Old-school hardcore overclocking

Absolutely amazing!!

sick stuff there :D

Solid: Ceramic caps have extremely low ESR, making them great for stuff like this... why replace them? Is there a better solid-state cap out there, or are you talking about the cylindrical ones?

s0lid

Thanks for reply :)

Please, do not confuse people. Ceramic capacitors - are small brick-type ones, most of them without legs, for mounting on PCB top layer. They hape excellent ESR, ESL and can work with high frequency, like 1-100 MHz range. They are just perfect for filtering and noise removal. But they all have small capacitor, most large (and expensive!) of them have 22-68uF. That's way too small for high-power apps with 500kHz-1MHz switching range. That's why you can find a lot of "usual" cylindrical caps in cans, with electrolyte. They offer good capacity - to thousands of uF, but with much higher ESR, ESL and worse temperature stability. Also can caps usually working good for low frequency, below MHz's, it's like to push huge tank to move. Once you get it moving, it's hard to stop. More frequency > more parasitic heating > less stable... So to lower ESR vendors install a lot of caps in parallel.

Little note: A lot of people saw "digital" VRMs, with only ceramic caps. That's because Volterra (vendor who invented this power units) use high frequency (1-2MHz) for switching, and thus their VRM can work without need of big capacitance for storing charge. Faster frequency - less component size, with same power rating. Remember old bulky 50/60Hz transformers for house mains? They are huge, even with power output only tens of watts. That's because of 50Hz. Increase clock to 5000Hz - and you get 100 times less space and weight for same power. But when you go with extreme overclock - low capacity makes thing bad, and advanced people, like Hipro5 always add solid polymer electrolyte caps to help VRM powering high-power load.

Ceramic caps are perfect for fast transients, big can caps are good for low speed filtering and energy storage. But when you use both - you get pros from each type, and lowering "bad capabilities" of can caps alone a bit.

I'm looking forward for your X48 VRM conversion :D

Tin EOF :worship::bows::banana3:

Crazy Great Old Skool Modding man :up: beats soldering crystals to OC :D
This thread is just tooo much fun :clap:

http://www.xdevs.com/images/chiller/7800gtx_1.jpg

Continue :D

Added extra wires to get less drop on Vgpu voltage under load. Also added FB line from near-GPU caps. Added some extra grounds.

Videocard review on current state. 7800GTX Radeon powa edition :D

http://www.xdevs.com/images/chiller/7800gtx_3.jpg
http://www.xdevs.com/images/chiller/7800gtx_2.jpg

By the way, this is the ONLY 7800 videocard with TWO PCI-Express power connectors :D

Quote:

---

Originally Posted by TiN_EOF

s0lid

Thanks for reply :)

Please, do not confuse people. Ceramic capacitors - are small brick-type ones, most of them without legs, for mounting on PCB top layer. They hape excellent ESR, ESL and can work with high frequency, like 1-100 MHz range. They are just perfect for filtering and noise removal. But they all have small capacitor, most large (and expensive!) of them have 22-68uF. That's way too small for high-power apps with 500kHz-1MHz switching range. That's why you can find a lot of "usual" cylindrical caps in cans, with electrolyte. They offer good capacity - to thousands of uF, but with much higher ESR, ESL and worse temperature stability. Also can caps usually working good for low frequency, below MHz's, it's like to push huge tank to move. Once you get it moving, it's hard to stop. More frequency > more parasitic heating > less stable... So to lower ESR vendors install a lot of caps in parallel.

Little note: A lot of people saw "digital" VRMs, with only ceramic caps. That's because Volterra (vendor who invented this power units) use high frequency (1-2MHz) for switching, and thus their VRM can work without need of big capacitance for storing charge. Faster frequency - less component size, with same power rating. Remember old bulky 50/60Hz transformers for house mains? They are huge, even with power output only tens of watts. That's because of 50Hz. Increase clock to 5000Hz - and you get 100 times less space and weight for same power. But when you go with extreme overclock - low capacity makes thing bad, and advanced people, like Hipro5 always add solid polymer electrolyte caps to help VRM powering high-power load.

