EVGA probably isn’t the first name you think of regarding high-quality PSUs, but they’re certainly no newbies considering they made what is basically the Tesla Roadster of PSUs – top-flight specs that almost everyone wants, but also massive overkill and beyond budget for most PC builders. Today, in contrast, we’ll be reviewing an EVGA Supernova 650 G3 PSU, one that the average PC builder can actually afford, yet is adequate for even a fairly high-end PC. Will EVGA be able to deliver on a budget?
Specifications and Features
It uses a HDB variable speed fan for very quiet operation and high reliability. It also boasts a lot about low ripple and very good voltage regulation, which we’ll take a close look at later. It is a single rail PSU like almost all modern PSUs on the market (multi-rail PSUs are becoming less common and true multi-rail PSUs almost unheard of) that claims to be Haswell ready and SLI/Crossfire ready. The Haswell ready is almost a given since it uses a DC/DC converter design which means the 12V output pretty much never gets a load small enough for it to operate in a region of poor transient performance. The SLI ready is mostly marketing since the provided cables will only accommodate at most 3 GPUs that each only has one power connector while most mid to high-end GPUs have at least 2 power connectors, but it would technically work for a video editing machine that favors multiple lower end GPUs over a single high end one. The warranty is a generous 7 years.
Below are the listed specifications and additional details about the power supply:
Packaging and Accessories
We’ll start by opening the box at which point we find a bag containing the PSU, another bag containing the cables, and an installation manual. Included with the cables is an adapter to enable the PSU without a motherboard, something that will come in handy later.
The cables are as follows, all removable:
28″ 4/8 pin EPS12V
24″ 6/8 pin GPU
24″ 6/8 pin GPU + 29″ 6/8 pin GPU
2x 20″ SATA + 24″ SATA + 28″ SATA
20″ Molex + 24″ Molex + 28″ Molex + 32″ Molex
4″ Molex to mini 4 pin
I am not a fan of the dual 6/8 pin GPU cable since it could put up to 300W of load on a connector designed for 150W. Additionally, since the PSU cannot be used without the ATX cable, I see little value in making that removable (though many PSU manufacturers do this for a completely modular design). Admittedly, uses for an ATX PSU without the ATX cable are very limited but they do exist – mining rigs and 3D printers come to mind as common examples. I would prefer the ATX cable be permanent, then use the space and cost savings to put the third 6/8 pin GPU connector on a separate cable. I can even go as far as suggesting the 4/8 pin EPS12V cable be permanent as well since it would be highly unlikely for a PC that needs a 650W PSU to not use it.
The Power Supply
Onto the PSU itself, we see that it’s a universal 100-240V design like almost all PSUs on the market. The full rated output is available on 12V, exactly what is desired for modern PCs that are largely 12V loads. The 5V can get pushed on PCs that have a lot of USB ports, but that’s a problem that’s solving itself as fast charge capable USB-C ports are powered by their own voltage regulators from 12V.
The fan has EVGA logos embossed onto the blades. I speculate they work like the dimples on golf balls to slightly improve aerodynamics. But as we’ll see later, that is not likely to play any significant role in thermal design.
On the front are the output connectors, along with the EVGA logo and a warning to not remove the cover.
The back side has the power input and service switch as we expect, along with a switch labeled “eco”. The remaining area is taken up by the rear vents.
Even the bottom is not so boring with the EVGA logo again and some stickers. No idea what use is a “warranty void if removed” sticker that doesn’t have to be removed to open the case.
Now comes the exciting part of opening it up and trying to understand how it works. Do beware that such an action is only to be done by experienced electrical engineers and as often said on Mr. Carlson’s Lab (a popular electrical engineering TV show), “There are high voltages inside and if you’re following along, you’re doing so at your own risk.”
The first thing we notice is a piece of plastic covering part of the fan. The intent is obviously to direct airflow closer to the center of the board, but it is clearly not a very aerodynamic way to do it. A substantially better way to do it is to set the board closer to the back so there is less airflow to redirect in the first place, then use a bent piece of plastic that works better for directing airflow.
Inside, we start with a fairly typical arrangement with an active PFC input stage going into a large 400V capacitor, complete with a relay to bypass the precharge NTC for an efficiency boost. But beyond that, things aren’t so typical anymore. Instead of the full bridge design common in modern PSUs of that power level, it’s a half bridge design that saves a lot of cost and complexity – two fewer transistors and one less gate drive circuit. The half bridge design used to be considered unreliable due to high ripple current through the input capacitors, but rest assured this PSU has done it right.
