Svarichevsky Mikhail - RSS feed Svarichevsky Mikhail - RSS feed en-us Tue, 10 Jun 2006 04:00:00 GMT Wed, 24 May 17 05:09:01 +0300 120 10 <![CDATA[Venus and Saturn today]]>
Good old A77ii+Sony 500F8]]>
Tue, 18 Apr 17 05:54:03 +0300
<![CDATA[Upgrade time! LGA2011 FTW #2]]> LGA2011 with i7-3820 (4 cores, 32nm), not waiting for Ivy Bridge. Recently ebay got filled with lots of used server CPUs on LGA2011 which were compatible with my motherboard. That was extremely tempting and much more interesting than upgrading to yet another boring quad core system.

I went for E5-2695v2 - 12 cores, 30Mb of L3 cache, 22nm (Ivy Bridge). That is one of the largest commercial dies - enormous 492 mm².
This all currently costs mere 295$ (just 4 years ago Intel was charging whopping $2336 for that).

Task manager looks beautiful now:

Quick benchmark results: Surely, both old and new processors are overclocked. For E5-2695v2 overclocking is really limited by multiplier and BCLK. Stability is rock solid - I was even able to reduce core voltage quite a bit so temperatures under full load are unbelievably low for 12-core CPU (<50°C).

i7-3820 @4.3GhzE5-2695v2@2.926Ghz HT OnE5-2695v2@2.926Ghz HT Off
x264 ffmpeg, FPS5311388
Linx, GFLOPs95197.42201.2
Typically, on parallelized tasks it's >2 times faster. With this amount of cores it is no longer possible to ignore multicore support in all applications I use. I even had to upgrade some of my image processing scripts using mparallel.

Now I can work in peace until we'll get some dual-socket systems with 16-32 core CPUs and HBMs...]]>
Wed, 12 Apr 17 05:47:30 +0300
<![CDATA[A look inside Russian 28nm MIPS CPU - Baikal-T1]]> Baikal-T1 - Russian implementation of dual core Imagination Technologies MIPS P5600 32 r5 with on-board 10Gb Ethernet. Baikal was the first to implement P5600 core in silicon.

CPUs itself (had to go through 4 pieces to get a good die):

Read more on →]]>
Tue, 28 Feb 17 20:46:39 +0300
<![CDATA[Time for decent SDR: Ettus N210 - 40Mhz of RF bandwidth]]>
Although it is surreal to see 50Mhz of spectrum in realtime (40Mhz bandwidth using UBX-40 - 10-6000MHz RF frontend) - it is only available on the host in 8-bit mode (it already requires 800Mbit/s - using otw_format=sc8 in device string). In 14-bit ADC mode we can "only" fit 25Mhz through gigabit Ethernet. But if you're ready to do custom processing on internal FPGA - one can do better.

Ethernet connection is really useful - now I can easily set it away from the PC, for example on a balcony - where I already have Ethernet cabling. Also it will not pickup any noise from the ground as Ethernet is isolated.

Tue, 14 Feb 17 03:56:35 +0300
<![CDATA[Time]]> ]]> Sat, 21 Jan 17 23:54:31 +0300 <![CDATA[Olympus 10x/0.3 WI - now with water immersion]]> previous post about my microscope I've mentioned that now there are no more cheap upgrades left. This microscope lens is first among more expensive ones : Olympus 10x/0.3 WI - with water immersion, was lucky to get it for 800$ on ebay.

Work with water immersion could allow to increase resolution (which is not used here - NA is only 0.3) and achieve slightly wider depth of focus for same NA - but (more importantly for me) it makes contamination of the specimen much less visible: dirt reflects/absorbs light very well on the surface to air and much less so in water due to much lower difference in refractive index.

Had to spent some time looking for proper containers for specimens which were flat enough (down to few microns). Finally, biological plastic Petri dishes worked well enough (glass ones were less flat but also usable). Very satisfied, fun lens :-)

First "production" photos here.

