NVIDIA GTX 1080 PCB Run Down

[Brian]

With NVIDIA having now officially launched their next generation Pascal based GTX 1070 and 1080 GPUs, there are a number of AIB partner boards coming to the market soon. We have a video today by YouTuber Actually Hardcore Overclocking that examines different custom PCBs and helps break down the differences between the various boards and their power delivery designs. If you've read about additional power phases that will be available on AIB cards that are lacking in Founder's Edition but never quite knew what that meant, this video should help. 

 

 

View full article

[Khenglish]

This guy mostly knows what he's talking about. He was right in ignoring the number of PCI-E power connectors. Those have zero impact on overclocking and more just makes the card more power efficient. He did properly identify phases, which many people don't, but he did annoyingly use the recent Nvidia nomenclature defining phases which is misleading. Ex. when saying 8+2 he's saying there's 8 core phases and 2 memory phases. This is only the recent meaning of that phrase. In the past 8+2 meant you have 10 core phases, with 8 switching in sequence plus another 2 switching in their own sequence, both summing up to deliver current to the core. For example the ATI 4890 was a 3+2 configuration for the core. Minor thing, but using Nvidia's misleading marketing terms annoys me.

 

For bigger things that I think he did wrong, he ignored cap quality and vdroop. Many "overclocking" cards try to have big impressive looking phases with big electrolytic caps that look cool. The problem is of the 3 main cap types you can have the big radial electrolytic caps are the lowest quality. Surface mount electrolytics tend to have superior ESR (effective series resistance. this matters for sudden load changes, or a phase switch) than the round radial electrolytics, but are more expensive and look less cool. Radials do tend to have higher capacitance than SMD, but usually electrolytic capacitance is overkill by nearly an order of magnitude on "overclocking" desktop cards, with ESR being more of a limiting factor. Even better ESR are ceramic caps, but these are the smallest and most expensive. Many of these boards completely lacked ceramic caps for the core on the front of the pcb. Maybe they made up for it on the back, but its obvious that many manufacturers are trying to make their boards look cool.

 

As for vdroop, many of these boards screwed over the memory. All high phase count cards except the zotac threw the memory phases way to the right of the card. This gives the current a long distance to travel, and even worse is the current needs to travel under the core phases, which will already be using many pcb layers. This also hurts core vdroop because pcb layers that the core would otherwise receive are needed to route the memory power. Then add in that 3 memory chips also have data and address lines between the core and power phases, and you have one overly worked pcb.

 

In short, the ASUS looked like crap to me. No top side ceramic caps and no SMD electrolytics. The powercolor card looked good for air and H20 with a good ceramic cap count, but seemed to lack the FETs to push the current or LN2. The EVGA should be good for LN2 if the backside makes up for the front side's lack of ceramics. Its electrolytic count is insane with tons of SMD caps, but it looks like it will have vdroop problems. The MSI looks decent with a good chunk of ceramics and high phase count, but lacks SMD electrolytics.

 

I don't get why he gave the Zotac so much crap. This card looked to be the best to me. That memory phase location is great. The PCI-E slot can route current up the left side so it can get a strong 12V connection that avoids messing with the core power side of the card and memory data and addressing. The Zotac will have the best vdroop of any card. It has a good count on ceramics and a few SMD electrolytics. Yes the FETs COULD be poor, but they could also potentially be better than the high/low combo chips on most cards.

 

Honestly the 25A high/50A low chip that he said Nvidia is using sounds like crap. Having a lower high side current limit than low side makes no sense. Yes high side conducts current way less often than low side so you want a smaller faster switching FET, but if you exceed that 25A the FET can still blow up. If the low side conducts 50A when it is on, then the high side also conducts 50A, so this chip is only rated for 25A due to the high side. I can see why manufacturers would avoid this chip. I'm surprised I didn't see the TI 83750 high/low combo chip on any cards as this has a 40A rating and is very commonly used on mobile cards.

[thelastpretorian]