Voltmodding the New Gridseed G-Blade Scrypt ASIC Miner

24 Apr
2014

gridseed-g-blade-voltmod-1

In the previous post with our first hands on experience with the new G-Blade miners we have mentioned a bit the topic about voltmodding the new Blade Miners. Now is the time to talk a bit more about it and share our experience based on the voltmod experiments we have already done on our G-Blade. We already said that the new G-Blade uses the same UPI Semiconductor UP1509 chip for controlling the voltage of the GC3355 chips as is used on the smaller 5-chip ASICs. So it is easy to do the voltage modification to the G-Blade as you are going to be essentially replacing the same resistor, though there is a bit of a difference in how things look on the new PCB. There are no element markings, so make sure you replace the right resistor marked on the photo above with a green circle with the 333 written on it. The resistor we are replacing is also a 33 kOhm one (~1.2V default), though here it is 603 type and not 402 like on the 5-chip ASIC devices, so it is easier to replace.

Do not be in a hurry to solder a new one before reading our advice. Since the default voltage of the GC3355 chips is 1.2V as on the smaller 5-chip ASIC miners here you can pretty much expect the same out of the box performance, though our tests have shown about 800 MHz to be the safest option in terms of how high you can go with no or very minimal HW errors, so a bit lower results than what we are seeing on the 5-chip ASICs. Also since the voltage regulators used on the new G-Blade are not that much powerful than the ones used in the 5-chip models you are not going to have the same safe headroom to increase the operating voltage of the chips, in fact you are actually pretty limited.

gridseed-g-blade-voltmod-2

We were quite disappointed in how Gridseed has designed the cooling of the new G-Blade miners, and especially the use of a powerful and noisy fan to compensate for the lack of adequate cooling for the three VRMs. The CG3355 chips can easily be cooled with a silent and much lower RPM fan, however it seems that the powerful fan is there just to be able to ensure that the mosfets are kept cooler without having any radiators directly placed on them or on the back of the PCB where they are located. We are getting about 60 degrees temperatures of the IR mosfets at the standard voltage and operating frequency of 800 MHz, however as soon as we remove the fan the temperature of the VRMs starts to get higher very quickly while the CG3355 chips still remain pretty cool. By replacing the resistor marked above with a higher value one and increasing the operating voltage of the chips you can easily overload the voltage regulators and burn them in a matter of just a second without the fan and even with the fan running you can still burn them if they get too hot. So the moment you decide to voltmod the new Gridseed G-Blade you need to also think of improving the cooling of the voltage regulators a lot and not try to push the operating voltage of the chips too much, or you risk the VRMs easily overheating and burning up. We have used an old motherboard chipset radiator glued on the back of the PCB (with thermal glue) where the voltage regulators are placed, so that it can help in keeping them cool and have also placed a smaller video memory cooler directly on top of the regulators on the other side of the PCB.

gridseed-g-blade-voltmod-3

We have slowly started increasing the operating voltage of the chips from the standard value of about 1.2V until we have reached 1.4V with the use of a 39 kOhm resistor and decided to stop there as we’ve already seen the voltage regulators getting quite hot. At this point we were aiming at a stable operating frequency of about 1000 MHz while keeping the device cooler and being on the safe side for the voltage regulators. With the 39 kOhm resistor replacing the standard 33 kOhm one we already saw the mosfets easily getting up to 90-100 degrees Celsius in about a minute (without the extra cooling). With our extra 2 radiators attached to the PCB and the chips we were able to bring the temperature below 70-75 degrees on the long run, so pretty much safe for leaving the miners run all the time like that, though you should continue to carefully monitor the temperatures. We have decided to stay at 39 kOhm, though pushing a bit more for like 40-41-42 kOhm with extra cooling might work, going for 47 or 49 kOhm may be a bit too much for the VRMs to handle. So be extra careful should you decide to go higher tan what we have used, though we do not recommend it, and monitor carefully the temperature of the voltage regulators as they may get very hot in no time and burn up without adequate cooling!