Ceramic caps are perfect for fast transients, big can caps are good for low speed filtering and energy storage. But when you use both - you get pros from each type, and lowering "bad capabilities" of can caps alone a bit.

I'm looking forward for your X48 VRM conversion :D

---

Well that's something new i didn't know, thx for sharing this info! :up:

What comes to my X48 vrm conversion, need to find out way to get phase controller turn on with only ESP connector connected, but i think it's voltage is taken from 5V or 3.3V. So not a big deal actually =P

Which PWM controller do you have? You can start dedicated thread, i'll try to help.

ISL6336

http://www.xtremesystems.org/forums/...d.php?t=234215 here's the thread :)

you are mad and I like it!

Did you measure the ripple of your extra psu? How high is Vpp?

Did you add the extra psu to the stock psu of the 7800gtx or did you deactivate the stock one?

Awesome stuff there :)

First time this noobs seen anything like this. Far out man, very impressed by you're knowledge :up:

loopy83

Ripple is quite high, it's because of wired connection to board and 2phase design (7800GTX is quite power hungry chip)
I've used external VRM because native onboard one was fully damaged, VGPU convertor just burned PCB away 4 layers INSIDE, I've posted damaged pic earlier in this thread. So I decided not to mess up with native VRM, and hook up Radeon's power :-D

So, I finished my benches with this mad 7800GTX :-D.

With core voltage 1.8V and Vmem = 2.4V, GPU chilled out by A/C unit chiller (-12 to -20°C antifreeze liquid in WC loop), and QX9650 @ ~4.6GHz 1.65V chilled to I've got:

http://hwbot.org/img/silver-trophy-hw.png
3Dmark05 - 12625

http://hwbot.org/img/silver-trophy-hw.png
3Dmark06 - 7298

http://hwbot.org/img/bronze-trophy-hw.png
3Dmark03 - 28105

http://hwbot.org/img/medal-hw.png
Aquamark - 191744

#12 3Dmark2001 - 65449 - http://hwbot.org/result.do?resultId=912039

Quite nice as for me :D

Lol man
hey you think it would be possible to replace memchips? fill it up with 0.8ns memchips 7800GTX with 1GB :D

Of course replacing memchips is possible, if pattern on PCB suits chips ;)

I've added memory to 8800GS 384MB card not a long time ago.

http://www.xdevs.com/images/g92gs2.jpg
http://www.xdevs.com/images/g92gs3.jpg

But there was blocked memory channels inside GPU, so just GPU didn't detected added rams :D Still if replace GPU from 8800GT on it, it will work. Maybe, LOL.

0.8ns mems are in FBGA-136 rectangular package. 1.6ns on 7800GTX is FBGA-144. So no way. :)

Quote:

---

Originally Posted by TiN_EOF

Of course replacing memchips is possible, if pattern on PCB suits chips ;)

I've added memory to 8800GS 384MB card not a long time ago.

http://www.xdevs.com/images/g92gs2.jpg
http://www.xdevs.com/images/g92gs3.jpg

But there was blocked memory channels inside GPU, so just GPU didn't detected added rams :D Still if replace GPU from 8800GT on it, it will work. Maybe, LOL.

0.8ns mems are in FBGA-136 rectangular package. 1.6ns on 7800GTX is FBGA-144. So no way. :)

---

But, with same package and replacing the ones that are on it, it coudl work? :D

There is no 0.8ns or even 1.1ns mems in FBGA-144, afaik.

Quote:

---

core voltage 1.8V

---

go for 2v+) i bench 7800gtx @ 2.2v :up:

GraduS

Спасибо, я уже набенчился с 7800 :)

So, 7800GTX experiments ended, but look, we have new zombie for our crazy experiments :-D.

That dead 9800GX2..

Sees only 1 GPU and bridge chip, and unlucky, primary GPU isn't working.