The two red-orange film capacitors next to the transformer are what take the ripple current, something they do much better than electrolytic capacitors do. Then the 12V is synchronously rectified by 4 MOSFETs and the 5V and 3.3V generated by two voltage regulator cards near the bottom right. There is no shortage of capacitors on the output, both on the main board and the connector board, which should make for very low ripple. Unusual for a common ATX PSU, the controller chips are all custom parts that make it impossible to find the datasheets.
Looking more closely at the output side, there’s an electrolytic capacitor crammed right up against the heatsink. The design engineer must have been short on time to get the layout out for review since there appears to be plenty of room to space things out. There also appears to be too little thermal grease on the output MOSFETs.
EVGA has been pretty good at gluing parts to guard against vibration, but they seem to have missed the fan control board and the two controller boards. Interestingly enough, the high voltage controller board does have a plastic insulator glued to it (detached when I pulled it back to check what’s underneath) and heat shrink to hold the insulating plastic in place without having to use glue. I suspect somewhere in the making of the assembly instructions, the instruction to glue the controller board to the nearby heatsink got misinterpreted as to glue the insulating plastic to the controller board. The low voltage controller board likewise can use a drop of glue to hold it to the nearby transformer.
Flipping the PCB over, the near perfect reputation of this PSU continues to show blemishes. The excess of solder on one transformer pin and a bit too little on another stick out like a sore thumb along with remains of flux that wasn’t completely cleaned. I will definitely be cleaning it up before reassembling it for future use, but EVGA should take note of it and review their QA checking. We also see that to improve voltage regulation, the voltage sensing is done at the output connector board.
Powering up the bare board with some load reveals that there are significant losses in the transformer as well as in the PFC circuit. Surprisingly for a PSU that uses custom control chips, probing with an oscilloscope shows no attempt to boost PFC efficiency by gating the PFC drive signals at light load. Blame it on the 80 Plus specification that does not provide any incentive to boost efficiency below 20% load for the Gold tier, while the PSUs that do feature PFC gating tend to be server PSUs where light load efficiency does show up on the system power usage specification.
Testing and Results
Onto the testing, the standby power draw was excellent at 0.14W. So little, in fact, that the TP-Link smart plug I was using initially was not able to detect it and I had to switch to a Mooshimeter to make the measurement. Now the adapter to enable the PSU is installed, at which point the idle power draw is 5.7W with eco off and 5.4W with eco on. Or about as much power as a few indicator LEDs, which is not much. It would be more accurate to label it the “silent” switch.
It’s time to see what a trusty oscilloscope makes of it. With the oscilloscope running on batteries to minimize noise and set to a 20MHz bandwidth, we get 7.93mV RMS ripple on 12V, 4.07mV RMS on 5V, and 4.19mV RMS on 3.3V. (For reasons beyond the scope of this review, the noise floor always increases with more bandwidth. Thus a 20MHz limit allows consistent readings when using oscilloscopes of different bandwidths). All of those measurements are done with a 1-ohm power resistor to take the voltage regulators out of a “period skip” mode that would give false readings. Those values are all excellent and are of particular interest to SDR experimenters trying to minimize noise.
Putting it under load, it performed about as well as expected, favoring high loads. The only meter I have capable of measuring such a high current – a Radio Shack clamp meter – reads about 2% low so a correction factor was applied. The voltage regulation is unmatched and didn’t move one bit while loaded – not surprising when looking at the design.
|Iout (12V)||Vout (12V)||Pout||Pin||Eff (%)|
As of October 2018, PCPartPicker lists the EVGA Supernova 650 G3 at just under $60, the cheapest name brand 650W PSU with at least an 80 Plus Gold rating. The cheapest name brand 650W PSU they list is about $48 and only has an 80 Plus Bronze rating, so not worth going cheaper.
The ripple readings are excellent, the voltage regulation is unmatched, and the efficiency is pretty good meeting its rated specifications. I would have liked to see better efficiency at around the 50-100W region where most modern PCs idle, especially considering they went to the expense of using custom control chips.
The dual 6/8 pin GPU cable can put up to twice as much load on the PSU side connector as suggested by the PCIe specifications, there’s an electrolytic capacitor crammed right against a heatsink when it could have been designed with more spacing, the soldering could be improved in a few spots, and the control boards can use a bit more glue to hold them in place.
Overall, I consider it a good PSU, although I would not recommend it for a LAN party PC due to weak spots in vibration damping. For a home or office PC, it is a great choice.