gridseed-g-blade-voltmod-4

Here is what we are getting after replacing the resistor from the standard value of 33 kOhm to a 39 kOhm one. Running the device (only one of the PCBs is connected on the photo) at 1000 MHz seemed quite Ok with little HW errors, though we have decided to go a step lower in order to lower the HW errors a bit more. After some additional testing we have settled for 988 MHz as it brings down the HW errors up to something about 1-2 per hour which is much more acceptable than getting an error about every minute or so at 1000 MHz. The local hashrate at 988 MHz we are getting per PCB is close to 3.4 MHS or almost 6.8 MHS total for the whole miner with the 2 PCBs modified, up from about 2.6-2.8 MHS per board or about 5.2-5.6 MHS without the mod and the miner running at 800 MHz. The power usage per PCB after the voltage modification is about 85W or a total power consumption of about 170W for the whole modified miner.

The results we are getting are not bad, but do require you not only to solder a resistor, but also to improve cooling of the device. You also need to be extra careful should you decide do go even higher than what we have achieved as the voltage regulators might not be able to handle well the additional load or the temperature. It seems that with the G-Blades things were already pushed close to their limits by Gridseed, so there is not that much of headroom left for users to squeeze some extra performance for free by voltmodding the units. We are a bit disappointed by that, we kind of expected more, especially after seeing what the GC3355 chips were capable of in the smaller 5-chip devices. We are going to be doing some more tests and experiments in the next few days to see if we can get something better than what we are already getting, but the chances for some significant improvement are not that good.

Again we advise anyone willing to do a voltage modification to his Gridseed G-Blade ASIC miner to be extra careful as these devices are more expensive and can handle less abuse from the users as compared to the smaller 5-chip ASIC miners, so proceed with caution! The smaller 5-chip miners allowed more headroom as they were designed to handle much higher power usage because of the BTC mining part, but for the new G-Blade miners Gridseed has decided not to support the SHA-256 mining part (the chips do, but the device cannot supply enough power, so do not try to activate BTC mining or your miner may burn). So the headroom for overclocking without doing some modification related to the voltage regulators is actually very little with the new G-Blade ASIC devices, and working only on improving their cooling does help only a bit. Reworking the power part of the miner could open up some additional possibilities, but that is not something that most users can easily do themselves unlike the replacement of a single resistor for example. In short, if you are expecting performance boost on the bigger G-Blades like on the smaller 5-chip devices with a voltmod, you will be very disappointed, and the fact that after voltmodding the Blade Miner you might have trouble cooling it down properly for long term operation might not be worth the extra performance boost you will get.



Other Similar Publications:

21 Responses to Voltmodding the New Gridseed G-Blade Scrypt ASIC Miner

Stanley

April 24th, 2014 at 17:30

Thanks for the tips. I’m not really good at soldering so I may just skip the volt mod to avoid frying my blade. However I have a question. Since those GC3355 chips are naked, I wonder if adding a simple heat sink on top of those chips will do any help on overclocking a bit higher than 800Mhz?

admin

April 24th, 2014 at 17:40

Stanley, if you want to add heatsinks on top of the chips you would have to find individual ones with the size of each chip as there are elements on the PCB around the chips that are taller than the CG3355 chips, so no option to use a larger ones for multiple chips. With that said, adding heatsinks on top of the chips will most likely not help at all in getting better overclock, at least on stock voltage… the chips are pretty cool with the stock cooling solution. The only problematic part for the cooling seems to be the voltage regulators and they become even more problematic after applying a voltmod. Without modding you can try 825 and 838 MHz as well, could get you a bit better performance with just a little bit of HW errors more that at 800 MHz.

geetash

April 24th, 2014 at 21:26

image shows 5 H/W errors for the 5min mining time, thats quite a lot.
can we add a heatsink or something on top of the voltage regulation ?

admin

April 24th, 2014 at 21:48

geetash, actually one HW error per minute is not that bad, but far from great, we do not like that very much as well, so we are going to try to improve the performance with more testing. Do note that this is a result with automatic switching pool, mining for a fixed coin does usually give better poolside performance and less errors – stale shares and HW errors too. This result is actually with the two heat sinks attached as you see on the photos above. Going a step lower in the operating frequency to 988 MHz does help in getting much lower number of HW errors, the result is similar to using 838 MHz instead of 850 MHz without any mods on the G-Blade.