Resoldered G92's GPU's on both half PCB's of card, and bridge too, but no success, final is the same.

What to do next?

YES, KILL working 9800GX2 for knowledge and new ideas :-D
That's joke...taking second working GX2 and disassembling it for parts. I tried to test "dead" halfs with alive ones.

Simple checks, tens of screw-descrew of PCB's , and gotcha - that freakin interconnection FCB (flexible curcuit boards) between two parts of GX2 is dead. Using good FCBs from alive card on dead GX brings life to that pieces of hardware.

Heated dead FCB's to some very good temps... No way, assembled card isn't working still.
But if second half is laying near connected, but not mounted on HSF - card works.

http://www.xdevs.com/rep/gx2_1.jpg

Now I need another dead 9800GX2 to replace FCB's and continue evil sciences :-D

Tried Quad, but forgot that XP not supports it :D

http://www.xdevs.com/rep/gx2_3.jpg

BTW, there is XBOX GPU heatsink on laying part of GX :D

And NVIDIA Bridge chip is fool, P5B deluxe mobo used for test have only 1 PCI-E x16 and second is PCI-E x4. But that chip gives x16 on all 4 GPU's. LOL.

http://www.xdevs.com/rep/gx2.png

again nice work :clap:

Omg dudes you crazy

Simply WOOT!!!!

Crazy!!!! Nice job!

Lots of respect, I'm really enjoying your writing and photos! Please keep more information coming, I'm hungry for it :)

Wow, that's very Xtreme! :up:

I can Never get enough of this thread :up:

If a bios flash doesn't fix my 3870X2 it may be time to play :rofl:

New frankenstein card arrived :-D

Zotac (crappy cheapo stuff) GeForce 8600GT 256MB DDR3.

Stock VRM's based on APW7067N chips, pair of DPAK FETs and inductors. Both single phase.

Hooked up FB voltage mods to these APW's, set Vgpu= 1.77V , Vmem = 2.3V.
Mounted on aluminum waterblock in A/C chiller loop.
Added GOOD capacitors on the back of card, just to be sure.

http://www.xdevs.com/images/chiller/ch0.jpg

First try..testing with FurMark during 6000sec.

GPU clock ramp up from stock 570 to 700....800....900...1000. huh, DRAMs - 700-800-900-950--1000..errors...back to 950.
Tried aquamark03 (my favourite) at 1000/2200 shader/950 drams...got black screen when gray dust appears from tubes. Reboot - no go, black screen. Power off...power on - ok, working.
again tests - the same...even at 900MHz core power VRM for GPU shutdowns completely and latch off till power down. Overcurrent protection here, no doubt.

Crappy APW7067 don't have any external pin for OCP control.
So card can't handle power for clocks above 850/950.

So....frankenshtein edition appears again, 4850 power for 8600GT. This adds PCI-E power connector to 8600GT too :-D.
1.8V, 2.2V for memory. Everything is ok, no shutdowns.

http://www.xdevs.com/images/chiller/fr86_1.jpg
http://www.xdevs.com/images/chiller/fr86_2.jpg
http://www.xdevs.com/images/chiller/fr86_3.jpg

Card works with clocks up to 1020/2280/972 quite good.

And after few hours, with next rig:

Intel SLAN3
Swiftech Apogee GT with -30°C Liquid
ASUS P5B Deluxe no mods
2x1G D9DQW's
Zotac 8600GT 256MB DDR3 frankenshteined and cooled by Waterworker AL WB with -30°C Liquid
Xtremelabs.org Chiller A/C unit for liquid loop cooling.
Enermax Revolution85+ 1250W

Got next results:

01: 62606 , 1012/972, CPU 4533MHz
03: 27211, 1012/956, CPU 4600MHz
05: 17500 , 1000/956, CPU 4600MHz
#5 06: 9147 , 990/950, CPU 4507MHz
aq: 225313 , 1012/1018, CPU 4600MHz
#2 vn: 3348 , 1000/950, CPU 4480MHz

TiN, your hardware mods are sick! :shocked: Congrats! Keep going

Awesome :shocked:.

