Test Equipment Anonymous (TEA) group therapy thread

[Saskia]

Anybody got a manual for an Anritsu MS420k ?
Thanx

[MathWizard]

Why doesn't a company like Tekronix, also make DMMs ? (or do they ?)

If the HPAK 34401A is so good and popular, why is there "bugger all" on utube about it, or other 6.5d DMM's ?

I suppose if I looked for each brand/model I'd find a lot more videos, but overall, there's not much on utube about them in general it seems

Keithley is owned by Tektronix. The 34401A is an older model so not too much is on Youtube, although I must admit there's not a lot on the DMM6500 either. Other than Shahriar's excellent review of course.

Right I forgot I learned that recently

[ch_scr]

Anybody got a manual for an Anritsu MS420k ?
Thanx

KO4BB has them: http://ftb.ko4bb.com/getsimple/index.php?id=manuals&dir=Anritsu-Wiltron
Seems like "K" is just the 50 Ohm version, so the generic manual fits both.

[tggzzz]

It’s performance per watt and scalability that concerns me. ARM far outpaces everything on both. Intel runs out of steam around 24 cores a partition meaning you have to go NUMA. ARM is flat to 128 cores.

https://en.wikipedia.org/wiki/Fugaku_(supercomputer)

Yes that’s ARM. Right at the top of the top 100 supercomputer list.

Last millenium I remember seeing predictions MIPS/W would be a key consideration for the HPC mob. That was one of the reasons I was never very interested in the Itanium: their strategy was to waste power on calculations that would be discarded because they turned out to be wrong.

The HPC mob are strange, and push computing in all sorts of ways, and eventually conventional computing catches up. They achieve throughput (not speed) by having embarassingly parallel problem sets, and it will be a long time before desktop computing is like that. A core prerequisite is applications and parts of the operating system being (re)coded in languages which presume multicore parallelism - and that excludes C/C++.

Having said that, maybe Apples' tight control of the hardware and software will enable their compilers and languages to be more efficient than the PC world jungle.

[BU508A]

BU508A, I like what you offer but the last DMM has recently suffered the EEVBlog youtube effect, sellers has raised these prices from cheap to beyond what 5.5 should cost. I have a nice source of HP 3478A from an Canadian seller for decent price and it's tempting. Also, I should think perusing each canadian seller who sells TE to find any gems they might had it like my Fluke 8502A which I can repair once I get my DE-5000 LCR meter.


Currently, I am also thinking of purchasing 34401A due their great linearity and solid retention of drift to a point it can be considered as 1-year speced meter or also Keithley. My budget for this meter is $600

And I am watching the video on this topic
https://www.youtube.com/watch?v=I2HcJ80OyrA

Well, you've mentioned 5.5 digit DMM. I knew about this EVBlog effect but sometimes one can find a 8842A for a reasonable price.
And because I know of this EEVBlog effect, that's why I've suggested a 8520A. They are usually cheaper than the 8842A. I own one and I like it.

If it comes to 6.5 DMM there are mainly three meters which I'd recommend:

- the 34401A
- the 3456A
- the 3457A (I don't like the LCD and I would modify it for sure. The same goes for the 3478A)

[tggzzz]

FPGA is interesting. Perhaps. But then again it’s difficult to configure for workloads and the core density and power consumption isn’t that great either. ASICs are still way cheaper if you constrain and classify workloads.

There are many counterexamples to using FPGAs, in general purpose computing at least.

Start by remembering the experiences of the Cell processors, which had one big processor and hald a dozen small satellite processors. Even for just games consoles, they were a pain to program.

Then go back and look at re-microcodeable machines such as the Research Machines Orion. Nobody could be bothered to fiddle with the microcode, because there were too many practical limitations, e.g. having microcode for C and Pascal and X all resident simultaneously.

One thing that is true is x86 is a dead end for sure at this point. If you look at some of the ARM vs x86-64 compiler optimisations possible it’s pretty scary what ARM can do per cycle. I was throwing stuff in godbolt (thanks Cerebus) and it was flattening one of my loops to a one liner on ARM but 14 instructions on x86. I looked at the clock timings and ARM did it in 8 clocks as well. X86 I gave up on working it out as the damn thing has different cycle counts per address mode per instruction. Ugh. Kill me.

