Okay or no way? AMD RX 7800 XT with 560W PSU

[Ferrum Master]

The issue of course is whether the PSU is actually a quality unit with properly-set OCP!

Pretty hard set in stone, ATX standarts for that. It should not be longer, as in case of a real semiconductor failure, when the peak is past and smoke rises, resistance drops, it was too numb to detect the short, then the traces pop and melt, the PSU will not shut down and keep burning the PCB. What's so proper about it? It is a defensive mechanism not a feature. OCP is set way too high for many PSUs, actually the OPs Fractal PSU is also one of them.

It's not so bad.

This only works if you need more speed at the same wattage. If you wanna lower the wattage you need to lower the maximum clock as well. Usually, lowering it by 200 reduces wattage by 25 to 40 percent.

Depends on SKU. Also all that undervolting Mojo.... in reality all Radeons I've undervolted + underpowered recently were unstable doing it. Yes it works in some games, but it ain't stable. It is silicon lottery. Maybe you have a golden sample, okay... but for most, nope... then again we have full forums, hey my Radeon keeps crashing.

 

[Beginner Micro Device]

in reality all Radeons I've undervolted + underpowered recently were unstable doing it

I also spent a lot of time finding stable numbers. Yet it's satisfying to have your GPU draw ~10 percent less energy whilst performing at the same level. You can also decrease the power draw without touching the voltage curve at all.

hey my Radeon keeps crashing.

Yeah, AMD could do a way better job manifesting the frequency-voltage curves.

 

[Chrispy_]

Pretty hard set in stone, ATX standarts for that. It should not be longer, as in case of a real semiconductor failure, when the peak is past and smoke rises, resistance drops, it was too numb to detect the short, then the traces pop and melt, the PSU will not shut down and keep burning the PCB. What's so proper about it? It is a defensive mechanism not a feature. OCP is set way too high for many PSUs, actually the OPs Fractal PSU is also one of them.

I think you're confusing OCP and OVP. OVP is binary - too much voltage can instantly fry components so it needs to be a binary and immediate cutoff.

Too much current simply generates more heat, which is why it's not necessary to panic for small, brief excursions. If something is rated for 10 amps 24/7 for years on end, it's not going to suddenly explode if it gets 15 amps for 50 milliseconds. The rate at which the components heat up to the point of damage is much like a fuse. A 10 amp fuse doesn't instantly blow at 11 amps. It doesn't blow until several whole seconds at 15 amps and to get "instant" blow in the sub-100ms range, you probably need 20-30A or more to generate enough heat fast enough to reach the plasticity temperature of 10A fuse wire.

 

[Dr. Dro]

Yeah, AMD could do a way better job manifesting the frequency-voltage curves.

OT, but I often wonder if there is anything that AMD could not do a way better job at sometimes. It's frustrating.

 

[Ferrum Master]

Yeah, AMD could do a way better job manifesting the frequency-voltage curves.

You are forgetting that it is temperature dependent too, AMD doesn't dictate FAN curves for OEMs to use. You get the idea who to blame really?

I think you're confusing OCP and OVP. OVP is binary - too much voltage can instantly fry components so it needs to be a binary and immediate cutoff.

Certainly not. Totally different case.

I guess if you haven't seen a different class of fuses, as those vary a lot by speed, for each application and device there is a proper one, those are not made equal unless you put a crowbar in the fuse box.

But here we a looking at voltage curve rise across the internal shunt and the watchdog that monitors it. Regarding other statements about current generates more heat etc... you need to refresh some basic electronics. I am not sure what's that about, but it ain't related here.

We are talking about protection. Catastrophic failure means a short, short means infinite current doing simple math. How OVP is connected here? UVP, okay, then some math aligns. So when a semiconductor failure event occurs it is a short, but as everything burns fast as it wants to consume infinite current, explodes, pops, the short turns into resitor(it means current drops down), and the PSU treats it as a normal load and keeps working. They key word is in the timing how fast. The ATX standart is calculated and manufacturers are disobeying it. It is a flaw.

