Test Equipment Anonymous (TEA) group therapy thread

[0culus]

Wanted to call everyone's attention to this thread: https://www.eevblog.com/forum/projects/challenge-thread-the-fastest-breadboard-oscillator-on-the-mudball/

I got started by breadboarding a 2N2222 based Colpitts oscillator (shown in the thread), and after tweaking it have broken 20 MHz with about .5Vpp output. Looking forward to trying more designs...although I doubt I'll soon break joeqsmith's current record of over 13 GHz.  Even so, it's a lot of fun!

Put that TEA to use!! 

[Saskia]

how do you breadboard a BGA ...

hmm

as for the computing stuff I think FPGAs are a relatively cheap way to scale up in specialized applications.

too much stuff to learn. Bummer.

[Neomys Sapiens]

how do you breadboard a BGA ...

hmm

as for the computing stuff I think FPGAs are a relatively cheap way to scale up in specialized applications.

too much stuff to learn. Bummer.

What, why would you need to do that? There are premade FPGA experimental boards as well as some that can be plugged into a classic breadboard. Or did I miss something?

[tggzzz]

Bus throughput is already there. Also ARM scales up in cores and latency much much better.

Read this: https://www.anandtech.com/show/15578/cloud-clash-amazon-graviton2-arm-against-intel-and-amd/3

I think you discount PCIE 4.0/Zen2 too much; people have been making a living doing this for almost a year now and they are soon going to be in for a rude awakening as ground-up PCIE4.0 expansion hardware designs actually hit the market. GPU/CPU distributed processing is a huge advantage in content creation/rendering, so the external bus is still as important or more so than the internal mesh/memory bus. Not saying Neoverse memory architecture isn't equivalent (on paper)... but so far that's it. EQUIVALENT, and then ONLY OCed like fuck to get the mesh up to ~2TB/s. Not earth-shakingly better. It will be interesting to see how dual 128-bit stacks up against native 256-bit tho.   

BUTT... and this is a big butt... I'm really not sure how much of this recent server-centric explosion in hardware is going to trickle down to even high-end user workstations; I expect there will be HUGE pressure to follow Intel's example WRT new tech and "keep it in the priesthood" for probably as long as we're both still on the right side of the grass & flogging bits.

mnem
 

Add per-watt to everything and that’s where the money is. PCIE4.0 is nice but it’s really edge case stuff that benefits. Less watts = higher density. NUC with an EPYC in it on your desktop is what we’re talking here. Or a 42U compute rack with 6x the compute in it for the same power.

One killer for me is lots of cores. I work on stuff that is highly parallel and x86-64 craps our fairly quickly on core scalability. And also NUMA is a mess on x86-64. Those problems are gone on ARM.

Whatever the ISA, memory bandwidth becomes the issue.

Sun's Niagara processors had a good take on that for "embarrassingly parallel" workloads. Match the core cycle time to the memory, don't have out of order stuff, and use the space saved for more cores.

[bd139]

Or have a lot of I and D cache per core and keep your cache line optimisation as tight as a duck’s arse

Latency is where it all goes to shit. Not necessarily throughput. Cores stalled on cache reads are useless.

We had some matching stuff which was fun. 3x the throughput with half the cores just by mapping the streams and using very local state. Aka why C still wins. The C# stuff is horrible for this.

[tggzzz]

Or have a lot of I and D cache per core and keep your cache line optimisation as tight as a duck’s arse

Latency is where it all goes to shit. Not necessarily throughput. Cores stalled on cache reads are useless.

We had some matching stuff which was fun. 3x the throughput with half the cores just by mapping the streams and using very local state. Aka why C still wins. The C# stuff is horrible for this.

Latency is a key issue, and caches only help up to a point - or rather a cliff. That the neat point about Niagara, a core's cycle time was tied to memory latency. People were gobsmacked at the performance I got out of a telecom Java application sized to be one thread per core.

[worsthorse]

I've decided I am hankering for a logic analyzer because, well, why not? I already own a tek 308 and a tek 338 with all the probes, etc but they were made for a slower, kinder era when 20MHz was RFF¹ and bus decoding wasn't often required.  Though I could buy some new-fangled USB-based LA the size of a pack of cards, where's the fun in that?

So... LA fans, what sort of box should I consider getting? Figuring that while faster is better, I can probably do with something of middling speed and I probably don' t need sixty-four channels.  For bonus points, tell me whether getting the probes and lead sets for your favorite is going to cost me four times as much as the analyzer.