All valid - and Nvidia is now a bigger company than Intel!

One of the x86's disadvantages is that there are so many ISA variants; unless you ship source code, you have to make a stab at which one(s) exist in themachine running your library code. The linux mob are beginning to consider defining subsets of the ISAs, and trying to deal with those https://www.phoronix.com/scan.php?page=news_item&px=Linux-x86-64-Feature-Levels

[tggzzz]

After some PM messages with Specmaster, I came up at my conclusion that I need make the DIY reference box of and of voltage and I must order a new DMM at least of 5.5 digits and it could come with certs to help "transfer" the standard from it to my HP 3478A. It should be good enough for hobbyists, I just don't become an voltnut when I am supposed to make projects

Do examine all the posts on those subjects in the Metrology sub-forum. As you are a voltnut, you will need to understand the subtle points, e.g. the units used to express the drift of voltage sources both inside and outside DMMs.

Have fun, and welcome to the rathole. There's a standard picture used to illustrate that in this thread

[tggzzz]

Edit: anyway it’s 2am. I’m going to bed 

An hour later I woke myself up, which nowadays requires specialised equipment.

Why? To go over the hill and far away and see this:



That's the first time I've seen noctilucent clouds. Apparently they are unusually common at the moment, for unknown reasons.

And the comet Neowise is just visible, if you look carefully and zoom into above the bush near the guy line. It is the first I have seen in 34 years. (Yes bd, the fens are ideal for skywatching since they have lots of it and relatively little light.)

Next visible, assuming clear skies and that it survives perihelion, 11-23rd July in the night sky.

[mansaxel]

Still having no luck on eBay with TEA, but I sure am having success with tools.

<snip>

A Mitutoyo depth gauge with a full set of rods for 0-150mm (0.01mm resolution). Posted as "Garage Clearout ... This was my father in laws - I’m not too sure if it’s ever been used. Some of the protective sponge has started to deteriorate a bit.".

I managed to snatch a standard Mitutoyo caliper for my wife at something like 20€ and when I went back online to grab one for myself too, the market seemed to have dried up.  Just my luck.

But, I never get loud protests when I buy something, even though that Solartron's been sitting inside the house for like two weeks while I'm procrastinating a mains plug search.

[mendip_discovery]

They even provide APIs now!

I still use pencil and paper when planning world domination. Take that suckers.

hmm, bd139 has just ordered 5 notebooks and 200 pencils. Better keep an eye on him after what happened last time.

[AVGresponding]

Will somebody please save me from myself. I've got the money but not the time nor the space. I already have £0.99 nixies that I picked up about 15 minutes away from this vendor.

https://www.ebay.co.uk/itm/Solartron-Digital-Voltmeters-A203-A202/153999947342

Bah, collection only? Best way to get less for it than it's worth. How hard is it to bung it in a cardboard box with some bubble wrap?
This HP system psu will go for silly low money for the same reason:
https://www.ebay.co.uk/itm/HP-agilent-keysight-6624A-system-power-supply/303614195440

BU508A, I like what you offer but the last DMM has recently suffered the EEVBlog youtube effect, sellers has raised these prices from cheap to beyond what 5.5 should cost. I have a nice source of HP 3478A from an Canadian seller for decent price and it's tempting. Also, I should think perusing each canadian seller who sells TE to find any gems they might had it like my Fluke 8502A which I can repair once I get my DE-5000 LCR meter.


Currently, I am also thinking of purchasing 34401A due their great linearity and solid retention of drift to a point it can be considered as 1-year speced meter or also Keithley. My budget for this meter is $600

And I am watching the video on this topic
https://www.youtube.com/watch?v=I2HcJ80OyrA

Well, you've mentioned 5.5 digit DMM. I knew about this EVBlog effect but sometimes one can find a 8842A for a reasonable price.
And because I know of this EEVBlog effect, that's why I've suggested a 8520A. They are usually cheaper than the 8842A. I own one and I like it.