Jabberwock

April 25th, 2014 at 00:03

Hi, you can push a little higher the clock with 42k resistor.
With 39k, I was able to get 1025Mhz with few HW errors.
42-43k gives 1063Mhz for 7.2Mh/s and almost no HW errors
302 watt at wall

admin

April 25th, 2014 at 00:18

Jabberwock, have you measured the temperature of the voltage regulators with the 42K resistor mod? Also have you improved the cooling of the regulators? Do check the temperature of the choke as well as it also tends go get quite hot.

Ryan

April 25th, 2014 at 02:43

I followed your instructions word for word, from mod to cooling, and it was working perfectly at first. After about 2 hours G-Blade died with several components melted. For cooling I had a heatsink just like the ones in your picture, cooling paste that I’ve used extensively before on my gridseeds with great results, plus the stock fan was working (quadruple checked) and had an extra larger fan from the outside blowing in a freezing basement mind you. As you mentioned these devices do not have any overhead for overclocking and pushing them any further than stock is just too risky and consequently dumb. This is not a gridseed mod, these were built to prevent people from overclocking and anyone trying to get around that will likely meet the same fate I have (almost $2k LOST). Accounting for variances in devices and modding expertise I don’t see anyone running a modded G-blade like the one I did from here and actually seeing the device surviving days not to mention the week.

TC

April 25th, 2014 at 04:21

what about replacing the VRM’s with higher rated ones?

admin

April 25th, 2014 at 09:27

Ryan, do you have only the VRMs blown up, if so there is a chance that they can be replaced and the device to be revived to life. We have actually done that on a 5-chip ASIC that we have pushed way too much and the voltage regulators were not able to handle the extra load. Can you please post some photos?

Again, we advice anyone willing to do this mod to be extra careful and monitor the temperatures of the chips carefully. While these little chips are quite tough and rated with fairly high specs and meant to be able to operate at high temperatures, their small size makes it hard to keep them cool, especially on the long run. The better idea would be not to try to replace them with more powerful ones (still hard to cool them down), but to actually have a second phase just like motherboard manufacturers do. Do note that the choke near the VRMs also gets very hot and it may also be a problem if you go too high with the voltage.

Jabberwock

April 25th, 2014 at 11:31

Admin, you can follow my thread on bitcointalk.

I haven’t been able to measure temperature, and one of your thermographic picture would be nice to see hot/weak spots.

as you said, I also noticed Ferrite beads and the inductor are also getting hot.
The last part is the power socket that isn’t rated for that much power and needs to be double for security purpose.

A second phase would be great, but in will be possible only with a complete new design of the boards

here are my best results so far. I don’t think it can be pushed much more

Ryan, I’d check components in this order: the 16V capacitor, the VRM, the ferrite beads.
My 16V capacitor blew after a around 1 hour on one of the boards, replace it with 25V 220uF one if you can, it’s safer anyway. Ferrite beads are subject to explode on the pods, can be the same on the blade as they are getting really hot.

Jabberwock

April 25th, 2014 at 11:32

link not showing for picture

http://i61.tinypic.com/k0fwba.jpg

admin

April 25th, 2014 at 12:09

Jabberwock, Thermal images are coming shortly…

Scott

April 26th, 2014 at 11:54

Hi guys,
I have been researching this 30V Single N-Channel HEXFET Power MOSFET heat issue along with the crappy cooling on my new G-Blade unit. In fact, my Q6 fried on my new/stock unit after a few hours of normal use at 800MHz. Call it a rare fluke or a poor power design, something needs to be done to increase the robustness of the power circuit feeding these expensive miners. This is when I realized it was time to put on my ham radio/electronic design hat and get to work. In reviewing the overall design of the G-Blade, I came to the following conclusions:

1. I agree with a lot of what the OP says.

2. The G-Blade would be better cooled and controlled in my setup as two separate devices, which it electrically already is. So, I unscrewed the 6 post securing both halves and laid the miner cards side by side.