¹Really Freakin' Fast

[bd139]

Or have a lot of I and D cache per core and keep your cache line optimisation as tight as a duck’s arse

Latency is where it all goes to shit. Not necessarily throughput. Cores stalled on cache reads are useless.

We had some matching stuff which was fun. 3x the throughput with half the cores just by mapping the streams and using very local state. Aka why C still wins. The C# stuff is horrible for this.

Latency is a key issue, and caches only help up to a point - or rather a cliff. That the neat point about Niagara, a core's cycle time was tied to memory latency. People were gobsmacked at the performance I got out of a telecom Java application sized to be one thread per core.

Similar to what I was doing with pthread_setaffinity_np

Interestingly someone then migrated it to a NUMA virtual machine. That was hilarious.

[Saskia]

@wh
HP 16702 B ?
It's your boat anchor ...

[tggzzz]

I've decided I am hankering for a logic analyzer because, well, why not? I already own a tek 308 and a tek 338 with all the probes, etc but they were made for a slower, kinder era when 20MHz was RFF¹ and bus decoding wasn't often required.  Though I could buy some new-fangled USB-based LA the size of a pack of cards, where's the fun in that?

So... LA fans, what sort of box should I consider getting? Figuring that while faster is better, I can probably do with something of middling speed and I probably don' t need sixty-four channels.  For bonus points, tell me whether getting the probes and lead sets for your favorite is going to cost me four times as much as the analyzer.

Probes typically come in two parts. A fancy ribbon cables with around 8 signals and ground, and short grabbers.

You must get the right fancy ribbon cables. The grabbers are around £1-£2 each. I marginally bleeder the Tek grabbers to the HP grabbers.

Speed is important, but not everything. Just as important are the triggering and filtering capabilities. If you are probing a synchronous system, the LA treats a clock as the same as a signal, so to detect the clock reduces the effective speed by at least a factor of 2, realistically 4.

The arming and triggering capabilities can be complex, so an appropriate GUI is highly beneficial.

[worsthorse]

I've decided I am hankering for a logic analyzer because, well, why not? I already own a tek 308 and a tek 338 with all the probes, etc but they were made for a slower, kinder era when 20MHz was RFF¹ and bus decoding wasn't often required.  Though I could buy some new-fangled USB-based LA the size of a pack of cards, where's the fun in that?

So... LA fans, what sort of box should I consider getting? Figuring that while faster is better, I can probably do with something of middling speed and I probably don' t need sixty-four channels.  For bonus points, tell me whether getting the probes and lead sets for your favorite is going to cost me four times as much as the analyzer.

Probes typically come in two parts. A fancy ribbon cables with around 8 signals and ground, and short grabbers.

You must get the right fancy ribbon cables. The grabbers are around £1-£2 each. I marginally bleeder the Tek grabbers to the HP grabbers.

Speed is important, but not everything. Just as important are the triggering and filtering capabilities. If you are probing a synchronous system, the LA treats a clock as the same as a signal, so to detect the clock reduces the effective speed by at least a factor of 2, realistically 4.

The arming and triggering capabilities can be complex, so an appropriate GUI is highly beneficial.

Yeah, I am inclined to get something Tekish that uses all the flying lead sets I accumulated for the 308/338 LAs.  At the low end the 1240/1241 might be an option.  In HP land, the 1562B might be another.   Both are definitely boat anchors. The 16702B? Maybe though I'd have to spend some time figuring out the options/add-ons to make a usable system. Same goes for the TLA714. 

For what I am likely to use it for, synchronous operation is the most common scenario.

[0culus]

@worsthorse: I have a Tek TLA715 that has two 128 channel plugins. I only have two probe cables however, because those are the most expensive part. It was more than sufficient for my purposes and it saved my bacon when I was working on my MS after my design broke Vivado's internal logic analyzer IP and I had to instead route signals of interest to board pins and use the TLA715. There's a guy on ebay who I got mine from, who refurbishes them with a compact flash card instead of the spinning disk and installs XP professional which is the newest Windows supported.

As the above post points out, LA triggering isn't really trivial. You'll want to sit down and really understand how it all works. At the end of the day, the triggers in most modern LAs are really just a user-configurable state machine. It's quite powerful.

[worsthorse]

@worsthorse: I have a Tek TLA715 that has two 128 channel plugins. I only have two probe cables however, because those are the most expensive part. It was more than sufficient for my purposes and it saved my bacon when I was working on my MS after my design broke Vivado's internal logic analyzer IP and I had to instead route signals of interest to board pins and use the TLA715. There's a guy on ebay who I got mine from, who refurbishes them with a compact flash card instead of the spinning disk and installs XP professional which is the newest Windows supported.