On most parts it needs to be software configured in the watchdog IC, but most just don't do that, contrary to decrease RMA rates they set it too high to get rid of some clients using their PSU's on the limit, simple as that.

Now while I am writing we have a second fanboy rant... lol...

 

[freeagent]

OT, but I often wonder if there is anything that AMD could not do a way better job at sometimes. It's frustrating.

Change their name back to ATi and it will sort itself out

Even some Nvidia fanboy came in.

Why do you gotta be like that? Maybe you need to take a break?

 

[Beginner Micro Device]

You are forgetting that it is temperature dependent too

Spoiler: The dies themselves are within safe, sometimes overly safe values.

Cut-down GPUs usually have a massive overkill in terms of VRM capacity. VRAM? Overheating of this part is usually an RTX 3090 meme, not the case for most latest AMD GPUs and non-3090 nVidia ones.

The main concern still is pushing too much voltage and/or current upon the GPU at the peak load and insufficient delivery under light loads which leads to massive power spikes when GPU is 100% utilised and to crashes when load isn't as apparent. Wild temperatures are a rare case.

 

[Ferrum Master]

Why do you gotta be like that? Maybe you need to take a break?

I cannot see how praising other maker or even bashing AMD helps understanding PSU compatibility and qualities with certain AMD GPU. Or I am not understanding something about moderation.

Cut-down GPUs usually have a massive overkill in terms of VRM capacity. VRAM?

It is much more... hotspots are the new kids in the block also. Quality, evenness, bending, Fan Curves, ambient temps, airflow... it all contributes for the clock stability, and it massively differs from user to user even for cut down SKU's.

 

[freeagent]

I cannot see how praising other maker or even bashing AMD

No one was bashing AMD? I certainly was not, the person who suggested a 4070 instead certainly was not..

We know these GPUs spike pretty hard, not a new concept. But if he runs a low core count CPU, with very little attached, we just said he might be ok.

But lets be real, at under 600w he will have a problem at some point.

I do not understand why so much emotion has to be placed into hardware. It is just hardware.

 

[Dr. Dro]

My suggestion for the 4070 was with power efficiency in mind + the studio drivers optimized for the OPs productivity suite which AMD doesn't provide.

Said characteristics make of the 4070 a better choice of product if gaming isn't the strict focus, it won't run a risk of overloading the power supply and behave better in these applications. The 50 watt differential here correspond to almost 10% of the PSU's capacity.

Arguably, the 7800 XT is a better card for games, as long as you're not too keen on RT, but in this segment it's non issue, IMO.

It wasn't a fanboy pitch; and while I do have my differences with AMD as of late (especially since they can't stop bungling it at every turn), it was a mere suggestion.

 

[freeagent]

I know my Ti can do about 350w, with swings up to 400w, with my OC, I cant imagine the vanilla being too far behind.

This is not usually in game though, but with F@H. Though some benchmarks can push it close.

 

[Ferrum Master]

I do not understand why so much emotion has to be placed into hardware. It is just hardware.

I work daily in RMA business and see bad choices made by people not understanding basic things. It is not only about it should work.

If others are ignorant, okay. But at least someone will re read it after googling and understand what's what especially regarding safety.

Regarding nvidia and complaints about AMD, well I see it totally out of place for this topic.

 

[Dr. Dro]

I work daily in RMA business and see bad choices made by people not understanding basic things. It is not only about it should work.

If others are ignorant, okay. But at least someone will re read it after googling and understand what's what especially regarding safety.

Regarding nvidia and complaints about AMD, well I see it totally out of place for this topic.