I saw this thread on the 31/10/09 but then the following day I couldn't find it again and spent 1-2 hours looking for it. Finally found it again.

Amazing mods their :clap:.

that card looks really crazy. awesome work :clap:

62606.0 :clap: :up:

Some little update again :)

Not frankenshtein this time. Started some more advanced tech engineering now :D

I'm working on quite big R&D industrial project ATM, and decided to make some overclocking fun with my new digital controller.

Already soldered by hand few prototypes, ready for firmware programming.

http://www.xdevs.com/kb/HC1/HC1_es_2.jpg

It has 8 independed analog input channels (all with protection and attenuator curcuitry), onboard +3.3V power convertor, some memories, real-time clock with backup battery, custom I/O hub and nice 132x32 pixel graphic LCD. PCB suits 5.25" computer bay.

I've got three 8800Ultra videocards, and decided to bench them in tri-sli on Maximus III Formula with Intel Xeon 3470 LGA1156 CPU.
All videocards already voltmodded, and will be cooled with waterchilled setup with A/C unit, together with CPU.

Programmed simple code for reading 6 ADC channels and showing them on LCD

Initial tests with Radeon 4870.
http://www.xdevs.com/images/3sli/voltunit5.jpg
Good, it's working, readings calibrated to my 0.05% accurate Fluke 87V.
There are three lines dedicated to 3 videocards, all with two voltages, for GPU and memory respectively.
Readings updated 5 times a second for smooth readout.

Next hook up to 8800U's.

http://www.xdevs.com/images/3sli/voltunit1.jpg

Just controller powered. Vgpu and Vmemory sense wires soldered to caps near GPU, with ground for precise measurement.

http://www.xdevs.com/images/3sli/voltunit2.jpg

Side view, you can see Vgpu mods too. Decided not to do Vmem mod, coz these cards can do 1.2GHz mem without any mod.

http://www.xdevs.com/images/3sli/voltunit3.jpg

Now brief tests with two cards aircooled.

Better view of controller:

http://www.xdevs.com/images/3sli/voltunit4.jpg

:) If anybody interested, I can make some controllers like this in some future. Also possible temperature channel add, more channels, more controls, etc.
Will continue with 3-way SLI setup next days.

Very nice controller

i wanted to make a controler like this but using Arduino hope make it soon. :) Gratz Crazy HardModder :P

:clap: Great I can wait till you get a finished product :up:

Once you go 2 cards it's hard to go back :D

http://www.xdevs.com/images/3sli/chill3w2.jpg
Single card with chilled out waterblock :-D.

Time to say three. Slave cards on air, initial tests. Controller now reads all three cards ;)
http://www.xdevs.com/images/3sli/chill3w3.jpg

Frosty :)
http://www.xdevs.com/images/3sli/chill3w1.jpg

Chilled card do 750/2000/1200MHz with 1.55Vgpu,2.15Vmem. Rivatuner reads +2°C idle, +18°C load.
Aircooled 700/1200MHz with 1.5Vgpu,2.2Vmem.

Will continue with all cards under subzero.

So far
№3 Aquamark multigpu - http://hwbot.org/compare.do?resultId=929080 - 18.9 boints :D

:clap: lol that is chillin out !

:eek::eek:

:yepp:...man, you rocks. I love the first part, 6600LE powa by 4850. hehe, nice things. BTW, about changing memory chips, is that possible?

Respect :up:

Kicking it since 939, nice going mate!

TiN_EOF You Sir are a BEAST!!! what do you depend on for your system PSU?

Just the Enermax Revolution85+ 1250W?

Yes, it's more than enough. Still I have second revo if any need :D
Thanks to Enermax Germany for them :D

wow, I liked the LCD showing the infos of the VGA!

My jaws been on the desk for the last 6 pages, amazing work! Proper hard core!

Nice controller Tin! Good luck for 3 SLI action under cold :up: