-conditionally-approves-amds-35-bln-deal-xilinx-2022-01-27/
I always thought Threadripper handled this issue pretty well.
Bondrewd
Veteran
True that but AM5 isn't anything like SP3/5/6 mechanically.
Also you can still just accidentally drop the entire carrier frame on naked pins too.
Bondrewd
Veteran
https://ir.amd.com/news-events/pres...s-fourth-quarter-and-full-year-2021-financial
Baddaboom.
$21.5B FY22 guidance without even Xilinx.
Baddaboom.
$21.5B FY22 guidance without even Xilinx.
She's pretty damn competent, and without resorting to a fem-centric pandering narrative. Clearly AMD is doing great, but I can't help but look at Apple's M1s and wish AMD had that kind of overall graphics performance on a single SoC.
Bondrewd
Veteran
Wish granted early next year.
The potent M1 Ultra chips are targeting devices *starting* at $4k. How large is that market?
- Status
Similar threads
- Locked