Ooooo, and it supports secondary addressing? NONE of the low-cost interfaces support secondary addressing! (At least last time I looked around!)
That's a big deal for those of us with R&S CMU200's / CRTU's, which require secondary addressing to make it act like a spec-an with tracking generator. And I see that in your list of tested equipment!
Fantastic. And the price of prebuilt units is super reasonable. I'm in.
You are correct. UsbTmc exposes no multi device capability. Also when I made first versions many years ago I had no "garden hose" GPIB cables and saw they were more expensive as the whole adapter.
V3 which will come early next year supports GPIB daisy chaining/multi device support via Ethernet VXI11 & HiSlip.
The V2 adapter shown here supports trigger, SRQ, serial poll, trigger, clear, local lockout, goto lockal, pulse indicator, all features though.
If you absolutely need common synchronous trigger to a chain of GPIB instruments which is quite seldomly absolutely required, you'll have it in next version V3.
Great work making this available for sale. NI and Keysight are the only two traditional vendors of this kind of thing and they’re priced close to $1k these days. There are tons of knock-offs on Ebay that won’t last more than a few months if they ever work.
Even worse, the windows drivers know they're knock offs and let you use them for a bit and then error out. It's the perfect lesson. They give you that high of "yes, I got one over on the big guys, we're cruising now" and then "I don't know boss it's just all of a sudden not working anymore. Yes I know we're in a production crunch. Yes we should've just bought the real one". On Linux though, no issues for me
FWIW, You can buy legit used 82357s for quite a bit less than 1k. Anecdotally, I've never had one fail in 20 years. Probably bought a dozen over that time frame. All used daily.
The step to go to ARM or risc v based controllers is easy. That's what I did with V3. Gives a tremendous performance boost and costs 1/2 of that old AVR.
If you control your application stack, you can pivot from expensive NI GPIB hardware to something cheaper. Besides prices, you get the advantage of not requiring a 1.5GB Windows-only NI MAX install on a PC just to connect to the device.
Since I hadn't heard of open source hardware, I bought 5 of these [0] instead around 2020 and they've been reliable for daily lab use. You just connect to them over TCP and send a couple of commands to select the GPIB addr to correspond with; after that, you can send SCPI commands the same way you always do.
A good way around application overloaded applocation stacks is R&S Visa which is lean and mean while still exposing full visa to applocations like labview. Or going with something like python usbtmc.
If you also have other instruments to control e.g. with usb or ethernet you'll likely anyway look into a visa.
Actually world would need an open source visa stack. Right now you find many applications doing their own thing, but then not supporting vxi11, hislip or usbtmc.
Visa is something which solved that problem for years already but the most standard package NI Visa is overloaded and scaring people away. For that reason I propagate the usefulness of R&S Visa a lot. Lean and mean, supporting all OS and procotocols as it is supposed to be.
Interesting. I built an AR488 a while back (https://github.com/Twilight-Logic/AR488) and it worked alright for my needs. I'd be interested to see a comparison of the two projects.
Ooh! The faster speeds would be very welcome, I was bothered by the slowness of the AR488 but I assumed that was just how GPIB was (I had no baseline to compare to). I'll switch over when I get a chance.
That user's project also looks very interesting - My TDS684A's CRT seems to have died, and rather than fix it I could switch to using a software scope.
The upcoming V3 adapter reaches even more than 1 MByte/s with my fastest instruments. Unless you connect long GPIB cables to it, because capacitive load slows down GPIB as it self regulates speed down in the way GPIB is designed.
Am I wrong in understanding that each physical device supports only a single GPIB device at a time? The enumeration and discovery scheme is certainly novel, but one device per hardware would seem to prevent using some of GPIB’s advanced features like parallel polling and low latency triggers.
Linux gpib also supports my adapter natively with a special linux gpib firmware you can download from the page above. It also supports multiple instruments connected to one adapter then.
I finished my Master's a couple years ago, and in our IC probing lab, it was almost all GPIB. All of the pieces of test equipment, except for a single cutting-edge Keysight scope (which used Ethernet instead), were connected together via those chunky but satisfying and robust cables, usually controlled via MATLAB or some other IC measurement software package.
Yes, those cables are often called "garden hose cables". The only benefit of that bus infrastructure under a nowadays view is that it gives the ability to synchronously trigger multple instruments which is very rarely used or required.