As the above post points out, LA triggering isn't really trivial. You'll want to sit down and really understand how it all works. At the end of the day, the triggers in most modern LAs are really just a user-configurable state machine. It's quite powerful.

Good info, thanks.

While it has been a long time, I remember how complicated triggering could could get. In my first design job, I used a LA way more than an oscilloscope as I was designing FSMs and 2900 bit slice processor-based systems.  The software tools were very primitive and a logic analyzer saved my butt way more than one time.

EDIT: @oculus, if you have a chance, PM me the name of that seller. thanks!

[tggzzz]

I've decided I am hankering for a logic analyzer because, well, why not? I already own a tek 308 and a tek 338 with all the probes, etc but they were made for a slower, kinder era when 20MHz was RFF¹ and bus decoding wasn't often required.  Though I could buy some new-fangled USB-based LA the size of a pack of cards, where's the fun in that?

So... LA fans, what sort of box should I consider getting? Figuring that while faster is better, I can probably do with something of middling speed and I probably don' t need sixty-four channels.  For bonus points, tell me whether getting the probes and lead sets for your favorite is going to cost me four times as much as the analyzer.

Probes typically come in two parts. A fancy ribbon cables with around 8 signals and ground, and short grabbers.

You must get the right fancy ribbon cables. The grabbers are around £1-£2 each. I marginally bleeder the Tek grabbers to the HP grabbers.

Speed is important, but not everything. Just as important are the triggering and filtering capabilities. If you are probing a synchronous system, the LA treats a clock as the same as a signal, so to detect the clock reduces the effective speed by at least a factor of 2, realistically 4.

The arming and triggering capabilities can be complex, so an appropriate GUI is highly beneficial.

Yeah, I am inclined to get something Tekish that uses all the flying lead sets I accumulated for the 308/338 LAs.  At the low end the 1240/1241 might be an option.  In HP land, the 1562B might be another.   Both are definitely boat anchors. The 16702B? Maybe though I'd have to spend some time figuring out the options/add-ons to make a usable system. Same goes for the TLA714. 

For what I am likely to use it for, synchronous operation is the most common scenario.

The Tek grabbers and HP grabbers are interchangeable (not the flying leads, though). Connecting them to the UUT can be a pain, of course; the Tek ones grap 1.27mm, 0.05" pitch leads. Sometimes people put a multiway connector on their UUT, and just plug the flying lead into that.

I have an Agilent 1682D, 800MHz asynchronous clock rate (IIRC with half the pins, 400MHz with all of them) so it can resolve 1.25/2.5ns intervals. It isn't a boat anchor; it is a Windows machine with an extra largePCB containing the Xilinx FPGAs.

[Neomys Sapiens]

I've decided I am hankering for a logic analyzer because, well, why not? I already own a tek 308 and a tek 338 with all the probes, etc but they were made for a slower, kinder era when 20MHz was RFF¹ and bus decoding wasn't often required.  Though I could buy some new-fangled USB-based LA the size of a pack of cards, where's the fun in that?

So... LA fans, what sort of box should I consider getting? Figuring that while faster is better, I can probably do with something of middling speed and I probably don' t need sixty-four channels.  For bonus points, tell me whether getting the probes and lead sets for your favorite is going to cost me four times as much as the analyzer.

Probes typically come in two parts. A fancy ribbon cables with around 8 signals and ground, and short grabbers.

You must get the right fancy ribbon cables. The grabbers are around £1-£2 each. I marginally bleeder the Tek grabbers to the HP grabbers.

Speed is important, but not everything. Just as important are the triggering and filtering capabilities. If you are probing a synchronous system, the LA treats a clock as the same as a signal, so to detect the clock reduces the effective speed by at least a factor of 2, realistically 4.

The arming and triggering capabilities can be complex, so an appropriate GUI is highly beneficial.

Yeah, I am inclined to get something Tekish that uses all the flying lead sets I accumulated for the 308/338 LAs.  At the low end the 1240/1241 might be an option.  In HP land, the 1562B might be another.   Both are definitely boat anchors. The 16702B? Maybe though I'd have to spend some time figuring out the options/add-ons to make a usable system. Same goes for the TLA714. 

For what I am likely to use it for, synchronous operation is the most common scenario.