3. I found that two lower RPM and whisper quiet [CoolMaster] 120mm 12VDC 0.15A case fans can be used side by side face down and cover over 85% of the card nicely – making it look very similar in size and appearance to a full sized dual fan GPU card.

4. I will be swapping out the stock DC JACK for a more robust 20A 2.5mm/5.1. capable polycarbonate/nylon based jack.

5. Like the OP, I have some device heat sinks on order that will sit on top of the power MOSFET’s and help dissipate heat.

6. Since I need to replace Q6 that was bad on one of the two cards that make up the G-Blade, I did some research and found there is a more robust version of the same 30V MOSFET – It is the 5303 (IRFH5303TR2PBF) and is rated at 82A instead of the standard 5302 40A. Likewise, the heat dissipation is better on the 5303 and it only costs about $1 more than the 5302. I will receive the devices this coming Thursday.

At any rate, when all is said/done, I should have two side-by-side G-Blade miners

zyberguy

April 26th, 2014 at 19:56

@Jabberwock I cannot find your thread on bitcointalk. Do you have a link?

Jabberwock

April 28th, 2014 at 15:24

zyberguy, thread is named ” GRIDSEED G-BLADE Overclocking 7Mh/s ”
for some reasons, BTCtalk moved it in a wrong place.

Kebabman

May 8th, 2014 at 07:28

In a careless moment I wired the 12V DC input the wrong way around, FB29 and FB28 promptly immolated themselves when I plugged it in :( the LEDs on the board still light up but it’s not recognized by the computer…what are the components at FB29 and FB28? Wondering if I might have escaped further damage and can just replace the damaged components?

admin

May 8th, 2014 at 10:05

These are Ferrite Beads. The good news is that they have probably saved your miner by acting like fuses and replacing them should most likely bring things back to normal. Not sure what are the ones used on the G-Blade, but could be pretty much the same as the ones used on the 5-chip miners, so you can try with 300ohm @ 100MHz, 3A.

Ryansgt

May 17th, 2014 at 00:50

Scott, i’m looking at the pic from the post and it looks like the mosfets that are on there are 5300, did yours come with 5302? also, have you found a source for 5303. Seems like with a more robust power system, these chips could be pushed pretty far.

Kebabman

May 17th, 2014 at 05:21

Admin, wanted to say thank you for pointing me in the right direction, the ferrite beads are actually a larger size than I expected (I bought the smallest ones of the spec you mentioned to start with, they we about 50% smaller than required), but once I found the right ones swapping them out made the blade fully functional again. Thanks! :)

btw any tops on soldering onto a PCB? I have a needle point soldering iron but I found it very hard to get the solder to attach to the PCB, it just wanted to stick to the iron, I got it done eventually but there must be an easier way. I want to try volt-modding my 5-chip miners but I don’t want to mess them up.

admin

May 17th, 2014 at 20:22

Ryansgt, our unit came with two 5300 and one 5302, though we’ve seen that Gridseed likes to use different components, probably if they can source larger quantities cheaper… on the smaller 5-chip models we’ve seen a lot of variance in the mosfets used, so on the G-Blade it is possible that they use different ones on different batches as well.

Kebabman, the right ferrite beads were the 805 type if I remember correctly. For easier soldering you could try using solder flux for the voltmod, it should make things easier, though with hot air station it is even easier to replace small SMD elements.

simon bliss

September 27th, 2016 at 17:26

https://www.scan.co.uk/products/92mm-zalman-zm-sf2-fan-sharks-fin-blade-inc-anti-vibration-silicone-pads-and-pins perfick for blades just snip the grey and solder on the red and black — loads quieter — update soobn on my solar and wind rig ;)

Leave a Reply

Your email address will not be published. Required fields are marked *

top