do you think AMD will restrict SPPTs even more in the future?

[GerKNG]

a 6900XT is bascially unable to overclock at all with it's heavily restricted powerlimits.
250 or 330W the card runs into a brickwall out of powerlimit and performs basically identical to a decent undervolt.
with MPT i pushed my nitro + from 303W to 375W and it actually runs 250Mhz higher than before (still powerlimited)

but since AMD has cut down the SPPT modding ability by a lot compared to RDNA1 do you think they will maybe lock it completely in the future? (like nvidia)

 

[eidairaman1]

What is sppt?

a 6900XT is bascially unable to overclock at all with it's heavily restricted powerlimits.
250 or 330W the card runs into a brickwall out of powerlimit and performs basically identical to a decent undervolt.
with MPT i pushed my nitro + from 303W to 375W and it actually runs 250Mhz higher than before (still powerlimited)

but since AMD has cut down the SPPT modding ability by a lot compared to RDNA1 do you think they will maybe lock it completely in the future? (like nvidia)

Actually the problem is neither company allows bios mods now. As of GTX 1000 and Vega, you cannot alter bios code, you can only cross flash bios as long as certain criteria in each bios matches up. It is why many were successful with my assistance to flash 5700s into 5700XTs.

Nvmd, i just realized what sppt is. Just review igorslab from time to time

 

[The red spirit]

but since AMD has cut down the SPPT modding ability by a lot compared to RDNA1 do you think they will maybe lock it completely in the future? (like nvidia)

I still don't understand what's the point for them to restrict that, it's not like they are losing money or sales due to that. All they do is pissing off some tweakers, who would tweak their cards in some other ways then.

 

[delshay]

What is sppt?

SPPT= Soft Power Play Table.

@Thread

Yes they should lock it to reduce RMA.

 

[nguyen]

I still don't understand what's the point for them to restrict that, it's not like they are losing money or sales due to that. All they do is pissing off some tweakers, who would tweak their cards in some other ways then.

Market segmentation, AMD wants people to buy the binned 6900XTXH chips that are selling for >2300usd.

but since AMD has cut down the SPPT modding ability by a lot compared to RDNA1 do you think they will maybe lock it completely in the future? (like nvidia)

What are you talking about? I bypassed the power limits on my 1080Ti, 2080Ti and 3090 by flashing the XOC BIOS (3080Ti also has XOC BIOS).

 

[Mussels]

Market segmentation, AMD wants people to buy the binned 6900XTXH chips that are selling for >2300usd.



What are you talking about? I bypassed the power limits on my 1080Ti, 2080Ti and 3090 by flashing the XOC BIOS (3080Ti also has XOC BIOS).

we can only crossflash, not bios mod


that said, AMD will probs end up in that exact same situation, preventing crossflashing is a time consuming process, and makes BIOS updates for bugfixes/features really hard

 

[The red spirit]

Market segmentation, AMD wants people to buy the binned 6900XTXH chips that are selling for >2300usd.

Most people will just buy those if they want, they certainly won't risk bricking their card by flashing it. I personally flashed my card to make it quieter and to consume less power (aka fixing PowerColor mistakes). I personally would love to see something as silly as 6900 LE version, which has same powerful core, but has a lot lower power limit, so that it runs very efficiently and a lot cooler. For me all this gamer marketing shit is irrelevant and horsepower race is pretty much over. If AMD doesn't want to make stuff that I want, at least they shouldn't be dicks about letting me modify it. When I buy a card, I only want hardware, not some specific tune with it too.

 

[GerKNG]

Market segmentation, AMD wants people to buy the binned 6900XTXH chips that are selling for >2300usd.



What are you talking about? I bypassed the power limits on my 1080Ti, 2080Ti and 3090 by flashing the XOC BIOS (3080Ti also has XOC BIOS).

show me a 3080 with no voltage cap and unlimited powerlimit.

AMD usually does this without even flashing a bios.

