4 ms·

The GPU runs at 1V (sub 1V usually). The issues with the power delivery and the nvidia connector are their own.

by xxs 3y ago

CamperBob2 3y ago
That's not how it works. The GPU receives power at 12V via the PCIe 8/6+2 connector at 20-30 amps or so. It then steps that down to ~1 volt at the point of load, at hundreds of amps, but that's irrelevant. The problem is that safely handling "20-30 amps or so" cannot really be done with cheap consumer-level Molex connectors, or at least should not be done, ideally. More expensive connectors are needed, so what they are doing now is asking for trouble. However, the voltage doesn't determine how hot it gets; only the current influences that. If they were distributing power at 48 volts, the current would be about 1/4 what it is now, and overheating at the connector would not be an issue.
- xxs 3y ago
  says 600W GPU, that's 50A. Screw based connectors would do just fine - delivery 48V would require to step it down to some other (sub 20) voltage first. That part would be at the GPUs. Also I do understand how it works.
  - CamperBob2 3y ago
    Yes, certainly, using better connectors would head off overheating problems. That's not free. Something else that isn't free is extra copper in the wire harnesses from the PSU. Meanwhile, it's true that point-of-load regulation from 48V directly to 1V will be less efficient (well, maybe; see nimish's link below). But by the same token, regulation in the PSU would be more efficient if it didn't have to go all the way down from 150-300 VDC on the primary side down to 12 at the output. A good compromise would probably be the geometric mean between a few hundred volts and the voltage at the load with the highest current. Assuming 250 volts on the primary as a compromise between 120/240VAC operation and 1V all the way downstream, that would be about 16 volts. Doesn't seem like a big improvement, but it would be enough to lower the current at the GPU input connector by about 1.8x compared to 12 volts. Either way, sticking with 12 volts in any new standard is just stupid. 48 has another advantage, which is that automotive is going to (finally) start moving to it over the next few years. There could be a lot of room for parts-sharing between the two industries.
    - sokoloff 3y ago
      I’ll take the over on 3 years for automotive to switch to 48VDC.
  - amluto 3y ago
    At 50A, those screw connectors should be properly torqued. 50A is serious current. (20A is more than enough to start fires due to small-scale arcing.)
    - xxs 3y ago
      >properly torqued Indeed - drop the phillips head already. Hex or torx, please. 50A is really on the very high side, though.

The GPU runs at 1V (sub 1V usually). The issues with the power delivery and the nvidia connector are their own.

by xxs 3y ago

The GPU runs at 1V (sub 1V usually). The issues with the power delivery and the nvidia connector are their own.

That's not how it works. The GPU receives power at 12V via the PCIe 8/6+2 connector at 20-30 amps or so. It then steps that down to ~1 volt at the point of load, at hundreds of amps, but that's irrelevant. The problem is that safely handling "20-30 amps or so" cannot really be done with cheap consumer-level Molex connectors, or at least should not be done, ideally. More expensive connectors are needed, so what they are doing now is asking for trouble. However, the voltage doesn't determine how hot it gets; only the current influences that. If they were distributing power at 48 volts, the current would be about 1/4 what it is now, and overheating at the connector would not be an issue.

xxs 3y ago
says 600W GPU, that's 50A. Screw based connectors would do just fine - delivery 48V would require to step it down to some other (sub 20) voltage first. That part would be at the GPUs. Also I do understand how it works.
- CamperBob2 3y ago
  Yes, certainly, using better connectors would head off overheating problems. That's not free. Something else that isn't free is extra copper in the wire harnesses from the PSU. Meanwhile, it's true that point-of-load regulation from 48V directly to 1V will be less efficient (well, maybe; see nimish's link below). But by the same token, regulation in the PSU would be more efficient if it didn't have to go all the way down from 150-300 VDC on the primary side down to 12 at the output. A good compromise would probably be the geometric mean between a few hundred volts and the voltage at the load with the highest current. Assuming 250 volts on the primary as a compromise between 120/240VAC operation and 1V all the way downstream, that would be about 16 volts. Doesn't seem like a big improvement, but it would be enough to lower the current at the GPU input connector by about 1.8x compared to 12 volts. Either way, sticking with 12 volts in any new standard is just stupid. 48 has another advantage, which is that automotive is going to (finally) start moving to it over the next few years. There could be a lot of room for parts-sharing between the two industries.
  - sokoloff 3y ago
    I’ll take the over on 3 years for automotive to switch to 48VDC.
- amluto 3y ago
  At 50A, those screw connectors should be properly torqued. 50A is serious current. (20A is more than enough to start fires due to small-scale arcing.)
  - xxs 3y ago
    >properly torqued Indeed - drop the phillips head already. Hex or torx, please. 50A is really on the very high side, though.