If it comes to 6.5 DMM there are mainly three meters which I'd recommend:

- the 34401A
- the 3456A
- the 3457A (I don't like the LCD and I would modify it for sure. The same goes for the 3478A)

Though I really like my Fluke 8840A (same as 8842 but with 0.005% instead of 0.0035% basic accuracy), be aware it does NOT have a uA range, so low current measurements are out the window. Your LSD is 10uA on the lowest current range.
Out of my entire meter collection, the only 5.5 digit one with decent low current capability is the Thurlby 1905a, with a 200uA range and 1nA resolution.



Here's a nice calculator for the flyers amongst us:
https://www.ebay.co.uk/itm/Heathkit-Model-OCW-1401-Aircraft-Navigation-Computer/293637935305

[bd139]

Well it's not strictly true that ARM sit at the top of the supercomputer CPU list as that A64FX CPU was a collaboration between Fujitsu and ARM.

https://en.wikipedia.org/wiki/Fujitsu_A64FX

I have a pretty powerful iPad but find that whole user experience is still somewhat lacking in comparison to my AMD 8350FX powered desktop PC. IMO there is still much to be done before a tablet regardless of who makes it and what the CPU is that powers it, can really rival a proper decent desktop.

You can connect a 4K display, keyboard, mouse and external storage to the iPads now. And multitask effectively. Killer for me is I’ve got zoom running on a conference call, taking notes with goodnotes/pencil and dealing with email at the same time. Plus also you can use it for live annotation working with the desktop.

I agree really but it’s 90% of the way there now. Enough to replace my laptop. But not the desktop.

[mansaxel]

No, these latest problem children are FTTH (Fiber To The Home) gateways.

Oh, APC connector (green shell SC connector). The only places you'll see them are over-engineering situations or cable TV networks. I've got the wonky adapters to 'scope them and we've got some at work, (including on one of our OTDR machines) but generally, I try to avoid them.

Do you have an idea whether the box is "just" an Ethernet device or does it do some PON on the WAN side? Since it's apparently is only one fiber (implying BiDirectional WDM) I'd be tempted to suspect Ethernet, but...

[mansaxel]

The HPC mob are strange, and push computing in all sorts of ways, and eventually conventional computing catches up. They achieve throughput (not speed) by having embarassingly parallel problem sets, and it will be a long time before desktop computing is like that. A core prerequisite is applications and parts of the operating system being (re)coded in languages which presume multicore parallelism - and that excludes C/C++.

They usually write things in FORTRAN. Or, perhaps Java now. When I worked at the HPC centre in Stockholm, we had one computer whose work was limited by the network bandwidth between 2 universities, because it sat on one of them, but the user base, and the home directories (AFS) they used to post their work in were on the other. Their driver scripts usually were in Perl.

Also, they build machines (ie. clusters) in different ways depending on workload. A task that can be easily subdivided into monotonic calculations (SETIatHome stuff) is easy on the interconnect between computers, but things like fluid dynamics or combustion chamber modelling (diesel engines) consist of calculations whose intermediate results influence and are influenced by other parts of the job. The partitioning into parallel tasks is done in a geometrical grid over the space modelled ("You shall make computations on the pressure/temperature/turbulence/whatever on this mm²"), and thus interdependences are created.

The former type of workload is easily solved with 10G Ethernet, which was 1G Ethernet when I worked there, but the latter needs lower latency still in interconnect. Back then Infiniband was very much popular. Last computer they bought is a Cray with a proprietary interconnect, but they're looking for a new one now. As can be expected, it is not so much the bandwidth but the latency that are the crucial specs.  A bad case for the "C5 Galaxy full of tapes" type of transport.

