Monday, July 08
What's That Lassie Edition
- Having a back button on your mouse is really handy right up until you click it in the middly of typing a long post and find that while the backspace key no longer erases everything, the back button sure does.
- So, I have this great idea for -
Available right now starting at €19.95 for the Duinomite Mini. (As a side note, the original designer of the Maximite is not entirely happy with Olimex.)
It uses an 80MHz Pic32MX795 - somewhat faster than the model I chose, but not available in DIP package, and slower and more expensive than the newer Pic32MX4 or Pic32MK that I considered as upgrades.
I think my video design is neater. The Maximite has the advantage, though, of actually existing.
- My video design is more flexible and capable (on paper) than the Maximite, but it does require two separate - what's that Lassie?
TimmyMicrochip has a dual-core microcontroller that puts everything I could want in a single package?
The Microchip SAM4C has two Arm cores running at 120MHz. CPU0 is an integer-only Cortex M4 with 2MB flash and 256KB RAM, and CPU1 is a floating-point Cortex M4F with 32KB code RAM and 16KB data RAM.
It's A$13.20 qty 1, so almost exactly twice the price of the 28-pin Pic32 I started this idea with. But it's more than twice the clock speed, and has two cores, and has eight times the flash and four-and-three-quarters times the RAM. It's surface mount, but you only have to deal with one of it.
Now, the 16KB data RAM on CPU1 is not remotely big enough for our frame buffer, but CPU1 can read and write CPU0's 256KB RAM. (It can't directly access the flash memory, which is why it has its own 32KB of code RAM.)
So, if we up our base resolution to 480x270 - which is what I wanted all along, and which fits in pretty well with a 120MHz clock - the frame buffer in 256-colour mode will use about half our 256KB.
CPU0 can draw merrily away in the frame buffer (rather than sending draw commands over a serial bus) and CPU1 will read from there, perform pixel transforms, and write to the line buffer, which it can then push out over its own DMA channel.
If we also up the colour depth to 15/16 bits, we'll need a 2KB lookup table and two 1920-byte line buffers (for 960x270 high-resolution mode, 2 bytes per pixel). Which is no problem even for our limited 16KB. We're pushing 2.25x the number of pixels of the original design, but wih 2.25x the clock speed, so we're good.
- Or you can do it the hard way.
On the other hand, this thing is plotting the Mandelbrot set using nothing but TTL. It has eight instructions.
Update: Now with a Basic interpreter. Running on TTL chips. With 8 instructions.
You save your programs by writing them to the buffer on the keboard controller, and load them by telling it to type them back in.
This one does have 64 colours, which is more than some of the retro models with fancy video hardware manage. It's amazing what they've managed to get out of this thing.
- Or you could do it the easy way.
- Or the crazy way.
This is an Apple I clone. Yes, Apple One.
The video controller is, um, it's a Parallax Propeller. It's not actually a video controller, not as such, it's more of a, well, more of an 8 core 80MHz RISC microcontroller and a several hundred times faster than the 6502 it's attached to. And it's pretending to be a serial terminal.
- What about news, you ask? What news? It's Monday in Australia after a long weekend in the US and there isn't any oh wait.
- AMD's Navi looks to be pretty competitive. (AnandTech)
Nvidia countered with their RTX Super but then AMD counter-countered with a price cut.
As a result, the 5700XT now lines up with the RTX 2060 Super, which it consistently beats. The 5700 is the same price as the original RTX 2060, and it's both faster and has more memory.
Pretty solid cards unless you're planning to use them for compute tasks in which case you might want to wait for a driver update or two first.
The RTX 2070 Super is faster than the 5700XT - but not by much, and it costs 25% more.
- On the CPU side of things, things are more clear-cut: Intel is toast. (AnandTech)
AMD's 3700X is competitive with Intel's i9-9900K, while costing 35% less and using 40% less power.
The 12-core 3900X simply runs away with the ball any time you fire up a multi-threaded task, and actually uses 15% less power than the 9900K despite having four more cores and a 10% higher rated TDP.
Did AMD deliver? They delivered.
A note on memory and the Infinity Fabric: Ryzen 3000 CPUs support DDR4-3200 at stock, and up to 3733 clocked 1:1 with the Infinity Fabric controller. Above that it drops down to 2:1 and latency jumps up; AMD recommends DDR4-3600 as the best price/performance compromise.
As for latency, yes, it's up a little with the new I/O die. But in benchmarks and real-world tests the huge L3 cache - 32MB on the 3700X, 64MB on the 3900X - means that it simply doesn't matter.
- Motherboards? ASRock's X570 Taichi is a solid entry if not exactly inexpensive. (Tom's Hardware)
It only has gigabit Ethernet, and the reviewer was not enamoured of the chipset fan, but otherwise it's pretty solid.