I've encountered the 1241 and a TLA7xx (I don't recall the exact type). I found the 1241 to be rather self-explaining when I used it once, although this applies only to the basic machine - I'm not sure about the modules, for which not much documentation seems to exist. The TLA just struck me dumb with it's capabilities and complexity. So with a task at hand and no time to vanish for a week to play with it, I chose to use other means. The 1241 with a nice set of personality modules etc. would be something I would like to aquire too. On the other hand, it only supports chips which had been in use during its lifecycle at Tek. And maybe the capabilities of the TLA are the thing to go for - I just could not operate them on short notice. I chose to use a pattern trigger and scope and hem the conditions in by the test routine. The chip I was looking at was not a CPU but a version of the VIC and no specific support for it existed anyway.

[Neomys Sapiens]

The Tek grabbers and HP grabbers are interchangeable (not the flying leads, though). Connecting them to the UUT can be a pain, of course; the Tek ones grap 1.27mm, 0.05" pitch leads. Sometimes people put a multiway connector on their UUT, and just plug the flying lead into that.

Either that, or get one of those adaptors from Emulation Technology or likewise. Beats working with the bloody grabbers by far, especially if there are many of them.

[GreyWoolfe]

I'll see your one knob and... take that too..

I have a working one myself that I paid the princely sum of $5.00 USD.  Very interesting to operate.

[worsthorse]

I have an Agilent 1682D, 800MHz asynchronous clock rate (IIRC with half the pins, 400MHz with all of them) so it can resolve 1.25/2.5ns intervals. It isn't a boat anchor; it is a Windows machine with an extra largePCB containing the Xilinx FPGAs.

Nice box.

And, unfortunately, purchasing one would violate a cardinal rule around here: a single piece of TE cannot cost more than my truck.   

Now granted, my truck is a 2003 Ford Ranger which doesn't give me a lot of running room, but a rule is a rule. 

Will spend the day tomorrow looking at the manuals for the HP1652B and the Tek 1241 to see if either will meet my relatively simple itch for a new piece of kit.  Or mebbe the TLA715 or the HP16702.

[beanflying]

Feeding the Inner Bean. Late Shack lunch of leftovers in the fridge before going out to brave the local Shopping nuttery.  Cracked Pepper a splash of Olive Oil and STUFF.

My TEA is looking forlornly at me for a lack of usage 

[tggzzz]

The Tek grabbers and HP grabbers are interchangeable (not the flying leads, though). Connecting them to the UUT can be a pain, of course; the Tek ones grap 1.27mm, 0.05" pitch leads. Sometimes people put a multiway connector on their UUT, and just plug the flying lead into that.

Either that, or get one of those adaptors from Emulation Technology or likewise. Beats working with the bloody grabbers by far, especially if there are many of them.

My google-foo fails me there.

If that means something that clips on an IC, then they can be expensive and you have to have the right size. Fine for old DIPs, but there are many packages available now.

In many cases it will be necessary to resort to soldering on short flying wires, and clipping onto those.

The other thing to watch for in an LA is the sample depth, for the same reasons as with digitising scopes; 1024 doesn't cut it anymore.

You might benefit from protocol decodes as well, although there I usually manage with printf() statements.

I chose to wait until I found my LA at a decent price. I had an hp1630, but found it too limiting, and sold it.

[0culus]

Maybe he means something like the Pomona SOIC-Clips?

https://www.pomonaelectronics.com/products/test-clips/soic-clip-kit

I have that kit...don't need it often but it's awesome to have when you do need it. I'm sure you can get clips for other types of packages too though I imagine they aren't cheap.

[tggzzz]

Maybe he means something like the Pomona SOIC-Clips?

https://www.pomonaelectronics.com/products/test-clips/soic-clip-kit

I have that kit...don't need it often but it's awesome to have when you do need it. I'm sure you can get clips for other types of packages too though I imagine they aren't cheap.

That's the kind of thing I was thinking of, but for £87 you get one example of some IC.

For modern logic, the lead length can be an issue, too. For 1ns a rule of thumb is to keep stubs <1.5cm, and jellybean 74lvc1g* gates have transition times down to 300ps

[BU508A]

There is a nice LCR Meter from HP in the bay (NAWTS):

HP 4284A, sadly without the test-fixture box:
https://www.ebay.de/itm/1-St-HP-4284A-Precision-LCR-Meter-20Hz-1MHz-gebr-guter-Zustand-/402295151645

[bd139]

Think there's a fixture box on UK ebay. Sure I saw one the other day.

Edit: actually no ... it was with an LCR meter  https://www.ebay.co.uk/itm/203019848049

[beanflying]

Interesting Aussie stray evilbay listing. Not much info on it without a proper trawl but I gather it is an input attenuator of some sort for a HP chart plotter 

eBay auction: #184329170558