[Specmaster]

After some PM messages with Specmaster, I came up at my conclusion that I need make the DIY reference box of and of voltage and I must order a new DMM at least of 5.5 digits and it could come with certs to help "transfer" the standard from it to my HP 3478A. It should be good enough for hobbyists, I just don't become an voltnut when I am supposed to make projects

Thus, for hobbyists they dont need traceable certs except when starting with one used bench DMM which I believe is worth to get adjusted to any labs, otherwise a new bench DMM is enough. 
Any suggestions for 2nd New DMM ? Fluke 8805A is a bust because the damned seller did not reply to me even I did check his profile saying he sometimes logs in.

You could keep an eye open for a Solartron 7150 or a 7150plus, both are 6.5 digits meters but often sell on eBay for reasonable money because the LCD screen lighting goes dark after a while but in reality it is fairly easy to replace the LED's, I did that with my two. I used ultra bright LEDs as you can see in the photo makes them very readable indeed. You can also clearly see in the photo what I was saying about the last digits can fluctuate slightly and for the hobbyist does 1mV really matter, I suggest not. The photo shows a selection of my meters, all connected a AD584-M voltage reference set at 10V and shows my 3478A after I calibrated it myself.

[tonyalbus]

After some PM messages with Specmaster, I came up at my conclusion that I need make the DIY reference box of and of voltage and I must order a new DMM at least of 5.5 digits and it could come with certs to help "transfer" the standard from it to my HP 3478A. It should be good enough for hobbyists, I just don't become an voltnut when I am supposed to make projects

Thus, for hobbyists they dont need traceable certs except when starting with one used bench DMM which I believe is worth to get adjusted to any labs, otherwise a new bench DMM is enough. 
Any suggestions for 2nd New DMM ? Fluke 8805A is a bust because the damned seller did not reply to me even I did check his profile saying he sometimes logs in.

You could keep an eye open for a Solartron 7150 or a 7150plus, both are 6.5 digits meters but often sell on eBay for reasonable money because the LCD screen lighting goes dark after a while but in reality it is fairly easy to replace the LED's, I did that with my two. I used ultra bright LEDs as you can see in the photo makes them very readable indeed. You can also clearly see in the photo what I was saying about the last digits can fluctuate slightly and for the hobbyist does 1mV really matter, I suggest not. The photo shows a selection of my meters, all connected a AD584-M voltage reference set at 10V and shows my 3478A after I calibrated it myself.

very, very nice collection Spec!

[tonyalbus]

@specmaster  Thanks received your SM and OM for the HP!

[Specmaster]

After some PM messages with Specmaster, I came up at my conclusion that I need make the DIY reference box of and of voltage and I must order a new DMM at least of 5.5 digits and it could come with certs to help "transfer" the standard from it to my HP 3478A. It should be good enough for hobbyists, I just don't become an voltnut when I am supposed to make projects

Thus, for hobbyists they dont need traceable certs except when starting with one used bench DMM which I believe is worth to get adjusted to any labs, otherwise a new bench DMM is enough. 
Any suggestions for 2nd New DMM ? Fluke 8805A is a bust because the damned seller did not reply to me even I did check his profile saying he sometimes logs in.

You could keep an eye open for a Solartron 7150 or a 7150plus, both are 6.5 digits meters but often sell on eBay for reasonable money because the LCD screen lighting goes dark after a while but in reality it is fairly easy to replace the LED's, I did that with my two. I used ultra bright LEDs as you can see in the photo makes them very readable indeed. You can also clearly see in the photo what I was saying about the last digits can fluctuate slightly and for the hobbyist does 1mV really matter, I suggest not. The photo shows a selection of my meters, all connected a AD584-M voltage reference set at 10V and shows my 3478A after I calibrated it myself.

very, very nice collection Spec!

That is now out of date, this is my new collection of bench meters and now no 3478A or 3466A and have been replaced by Thurlby 1905A 6.5 and a Black Star 5.5 digit meters.

Actually this is a good comparison of the new LEDs fitted to the screens on the 7150 and the before photo with the poor screen lighting, chalk and cheese.   