There’s a fairly large second-hand market for old electronics test equipment. A lot of that stuff only had GPIB ports, even well into the 90s. Cheap Chinese entrants (Rigol, Siglent, etc.) only within the last decade started making an impact
I know that very well. Ethernet=security concern. Connecting your shiny new scope to company network=no way. Hard to discuss arround it in company environments or push for split network topologies.
Maybe someonw can explain me, but I never understood the appeal of GPIB for modern instruments (legacy instruments are of course "excused"). Electrically is a terrible interface that introduces ground loops with the control computer. Speed are laughable and it requires exensive and exotic adapters with complex sw stack (I wish this projects good success, it's needed!). Ethernet in comparison tick all my boxes. It's electrically decoupled by default (just use UTP cables), crazy cheap, very fast and with sane sw stack thanks to vxi-11. You can even bypass visa if you wish and open a plain TCP socket, no need for any library. What am I missing?
Not much, but consider latency: You can use the Group Execute Trigger (GET) to simultaneously trigger multiple instruments with both very low latency and very low latency dispersion. Think, easy-to-use sub-microsecond synchronization.
Ethernet and USB 4 may have orders of magnitude more bandwidth, but can’t achieve the same multi-device synchronization capability without side channel signals.
Now, sure, you can add the same capability with a programmable pulse generator connected via coax to the trigger input of all your instruments, but GBIP lets you do that with just the data connection (and you don’t always have a spare trigger channel). The only other protocols I know of with similar capabilities are PXI and PXIe, which are “PCI(express) in an incompatible form-factor, plus some extra signals for real time synchronization”.
sub-microsecond triggering should be doable with a level-2 cut-through switch and an ethernet broadcast no? I admit that ethernet is not really designed for that as the Phy is then the latency bottleneck.
Sure, in principle, but that takes effort and special equipment to set up. The point is that GPIB makes it easy (trivial, actually) with nothing more than the cables you normally use to connect instruments to get very low and predictable latency.
GPIB GET works by first configuring a subset of bus devices as listeners and then sending a single-byte message (it’s an 8 bit bus, so one bus cycle) with the ATN line asserted. It’s intrinsically low latency without any special effort.
Whether that makes it worthwhile to put GPIB on a new instrument in 2025 is a different question. I’m only addressing “what does GPIB give you”?
There are approaches to real-time ethernet (some industry implementations like profinet or ethercat, 802.1as from IEEE) but support is spotty and it requires specialized gear to be effective.
If you buy used equipment which doesn't have Ethernet or your company wants you to ise the stuff that is in the Lab since 10+ years there's simply no other choice. Or companies that see Ethernet as a potential security attack vector.
It's indeed not that GPIB is better than Ethernet. In tiny aspects that's argueable, but as general statement true.
I strongly suspect that this does not meet the strict timing requirements that GPIB has. Putting this on your bus is likely to violate both the T1 hard timing requirements and the impedance requirements.
Use of the "standard" set of 74-series buffers that everybody uses would meet impedance requirements and would also allow the usage of a much faster MCU which likely could be made to adhere to the strict T1 timing requirements (with the caveat that most microcontroller USB stacks are piles of garbage that demand that they get interrupt priority even when you tell them otherwise).
For a situation where the instrument is directly connected to the adapter like here there is no difficulty in fast T1 timing on a MCU.
V3 will have btw. fully compliant GPIB electrical specs without 74 series buffers.
48mA drivers are not required for a direct connection scenario as V2 supports. VIH is a bit violated, though not practically meaningful. So far I heard of no device not working after over 6 years and those who did not in early stages I managed to get running with sw updates.
This is not a one shot plain textbook GPIB simple thing implementation, it's years of iterations, debug, improvement.
Ooooo, and it supports secondary addressing? NONE of the low-cost interfaces support secondary addressing! (At least last time I looked around!)
That's a big deal for those of us with R&S CMU200's / CRTU's, which require secondary addressing to make it act like a spec-an with tracking generator. And I see that in your list of tested equipment!
Fantastic. And the price of prebuilt units is super reasonable. I'm in.
As I have a CMU200 myself I was majorly interested in getting it to work too :-)
You are correct. UsbTmc exposes no multi device capability. Also when I made first versions many years ago I had no "garden hose" GPIB cables and saw they were more expensive as the whole adapter. V3 which will come early next year supports GPIB daisy chaining/multi device support via Ethernet VXI11 & HiSlip. The V2 adapter shown here supports trigger, SRQ, serial poll, trigger, clear, local lockout, goto lockal, pulse indicator, all features though. If you absolutely need common synchronous trigger to a chain of GPIB instruments which is quite seldomly absolutely required, you'll have it in next version V3.
