Discussion in 'Article Discussion' started by bit-tech, 22 Nov 2018.
Wait, +12v cables are shielded?
If they won't out-right replace a PSU that they have confirmed has issues with an off-the-shelf gaming GPU (not a fancy enterprise product under very specific loads), then they immediately and permanently go straight onto my naughty list and I won't buy any of their products.
Seasonic's actual words are 'enhanced shielding,' though I guess going from no shielding to any thickness of shielding at all would constitute both an enhancement and the presence of thicker-than-normal shielding!
(I've never used the PSUs in question, so I have no idea whether the default cable is shielded or not, and Seasonic ain't saying either way.)
The issue with Fiji-derived cards (the Fury Nano is a particular culprit) having extremely spikey power draw is well known for SFF use: it's relatively common to use external AC-DC19V supplies combined with internal DC-DC (to get the ATX voltages) PSUs. The DC-DC PSUs are just voltage converters, so handling voltage smoothing with current variation flals to the 19V PSU. Those are invariable laptop bricks (cheap, packaged, and readily available) which are not designed to handle large current spikes, leading to the Vega Nano being particularly infamous as effectively unusable stably with an external PSU setup (unless you go for one of the 'open cage' Meanwell PSUs) . Even a handful of 'real' PC PSUs (e.g. lower powered SFX PSUs) can have issues when the current spikes are above the overcurrent protection threshold as the instantaneous currents demanded by Vega are right on the edge of the ATX current slew spec.
::EDIT:: Herped a derp, Fiji was die-shrunk to Vega not vice versa.
I've not verified this guys claim on reddit but (s)he's claiming that only a 550W PSU would trigger the Vega issue and AMD recommends 650/750W so maybe it's more a case of user error on that one, and (s)he also mentions using a PCIe power cable without capacitors for the Asus 970 issue, i don't have a clue what that means as i didn't even know you could get cables with capacitors.
The 'recommended PSU' ratings for components are written assuming bargain bucket PSUs that fail to confirm the to the ATX standard and print their 'rating' using the peak power output rather than sustained output. i.e. if you take a decent 500W ATX PSU it will output 500W indefinitely. If you take a cheap '500W' ATX PSU it will output 500W for a millisecond (or maybe two milliseconds if you are lucky) and then promptly catch fire.
I wonder if they mean something like additional ferrite-core chokes that some PCI-E 8pin cables have? But the purpose of those is high frequency noise suppression, which I don't think would help this issue a lot.
That may well be so but if you're going to use a 550W PSU when a component in your rig has a minimum requirement of 650/750W then you've only got yourself to blame if it doesn't work.
The new Vega 56 I bought recommends a 750w PSU but also goes on to say your power requirement may be more or less depending on the rest of your system.
Are the GPUs going outside of their own quoted power usage (assuming they actually state what that is)?
The point is with a decent PSU, you only need to worry about average power draw of the GPU, as the smoothing caps in the PSU will handily meet the current spikes, e.g. for a Fury drawing an average of 220W but peaking to 450W you only need to worry about average system power +220W to spec the PSU. With a cheap PSU, the box rating is for the peak draw not the continuous draw, so you'd need to spec based on that 450W peak figure.
This is why GPU (and other component) power figures appear out of whack: a manufacuter will put the average draw on the box as a "look at how little power we need!" smaller-is-better marketing number, while using the peak draw for the PSU requirement to account for users buying the cheapest possible free-with-your-cornflakes PSU.
Define average and define worry?
The simple matter is asking a 550W PSU with only a 45A rating on the 12v to supply 650/750W at 51A for over 10ms is probably going to trip the OCP, in fact I'd want it to shut down versus the alternative of a chinesium PSU frying my components and giving of the blue smoke of death.
Damn I have two Focus Plus power supplies. I'll probably be okay as both machines are APUs and generally not overly burdened. System estimated power draw <300W, PSU 550W.
From the ATX12V v1.31 specs (the most recent I have to hand) there is no defined overcurrent shutoff level, but there is defined a minimum current slew rate a PSU needs to meet (+/-1 amp per us) and transient step size (60% of rated load, e.g. for 600W rated supply on the 12V2 rail that would be 50A, so+/-30A or +/- 360W).
Either that's a typo or it's really out of date, I'll go for the former.
The thing is a standard is just that, it's a base level of quality or goal that's meant to be achieved, no ones going to come and get you in the dark of night if you don't meet it, and no ones going to lock you up if you exceed it.
We're talking about what Seasonic's specifications are here, not what the minimum is as defined by some standard, if everyone build PSUs without deviating from the ATX12V specs we wouldn't have chinesium PSUs and we'd have PSUs without any level of OCP, making OCP rather redundant.
Honestly i don't get what your gripe is, if people use a 550W PSU that's only rated for 45A on the 12v rail to power a component with a minimum requirement of a 650/750W PSU you can hardly blame the maker of the PSU.
The standard PCI-E cables come with caps at the card end.
I have a Focus Plus 750W running an overclocked 4790K and 980ti. So far no random crashing issues. I'm using the standard ATX and EPS cables with caps, but I ditched the standard PCI-E cables and replaced with 18-AWG wire as they were waay too long for the case/mod I have.
Seasonic failing to meet the ATX PSU spec.
Would they be the same ATX PSU specs that you quoted earlier that you appear to misinterpreted, i mean seriously didn't you suspect something was wrong when you claimed that a 600W PSU with a 50A 12v rail should be capable of outputting 960W and 80A.
Or are you going to say that a 600W PSU with a 50A 12v rail should be capable of, as the specs actually define, an increase in amperage of 40% (60% if you've got two rails but the focus plus doesn't so we're not) of the total amperage rating on that rail in no more than 1 microsecond.
In other words your supposed 600W PSU with a 50A 12v rail should be able to increase their amperage by at least 20A/μs, not as you're implying be capable of running higher than it's rated for short durations.
Yes, that is what I said several posts ago and what the ATX PSU spec says:
If the Seasonic units are triggering OCP over intermittent current spikes within the ATX specs, then they are failing to comply with the ATX specs. That's not exactly a controversial position.
My second point was that the "our GPU needs an X hundred watt PSU" recommendations are written assuming the customer will be buying the cheapest garbage that falls off the back of a van, not a spec-compliant PSU. This is also well known.
Separate names with a comma.