 

[nguyen]

Most people will just buy those if they want, they certainly won't risk bricking their card by flashing it. I personally flashed my card to make it quieter and to consume less power (aka fixing PowerColor mistakes). I personally would love to see something as silly as 6900 LE version, which has same powerful core, but has a lot lower power limit, so that it runs very efficiently and a lot cooler. For me all this gamer marketing shit is irrelevant and horsepower race is pretty much over. If AMD doesn't want to make stuff that I want, at least they shouldn't be dicks about letting me modify it. When I buy a card, I only want hardware, not some specific tune with it too.

High-end GPUs nowadays come with 2 BIOSes, you can use the Quiet BIOS with much lower PL for 24/7 usage and the XOC BIOS for some occational overclocking fun (or use your PC as a heater during winter).

 

[mtcn77]

and performs basically identical to a decent undervolt.

Yes, this is why tweakers tweak. Not because they can do better, but because AMD cannot do well by default. When they do listen to us and fix SPPT tables for good, the chips won't need further modification. We aren't at 100%.
The issue is the software controlling the gpu is not learning, like an analog vrm current does to sustain a given voltage. If AMD had a SPPT that indeed worked, it wouldn't require further user modification, but it is always possible the fringe case that they do it out of product personalization.
PS: the chips are already enhanced with circuitry to monitor vdroop events and can time phase clocks according to that. What is missing is the chip isn't cutting back voltage up to this point to enable more power budget at up top frequencies. So it is detuned by default, that is the overwhelming issue. You are just negating the whole profile table by manually adjusting the top profile.

 

[GerKNG]

yes undervolting is the best that you can get out of your card at the moment.
because it misses like 75W out of the box.

stock clock, 2800 Mhz, 3000 Mhz... always running into a brick wall of powerlimit with same clocks as stock.
undervolting gives you bascially stock speeds back with 20W less power consumption.

but using MPT and giving a 6900XT 350W or 375W actually let the card run as it wants to. almost 2.7 Ghz average in Timespy instead of 2.35-2.4.

 

[The red spirit]

High-end GPUs nowadays come with 2 BIOSes, you can use the Quiet BIOS with much lower PL for 24/7 usage and the XOC BIOS for some occational overclocking fun (or use your PC as a heater during winter).

But I don't care about that. I want LE (low energy) BIOS, which pushes 6900 or equivalent card to let's say 150 watts.

 

[GerKNG]

But I don't care about that. I want LE (low energy) BIOS, which pushes 6900 or equivalent card to let's say 150 watts.

well then buy a 6700xt and run it stock... same performance, saved around 700 bucks.

 

[The red spirit]

well then buy a 6700xt and run it stock... same performance, saved around 700 bucks.

Bigger core at lower clocks will always be more efficient and still faster than smaller core at higher clocks. Anyway, it's just an example. The point is that clamping down on BIOS modding is just bad.

 

[GerKNG]

just an example of how bad the RDNA2 cards actually are for anything except stock/undervolting.
stock undervolt vs 365W max stable OC in a game.
still running into an insanely heavy powerlimit with zero gains in performance.

 

[OneMoar]

the power limit is there for a reason
you can only draw so much power from PCIE connectors and unless you want your gpu to have 3 8Pin connectors 300`ish watts is all you have
and thats not even discussing weather or not the power supply can safely deliver 400W you need to realise just because the psu is 1000w or more doesn't mean you can run huge amounts of power down what is effectively 3 wires (3 hot 3 ground two sense)

this means you must make sure that the user connects the correct amount of cables to the power supply and doesn't daisy chain connectors else you will have a fire
remember gpu PPT is just the package there is anouther 20-35W there not measured via that sensor

the engineers that designed these are way smarter then you and have a wide variety of Reliability Safety Usability concerns other then make it go fast

 

[GerKNG]

the power limit is there for a reason
you can only draw so much power from PCIE connectors and unless you want your gpu to have 3 8Pin connectors 300`ish watts is all you have
and thats not even discussing weather or not the power supply can safely deliver 400W you need to release just because the psu is 1000w or more doesn't mean you can run huge amounts of power down what is effectively 3 wires (3 hot 3 ground two sense) this means you must make sure that the user connects the correct amount of cables to the power supply and doesn't daisy chain connectors else you will have a fire
remember gpu PPT is just the package there is anouther 20-35W there not measured via that sensor

two 8 pins =300W
(and since you love being in spec)
PCIe Slot 12V 66W, 3.3V 9W
150+150+66 = 366W
that's why i used 365W (fully in spec)


and please do not tell that anybody else at nvidia and their boardpartners (almost all board partner cards with 2x8 connectors are far out of spec)
even AMD with their 295x2 that pulls 430W out of 2 8 pins out of the box.