Great work making this available for sale. NI and Keysight are the only two traditional vendors of this kind of thing and they’re priced close to $1k these days. There are tons of knock-offs on Ebay that won’t last more than a few months if they ever work.
Even worse, the windows drivers know they're knock offs and let you use them for a bit and then error out. It's the perfect lesson. They give you that high of "yes, I got one over on the big guys, we're cruising now" and then "I don't know boss it's just all of a sudden not working anymore. Yes I know we're in a production crunch. Yes we should've just bought the real one". On Linux though, no issues for me
FWIW, You can buy legit used 82357s for quite a bit less than 1k. Anecdotally, I've never had one fail in 20 years. Probably bought a dozen over that time frame. All used daily.
Yes, it is easier to fall into the trap of getting a clone, even if you think you buy an original one.
used NI-GPIB on USB cost $100 on ebay. You don’t need $1000.
See knock-offs comment.
I wish AVR DU series had any sort of open source support, we could finally move on from 32U4.
https://www.microchip.com/en-us/products/microcontrollers/8-...
The step to go to ARM or risc v based controllers is easy. That's what I did with V3. Gives a tremendous performance boost and costs 1/2 of that old AVR.
the pic16/18 offerings with usb peripherals is much better tbh
If you control your application stack, you can pivot from expensive NI GPIB hardware to something cheaper. Besides prices, you get the advantage of not requiring a 1.5GB Windows-only NI MAX install on a PC just to connect to the device.
Since I hadn't heard of open source hardware, I bought 5 of these [0] instead around 2020 and they've been reliable for daily lab use. You just connect to them over TCP and send a couple of commands to select the GPIB addr to correspond with; after that, you can send SCPI commands the same way you always do.
[0]: https://prologix.biz/
A good way around application overloaded applocation stacks is R&S Visa which is lean and mean while still exposing full visa to applocations like labview. Or going with something like python usbtmc. If you also have other instruments to control e.g. with usb or ethernet you'll likely anyway look into a visa.
Works even well on Mac and Raspberry PI btw :-)
Actually world would need an open source visa stack. Right now you find many applications doing their own thing, but then not supporting vxi11, hislip or usbtmc. Visa is something which solved that problem for years already but the most standard package NI Visa is overloaded and scaring people away. For that reason I propagate the usefulness of R&S Visa a lot. Lean and mean, supporting all OS and procotocols as it is supposed to be.
Interesting. I built an AR488 a while back (https://github.com/Twilight-Logic/AR488) and it worked alright for my needs. I'd be interested to see a comparison of the two projects.
Here a comparison from a user: https://github.com/xyphro/UsbGpib/discussions/47
Ooh! The faster speeds would be very welcome, I was bothered by the slowness of the AR488 but I assumed that was just how GPIB was (I had no baseline to compare to). I'll switch over when I get a chance.
That user's project also looks very interesting - My TDS684A's CRT seems to have died, and rather than fix it I could switch to using a software scope.
The upcoming V3 adapter reaches even more than 1 MByte/s with my fastest instruments. Unless you connect long GPIB cables to it, because capacitive load slows down GPIB as it self regulates speed down in the way GPIB is designed.
Am I wrong in understanding that each physical device supports only a single GPIB device at a time? The enumeration and discovery scheme is certainly novel, but one device per hardware would seem to prevent using some of GPIB’s advanced features like parallel polling and low latency triggers.
Glad to see there's a REV 3 in progress that would support ethernet. That's the one thing that would make me go out of my way to build one of these.
Isn't this a better link? https://github.com/xyphro/UsbGpib/blob/9a8c18a7be1b17127496e...
At least it renders the Markdown.
I cut my teeth on IEE-488. I didn’t realize that it was still a thing.
Linux mainline kernel just had support for GPIB added. https://hackaday.com/2025/12/16/after-decades-linux-finally-...
Linux gpib also supports my adapter natively with a special linux gpib firmware you can download from the page above. It also supports multiple instruments connected to one adapter then.
I finished my Master's a couple years ago, and in our IC probing lab, it was almost all GPIB. All of the pieces of test equipment, except for a single cutting-edge Keysight scope (which used Ethernet instead), were connected together via those chunky but satisfying and robust cables, usually controlled via MATLAB or some other IC measurement software package.
Yes, those cables are often called "garden hose cables". The only benefit of that bus infrastructure under a nowadays view is that it gives the ability to synchronously trigger multple instruments which is very rarely used or required.