Maybe, but I'd argue that it's constructive to the topic to recommend an alternative which is arguably a better fit for OP's usage pattern and profile. They are a casual gamer who occasionally work with 3D modeling on their system. To that end, the Nvidia card has better driver support, more gaming features, is a better productivity solution and uses less power, which means that it's useful advice to recommend this alternative. By all means if they're willing to spend additional money replacing the power supply to accommodate the Radeon's increased power footprint and are fully aware that they will have to accept certain compromises by making this choice.

More power = higher strain on the power supply = higher temperatures = less stability, and it's already treading a fine line here, IMO. OP's case seems to be a Lian Li Air Mini, so while it's a small case, it has adequate airflow, so this may not be that much of a problem, for now.

Either way, this thread seems to be a couple of months old, so I'm unsure if OP already made a decision of their own regarding this.

 

[Freeco]

Any updates? Would be interested in hearing about your results

I have my Powercolor Hellhound RX7800XT a couple of weeks now.
As mentioned I did already do a 100mV undervolt of my CPU earlier, and also undervolted the GPU to 1060mV while running it in silent mode.
--> No problems what so ever with my 560W PSU... Everything runs rock solid.

 

[maxfly]

I have my Powercolor Hellhound RX7800XT a couple of weeks now.
As mentioned I did already do a 100mV undervolt of my CPU earlier, and also undervolted the GPU to 1060mV while running it in silent mode.
--> No problems what so ever with my 560W PSU... Everything runs rock solid.

Thanks for coming back and updating your experience. I had literally just started typing to ask if you ever came back to say how things worked out for you.

 

[Lore]

I have the same psu and a RX 7800XT . No problems here as well.

 

[Chrispy_]

Certainly not. Totally different case.

We are talking about protection. Catastrophic failure means a short, short means infinite current doing simple math. How OVP is connected here? UVP, okay, then some math aligns. So when a semiconductor failure event occurs it is a short, but as everything burns fast as it wants to consume infinite current, explodes, pops, the short turns into resitor(it means current drops down), and the PSU treats it as a normal load and keeps working. They key word is in the timing how fast. The ATX standart is calculated and manufacturers are disobeying it. It is a flaw.

On most parts it needs to be software configured in the watchdog IC, but most just don't do that, contrary to decrease RMA rates they set it too high to get rid of some clients using their PSU's on the limit, simple as that.

You appear to be mis-remembering or misinterpreting the ATX specification.

There's no point repeatedly arguing "the ATX standard" in your defence because the ATX standard specifically has a detailed section with formula for exactly how far a PCIe CEM device such as a graphics card can exceed the OCP limit and for how long. The standard absolutely, 100% includes precisely what you're arguing it does not have:

https://www.cybenetics.com/attachs/52.pdf​

Page 21 and 21 of the ATX standard​

Table 3-1: PCI Express* CEM Add-in Card Power Excursion Limits Table(45)​

I was paraphrasing earlier, but I'll say it again. Going over the OCP limit with brief transient current spikes is not an issue, and if it causes an issue then the PSU is faulty by design because the OCP isn't up to the ATX spec you keep quoting as gospel. It clearly has a table of logarithmic formulae and a graph defining, essentially that the larger the excursion beyond the limits, the shorter the duration of the trigger can be.

You can see that a 10ms transient spike is allowed to be 50% higher than the rated current, and a 20ms transient is allowed around 40% higher currents before OCP can be triggered. Table 3-3 on page 24 is more relevant to this post as it's a total system power draw that says a PSU has to handle power spikes at 160% of its rated load for 10ms and 120% of its rated load for 100ms.
It's literally the spec.

edit:
terrifyingly, for a 600W graphics card, 1800W is permitted, albeit only for 0.1ms
:O

 

[Dr. Dro]

I have my Powercolor Hellhound RX7800XT a couple of weeks now.
As mentioned I did already do a 100mV undervolt of my CPU earlier, and also undervolted the GPU to 1060mV while running it in silent mode.
--> No problems what so ever with my 560W PSU... Everything runs rock solid.

I'm glad it worked out for you in the end. Enjoy!