 

[OneMoar]

two 8 pins =300W
(and since you love being in spec)
PCIe Slot 12V 66W, 3.3V 9W
150+150+66 = 366W
that's why i used 365W (fully in spec)

and you are 100% sure that every hardware combo will tolerate being run exactly on the edge of the spec ?
so with vram and everything else not measured by the PTT sensor you are already pushing 330W

no good engineers leave margins for a reason if you wanna cut the margins off and push it to the bleeding edge thats on you but never at the cost of everybody else

 

[GerKNG]

and you are 100% sure that every hardware combo will tolerate being run exactly on the edge of the spec ?
so with vram and everything else not measured by the PTT sensor you are already pushing 330W

no good engineers leave margins for a reason if you wanna cut the margins off and push it to the bleeding edge thats on you but never at the cost of everybody else

and please do not tell that anybody else at nvidia and their boardpartners (almost all board partner cards with 2x8 connectors are far out of spec)
even AMD with their 295x2 that pulls 430W out of 2 8 pins out of the box.

and btw.. we talk about overclocking and not complaining about 5 watts.
i am happily running this card with 420W Powerlimit out of 2 8 pins with a dark power pro 12 titanium.

 

[OneMoar]

and please do not tell that anybody else at nvidia and their boardpartners (almost all board partner cards with 2x8 connectors are far out of spec)
even AMD with their 295x2 that pulls 430W out of 2 8 pins out of the box.

and btw.. we talk about overclocking and not complaining about 5 watts.
i am happily running this card with 420W Powerlimit out of 2 8 pins with a dark power pro 12 titanium.

so what exactly is your complaint your already 70+W over the spec what do you expect to gain in performance by adding anouther 25 or 35w ?
except maybe a visit from the fire department

your obsessing's over the power limit your thinking is its power limited it must be-able to go much faster that is simply not the case
at a certain point in the power/frequency curve the efficiency will take a noise dive, you are pretty close to that limit
now LN2 and the more extreme overclocking that is a differing kettle of fish

 

[Mussels]

Joining in on the power argument: My 3090 is locked to 300W (minimum voltage of 737mv @ 1.5GHz) and it burned out my goddamn PCI-E power cables.
Yes, these power limits exist for a reason, on dual cable cards.

My factory corsair cables are still safe, but it's a concern for these guys that any budget PSU's could catch fire

 

[OneMoar]

Flametastic

 

[Kabouter Plop]

Cables are rated for certain power limit to, settings a higher power limit on gpu past what AMD allows may not be a good idea

 

[delshay]

I still don't understand what's the point for them to restrict that, it's not like they are losing money or sales due to that. All they do is pissing off some tweakers, who would tweak their cards in some other ways then.

It's to stop user(s) blowing up they cards, then RMA it as if it's somebody else's fault..

When you increase the power, the MOSFETS will run hotter. Because the circumstances of the MOSFET has changed, you need to know can the current cooling plate & the thermal pads deal with this new level of heat. Better thermal pads with good pad compression can help, but you still need to know if the cooling plate can deal with it.

 

[The red spirit]

It's to stop user(s) blowing up they cards, then RMA it as if it's somebody else's fault why the card went faulty.

When you increase the power, the MOSFETS will run hotter. Because the circumstances of the MOSFET has changed, you need to understand can the current cooling plate & the thermal pads can deal with this new level of heat. Better thermal pads with good pad compression can help, but you still need to know if the cooling plate can deal with it.

If you are tweaking your hardware at all, something like this should be obvious. When you tweak, you pretty much accept that if you fuck up, you fuck up and there's no RMA. If you want RMA, you don't tweak. I would rather have some e-fuse, which detects me being a knob and makes RMA personnel not accept my card for free repairs.