There’s a fairly large second-hand market for old electronics test equipment. A lot of that stuff only had GPIB ports, even well into the 90s. Cheap Chinese entrants (Rigol, Siglent, etc.) only within the last decade started making an impact
I've worked places where the paperwork to hook something up via Ethernet drove us to use GPIB or RS485 for everything.
I know that very well. Ethernet=security concern. Connecting your shiny new scope to company network=no way. Hard to discuss arround it in company environments or push for split network topologies.
Crazy to think that the first HP9000 (PARISC) computers used GP-IB (aka HP-IB) for accessing disks!
I used one of those (and an even older little workstation that I forget the name) to program GPIB ("HPIB") systems.
This is my first ever engineering project, as a newly-minted electrical engineer (downloads a PDF): https://littlegreenviper.com/wp-content/uploads/2022/07/TF30...
Maybe someonw can explain me, but I never understood the appeal of GPIB for modern instruments (legacy instruments are of course "excused"). Electrically is a terrible interface that introduces ground loops with the control computer. Speed are laughable and it requires exensive and exotic adapters with complex sw stack (I wish this projects good success, it's needed!). Ethernet in comparison tick all my boxes. It's electrically decoupled by default (just use UTP cables), crazy cheap, very fast and with sane sw stack thanks to vxi-11. You can even bypass visa if you wish and open a plain TCP socket, no need for any library. What am I missing?
> What am I missing?
Not much, but consider latency: You can use the Group Execute Trigger (GET) to simultaneously trigger multiple instruments with both very low latency and very low latency dispersion. Think, easy-to-use sub-microsecond synchronization.
Ethernet and USB 4 may have orders of magnitude more bandwidth, but can’t achieve the same multi-device synchronization capability without side channel signals.
Now, sure, you can add the same capability with a programmable pulse generator connected via coax to the trigger input of all your instruments, but GBIP lets you do that with just the data connection (and you don’t always have a spare trigger channel). The only other protocols I know of with similar capabilities are PXI and PXIe, which are “PCI(express) in an incompatible form-factor, plus some extra signals for real time synchronization”.
sub-microsecond triggering should be doable with a level-2 cut-through switch and an ethernet broadcast no? I admit that ethernet is not really designed for that as the Phy is then the latency bottleneck.
Sure, in principle, but that takes effort and special equipment to set up. The point is that GPIB makes it easy (trivial, actually) with nothing more than the cables you normally use to connect instruments to get very low and predictable latency.
GPIB GET works by first configuring a subset of bus devices as listeners and then sending a single-byte message (it’s an 8 bit bus, so one bus cycle) with the ATN line asserted. It’s intrinsically low latency without any special effort.
Whether that makes it worthwhile to put GPIB on a new instrument in 2025 is a different question. I’m only addressing “what does GPIB give you”?
There are approaches to real-time ethernet (some industry implementations like profinet or ethercat, 802.1as from IEEE) but support is spotty and it requires specialized gear to be effective.
If you buy used equipment which doesn't have Ethernet or your company wants you to ise the stuff that is in the Lab since 10+ years there's simply no other choice. Or companies that see Ethernet as a potential security attack vector. It's indeed not that GPIB is better than Ethernet. In tiny aspects that's argueable, but as general statement true.
> Or companies that see Ethernet as a potential security attack vector.
It's sad how true this is. VXI gets corporate IT all prickly. An airgapped lab network would be safer but somehow they hate that idea even more.
GPIB isn't even on their radar. Test to your heart's content. It's not a "network".
Nothing, other than perhaps just much high-quality legacy equipment is out in the field that still works fantastically.
very impressive. better than anything on the market either NI or Keysight.
I strongly suspect that this does not meet the strict timing requirements that GPIB has. Putting this on your bus is likely to violate both the T1 hard timing requirements and the impedance requirements.
Use of the "standard" set of 74-series buffers that everybody uses would meet impedance requirements and would also allow the usage of a much faster MCU which likely could be made to adhere to the strict T1 timing requirements (with the caveat that most microcontroller USB stacks are piles of garbage that demand that they get interrupt priority even when you tell them otherwise).
For a situation where the instrument is directly connected to the adapter like here there is no difficulty in fast T1 timing on a MCU. V3 will have btw. fully compliant GPIB electrical specs without 74 series buffers. 48mA drivers are not required for a direct connection scenario as V2 supports. VIH is a bit violated, though not practically meaningful. So far I heard of no device not working after over 6 years and those who did not in early stages I managed to get running with sw updates. This is not a one shot plain textbook GPIB simple thing implementation, it's years of iterations, debug, improvement.