[Specmaster]

I have always wanted to get myself a reasonable ultrasonic cleaner that was large enough to take a PCB of a reasonable size and I have been jockeying with an eBay seller for a while trying to get 2 Litre one that was big enough at an affordable price. I make an offer, and he kept counter offering, so I decided to wait awhile. He recently dropped the prices to marginally above my best offer, so I struck, but me, being me, still went for an offer and this morning he accepted it and now I'm eagerly awaiting the arrival of this magnificent beast which boasts a tank size of 240 x 135 x 100 which will accommodate the PCB from the 4503, half at a time so it would require 2 passes but so much more efficient at cleaning than just scrubbing with a toothbrush and bending everything to get down to board level in places.

[tggzzz]

The HPC mob are strange, and push computing in all sorts of ways, and eventually conventional computing catches up. They achieve throughput (not speed) by having embarassingly parallel problem sets, and it will be a long time before desktop computing is like that. A core prerequisite is applications and parts of the operating system being (re)coded in languages which presume multicore parallelism - and that excludes C/C++.

They usually write things in FORTRAN. Or, perhaps Java now.

Fortran is standard, but I doubt Java is getting used for any core computation. Java does have a performance hit, albeit smaller than most people imagine. The HPC mob are very sensitive to performance

Also, they build machines (ie. clusters) in different ways depending on workload. A task that can be easily subdivided into monotonic calculations (SETIatHome stuff) is easy on the interconnect between computers, but things like fluid dynamics or combustion chamber modelling (diesel engines) consist of calculations whose intermediate results influence and are influenced by other parts of the job. The partitioning into parallel tasks is done in a geometrical grid over the space modelled ("You shall make computations on the pressure/temperature/turbulence/whatever on this mm²"), and thus interdependences are created.

The former type of workload is easily solved with 10G Ethernet, which was 1G Ethernet when I worked there, but the latter needs lower latency still in interconnect. Back then Infiniband was very much popular. Last computer they bought is a Cray with a proprietary interconnect, but they're looking for a new one now. As can be expected, it is not so much the bandwidth but the latency that are the crucial specs.  A bad case for the "C5 Galaxy full of tapes" type of transport.

Yes.

The key issue is the balance between communication and computation. That's true at many levels, e.g. the one you mention, and also even within multicore processors with their L1/2/3/DRAM memory hierarchy.

[tonyalbus]

I have always wanted to get myself a reasonable ultrasonic cleaner that was large enough to take a PCB of a reasonable size and I have been jockeying with an eBay seller for a while trying to get 2 Litre one that was big enough at an affordable price. I make an offer, and he kept counter offering, so I decided to wait awhile. He recently dropped the prices to marginally above my best offer, so I struck, but me, being me, still went for an offer and this morning he accepted it and now I'm eagerly awaiting the arrival of this magnificent beast which boasts a tank size of 240 x 135 x 100 which will accommodate the PCB from the 4503, half at a time so it would require 2 passes but so much more efficient at cleaning than just scrubbing with a toothbrush and bending everything to get down to board level in places.

 

[Robert763]

Another meter that is OK for general bench use is the Solartron 7150 / 7150 Plus. While often sold as 6.5 digit, and capable of displaying 6.5 the underlying specifications are that of a good 5.5 digit (0.002% +5 counts on DC) and was sold as 5.5 digit. Most have GPIB. Lot's in the UK as the military used them as their high accuracy meter for a long ime. Mod's to replace the dim yellow LED backlight with white and fit shrouded sockets are easy enough. Ebay prices can be high but at least one UK dealer has them on offer at £80-£90 inc VAT. North American ebay prices tend to be lower.

[Specmaster]

Just updated the 4503 meter with shrouded input jacks, taking great care not to damage the membrane on the front panel. The new sockets almost alliterate the word Max on the current jack but it is clear what it says even though.

[Kosmic]

Start by remembering the experiences of the Cell processors, which had one big processor and hald a dozen small satellite processors. Even for just games consoles, they were a pain to program.

The cell processor was an interesting architecture but was somehow a specialized device (targeting batch processing mostly). I'm not sure how and why someone decided to use the chip for their next console but it was completely out of place and misadapted. Fun fact, the main and satellite processors were 2 different type of cpu, so you had to compile your code in 2 pass.

[Saskia]

no discord tonite, it is no fun on an edge network.