Recent content by pavel

Extra cost, yes, but doing 22AC-48DC-12DC-1DC with good 220AC-48DC stage may still be better than 220AC-12DC with silicon. If 48DC is regulated, you may well try a fixed ratio topology for 48-12

Certainly, this is where I go for the technical direction. DC-DC will have to be much beefier and with higher input voltage to support a 200W+ GPU , and there will be no independent bridge to reduce complexity and layer count. I'm even thinking to forego going beyond default 4 USB ports on ZEN...

If I will go with an established x86 design shop, I will certainly ask them to do that if the have access to reference BIOS. If otherwise, in worst case I can "copypaste" it from a relatively similar board. Bios for ZEN is said to be very simple, with the only piece of code that needs to be...

I have quite a good understanding that a complex board can go up to $100k for something with a lot of high speed signals and layers, as you need more tries to debug the board, and buyers for that stuff tend to be well moneyed (telecom, carrier grade router makers, military etc.) That's why I'm...

The biggest thing I've ever developed myself was a simplest android settop box. Went through 3 prototype boards until I got everything running reliably. Usually, I only take upon microcontroller projects by myself. I'm not against dishing out $3 to $5k on getting the board done by a...

Hi guys, For quite some time I was thinking making a custom TR4 mobo for a workstation leasing business. Other boards out there are big overkills in my opinion. Other than the north bridge itself, things like sound card, PCIE/SATA expansions, WiFI, additional USB hubs, tons of "management"...

Do you know if it is a US only restriction or an international one (ICAO's) ?

There are empty pads for a connector near the barrel connector.

I recently bought a $30 NVME to type-c adaptor from Orico store in Shenzhen. If chipsets are the same, the price is ought to go down.

I'll introduce you to theory. So, the main thing in every DC-DC converter is a switch which can be any type of transistor, but usually a MOSFET. Secondary to it come capacitors and inductors that store the electricity while the switch is off. Theory of operations: the switch switches on until...

If we are determined to go further into the rabbit hole, we can think about other soft switching topologies. The main advantage of multiphase bucks is that they are easy to control, and that they deal with current spikes more or less gracefully, while being very simple. All kinds of resonant...

If we want to go even more adventurous, there is a "power-on-package" approach, where POL is placed right on the package. http://www.vicorpower.com/industries-computing/power-on-package-technology. That minimises "last centimetre" losses even further, but I suppose that the switching element...

Certainly not, at 90% worst case scenario, you will only be dissipating 50w of heat from 500w load. That's close to efficiency of existing Vcore VRMs, and the net efficiency goes up. Here are some studies from a vendor...

The idea is that we are throwing out a dedicated ATX psu and use DC-DC at main PCB using very high-end parts to minimise switching losses. GaNFETs can remain 92-93% efficient even in 1-2 megahertz region. Having size of passives minimised with this, losses in passives will go down as well.

Once it is settled that all "fast and hot" ICs are placed on a single big package, just like in a picture above, the remaining parts on the board are the IO and power. The power will be an issue, feeding a package consuming up to half a kilowatt will not be easy. As voltages will be around 1V...