The review mentions a Thunderbolt adaptor, but it is not included in the box. Some other models do have Thunderbolt either on board or via an adaptor - a first for the AMD platform - just not this one.
- iBrick, new from Apple! (Discontinued.) (Chromium.org)
A carefully formatted text message or email could soft-brick your iPhone. You could recover by wiping and reinstalling... At least until the next message arrived five minutes later.
It's fixed now.
- Time to update the firmware in your... Keyboard?! (Heise.de)
- An introduction to Dockerfile best practices.
1. Don't do that. Bad devops team! Bad! No biscuit!
- A Ruby library for checking password strength... You think you know where this is going, don't you? (ZDNet)
Actually, it's worse than that: It didn't just send off all your passwords to a remote server, it scanned your network to see if the password worked anywhere and downloaded and installed a bcakdoor in your own software.
This is not a flaw in Ruby at all, but in the way we distribute software modules. What the fix for this is, I don't know.
- Except that if your server is down, no-one can hack you. (One Angry Gamer)
- On fifth thoughts:
If I did add a third Pic32MX2 as a sound chip, mostly doing polyphonic waveform synthesis in software, maybe a little PCM, it would have enough RAM and CPU resources left over to act as a sprite generator. If sprites are 16x16 in 256, they'll be 256 bytes each, so you could have 128 of them at once and still have half the memory free for sound.
It would use the same LUT and line buffer approach as the video controller, and pre-order the sprites for efficient output to the line buffer. There would probably need to be a limit to the number of sprites that can coexist on one line. I might be able to smoosh them all together within the time limit (our software horizontal scan rate is only 5.5kHz), but it could get fiddly.
A $2.97 44-pin CPLD could act as the video output buffer and system controller. It's a surface-mount part, but it's PLCC, and sockets for those are cheap and easy to find. And it's in-system programmable so no worries about getting it wrong the first 97 times.
Of the 36 I/Os, most would be used right away merging the graphics and sprite data and sending it all out again, but there should be enough left for various control lines.
- Oh look it's a what the heck.
Shantae and Studio Trigger?
Video of the Day
You'll need to watch this one on YouTube I'm afraid because the entire channel has been deleted from BitChute and no I didn't get that the wrong way 'round.
Disclaimer: Back button, back button, wherefore art thou, back button?
I think there may also be something wrong with my motherboard because I had a magic smoke moment last month with my CPU AIO's water pump, and ever since then, when I shut it down it immediately powers itself back on. Unfortunately I didn't get to Micro Center before they opened yesterday so there aren't any Mini ITX X570 motherboards available.
Posted by: Rick C at Tuesday, July 09 2019 02:38 AM (Iwkd4)
Posted by: Pixy Misa at Tuesday, July 09 2019 09:44 AM (PiXy!)
In spite of the high voltage (and thus it idles in the mid-to-high 40s), it's still pulling fairly low wattage; in Guild Wars 2 it was only pulling 30W total package power.
It appears the clock increments are 100MHz, and also, some cores will drop down to 2GHz when idle, which is incredible for the former, and much nicer than first-gen on the latter, if you want to save power but also have performance (my 1600x would basically never drop lower than baseclock until I started adding lower-speed P-states, which is kind of annoying to do.
Posted by: Rick C at Tuesday, July 09 2019 11:06 AM (Iwkd4)
Posted by: Jay at Tuesday, July 09 2019 02:29 PM (vuQH5)
Now cast your mind way back to 1995, to the ancester of Intel's Core microarchitecture, the P6 and its first implementation, the Pentium Pro. A MCM with two or three dies, the CPU die and one or two L2 cache dies. Depending on the model, the dies could be manufactured in different or same processes.
Posted by: Kayle at Tuesday, July 09 2019 05:58 PM (magRz)
On the downside, higher theoretical latency. On the upside, as I like to say, it actually exists.
Posted by: Pixy Misa at Tuesday, July 09 2019 06:12 PM (PiXy!)
Posted by: Pixy Misa at Tuesday, July 09 2019 06:19 PM (PiXy!)
I guess a different integration that might make sense for desktop and laptop would be to integrate two CCXs and a cut down Navi core for the 4th gen APUs.
Posted by: Kayle at Tuesday, July 09 2019 06:30 PM (magRz)
These may be worth more than any net savings from making a different chip, especially with the rumors that the X570 is just a Ryzen IO die with parts fused off.
Posted by: Rick C at Wednesday, July 10 2019 12:32 AM (Iwkd4)
58 queries taking 0.1858 seconds, 297 records returned.
Powered by Minx 1.1.6c-pink.