Sat, 21 Jan 17 19:20:18 +0300
<![CDATA[Flying over Minsk and Zurich]]> Minsk:


On the way back it was all covered with clouds:
Sat, 03 Dec 16 19:30:44 +0300
<![CDATA[Winter is coming : 1212 hours timelapse (50 days)]]> 50 hour timelapse - now it's time for 50 day timelapse (1212 hours). Starting from green leaves to snow. Shooting on Xiaomi Yi, 1 frame per 30 seconds, and then - 16x faster in ffmpeg via frame averaging. Total video speed is 28'800х realtime. Original video was ~80 minutes, 28.7Gb (50Mbit/sec bitrate).

Tue, 15 Nov 16 07:02:59 +0300
<![CDATA[Lighting fluorescent lamp on a bench (with 400V/1A power supply)]]>

Heating filaments from one side, ~100mA:

If we crank up current a bit (to 220мА) - heating alone is enough to excite mercury and get white fluorescence. This mode of operation is very dangerous to the lamp (this is exactly the way I fried filament on the other side):

Now lamp could be started from ~150V DC, discharge current is 100mA with ~42V voltage drop. DCDC power supply is not fast enough to support discharge due to negative resistance of the lamp, so ballast resistor is required. Minimal resistor for stable discharge appeared to be ~74Ω. With lamps of this type one can disable filament heating during discharge - this will improve efficiency and expand lifetime.

Without heating lamp could not be striked even by 400V DC, higher and higher filament current reduces striking voltage to 100-150V. But that could be further lowered by preheating both filaments and going AC using H bridge.]]>
Sun, 13 Nov 16 23:16:01 +0300
<![CDATA[59kV is no joke]]>
This is ~10kV ~2mA discharge:

Mandatory warning: Anything above 50V can kill you in "suitable" conditions. Achieving that at 1-60kV is much easier.

When arc gets orange - things get scary, metal electrodes start to melt and evaporate. ~15kV ~12mA, 180W in arc:

Droplets of molten copper on the wire:

In some positions of electrodes "on the side" there is a "thick" discharge rather than thin arc:

Here are my safety gloves - only rated for works up to 1000V, yet better than nothing. Random spark or arc will more likely go around the finger.

Inside the supply there are 4 transformers with HV diodes between them in HV heat-shrink tube. Everything is glued with some hot-melt white glue, looks not very reliable yet works just fine. On mains input they didn't omitted "unnecessary" common mode choke - that's a good sign.

When running for max voltage had to isolate terminals with glass vials in order to suppress electric noise around sharp electrode features:

Initially power supply was limited to 42kV for reliability reasons. Manufacturer gave instructions how to unlock full 60kV (that requires cutting 1 resistor inside).

And here is where priceless HV experience starts. HV voltage probe has a note saying that negative lead should be at ground potential. What does that mean? It has decent isolation only on "hot" terminal. If you leave HV supply floating - negative terminal will go somewhere around -30kV, and positive to around +30kV (60kV total), but weak isolation of ground terminal in HV probe cannot handle that and there will be a breakdown. But if you tie one of HV terminals to ground - probe is now ok, but hottest transformer inside supply is much easier to breakdown. This is exactly what happened.

Last voltage I've seen while testing for maximum voltage - -59kV, after cranking up voltage a bit further - I've heard the scariest thing one might hear - sound of arc inside power supply. Instinctively you expect it go bang and fly into pieces across the room :-) Luckily for me, only 1 last transformer got damaged gracefully.

How you tie HV terminal to ground - use neutral or protective earth terminal? Surely protective earth, you only have 50% chance of finding neutral in EU outlet :-) This is my "ground" tool:

After transformer breakdown it no longer holds more than 20kV. I guess experience was well worth it. To repair it I've ordered 4 spare transformers with HV diodes from the manufacturer. They actually surprised me by including spare parts with the supply - power transistors(20N60C3), spare microchip with markings removed and some passives.

PS: This is how how 60kV cable looks like:

Thu, 27 Oct 16 15:29:56 +0300