Discuss Scratch

__Falcon-Games__
Scratcher
1000+ posts

Hardware Lessons

GIitchInTheMatrix wrote:

__Falcon-Games__ wrote:

GIitchInTheMatrix wrote:

Update: Replit is apparently moving hosting to a payed service(at least, that’s what it sounds like)
That’s to say, i may have to move it to Glitch(just like my main site)
You could move it to GitHub as it is just static pages and then I can also directly contribute as I have a GitHub account.
Unfortunately, my school blocked GitHub.
(Which is strange, as they just started a CS class)
Well that's unfortunate.
davidtheplatform
Scratcher
500+ posts

Hardware Lessons

GIitchInTheMatrix wrote:

Update: Replit is apparently moving hosting to a payed service(at least, that’s what it sounds like)
That’s to say, i may have to move it to Glitch(just like my main site)
Source?
GIitchInTheMatrix
Scratcher
1000+ posts

Hardware Lessons

davidtheplatform wrote:

GIitchInTheMatrix wrote:

Update: Replit is apparently moving hosting to a payed service(at least, that’s what it sounds like)
That’s to say, i may have to move it to Glitch(just like my main site)
Source?
https://blog.replit.com/hosting-changes
__Falcon-Games__
Scratcher
1000+ posts

Hardware Lessons

ajskateboarder wrote:

__Falcon-Games__ wrote:

CPU

Well more technically CPUs run one binary instruction per each cycle and the amount of cycles is controlled by the clock. Clock speed is measured in GHz specifically how many billion cycles run per second. Multiple cores are usually in your CPU to run multiple instructions at once.
And soon enough, thanks to Moore's Law, we will measure trillions of cycles a second someday
“Hey, my computer is pretty low-end, it's just 1.8 THz.”
__Falcon-Games__
Scratcher
1000+ posts

Hardware Lessons

So here's how a TPU and DPU works if you ever decide to make those pages.

TPU
A TPU which stands for Tensor Processing Unit is a specific kind of processing unit designed for training AI through Google's TensorFlow and usually is designed to do much more equations then the CPU does and do them fast too. It is designed for very low precision computing and does not have any display or monitor connected to it but does have an input/output system where inputs can go into the neural network and outputs is the result of the neural network. It is useful for convolutional neural networks which basically uses filtering and optimization to extract specific information from raw data.

TPUs are owned by Google and other companies can buy TPUs from Google to use them too for example AlphaZero, a famous Chess bot, used TPUs and Google uses them in Google Street View for text processing. Other vendors have also attempted to make things like Google's TPU. The TPU has had five generations so far with the latest being the TPU v5 announced in 2021 and plans to do a comparison this year against the Nvidia Hopper H100.

On an even more lower level, a TPU is an ASIC which stands for application-specific integrated circuit and an ASIC is an integrated chip which include microprocessors and microcontrollers but are designed for a specific application rather then general purpose and in this case the TPU is also AI-accelerated which makes it different then other ASICs out there.

DPU
The CPU used to do a lot of work with math and graphics rendering and data processing and a CPU couldn't do that alone so we made GPUs to cut graphics processing out into it's own unit, still a CPU cannot do so much math and data processing at once so a DPU standing for data processing unit is designed for data processing, now technically it does more then data processing. It does infrastructure tasks such as handling HTTP requests, TCP/IP requests, do encryption and decryption, all these other basic infrastructure tasks but a lot of these are related with data. It integrates a CPU with network interface hardware and allows the CPU to do more important and specific tasks while the DPU processes these basic tasks. It is also called IPU standing for infrastructure processing unit or SmartNICs as it takes responsibility off the CPU like a NIC does but to an even greater extent.

Cloud providers can benefit from DPUs a lot because they do a lot of network operations and the CPU can get very busy so a DPU can do the basic network operations and the CPU can manage more specific or complex cloud computing tasks which can make cloud providers be able to assist more customers which of course means more money. Now going more low-level, a DPU is a programable processor like the CPU and GPU and isn't common yet but will likely become more common in computers, it is a system on a chip (SoC) like the Apple Silicon but is a high performant multi-core CPU basically along with high-performance networking and flexible acceleration.

A DPU can do data packet parsing and implement an open virtual switch (OVS), RDMA data transport acceleration, bypass the CPU and feed networked data directly into GPUs, TCP acceleration, network virtualization, enable multimedia streaming, content distribution networks and 4K/8K video over IP along with much much more.
davidtheplatform
Scratcher
500+ posts

Hardware Lessons

GIitchInTheMatrix wrote:

davidtheplatform wrote:

GIitchInTheMatrix wrote:

Update: Replit is apparently moving hosting to a payed service(at least, that’s what it sounds like)
That’s to say, i may have to move it to Glitch(just like my main site)
Source?
https://blog.replit.com/hosting-changes
The free tier is staying the same. They're changing how some paid features work
god286
Scratcher
1000+ posts

Hardware Lessons

davidtheplatform wrote:

The free tier is staying the same. They're changing how some paid features work
No, you misunderstood the text. Here:

Replit wrote:

On January 1st, 2024, repl.co domains will transition to replit.dev and will only be accessible when someone is in the editor. If you are currently “hosting” anything on repl.co, we strongly recommend that you migrate to Deployments before the new year.
Replit will only work when someone is in the editor for websites, but the only free deployment is a static website which Github Pages can do much better.
davidtheplatform
Scratcher
500+ posts

Hardware Lessons

god286 wrote:

davidtheplatform wrote:

The free tier is staying the same. They're changing how some paid features work
No, you misunderstood the text. Here:

Replit wrote:

On January 1st, 2024, repl.co domains will transition to replit.dev and will only be accessible when someone is in the editor. If you are currently “hosting” anything on repl.co, we strongly recommend that you migrate to Deployments before the new year.
Replit will only work when someone is in the editor for websites, but the only free deployment is a static website which Github Pages can do much better.
I didn't see the static part, I just thought they were changing the domain name.

This isn't good for replit. I have a feeling they aren't doing very well financially

Powered by DjangoBB