Nvidia Ampere Discussion [2020-05-14]

Carry on then, my memory was wrong!

 

I would think that’s set in the bios and the aib would be setting board limits.

 

RTX 3080 CTD Issues

https://www.guru3d.com/news-story/g...ely-due-to-poscap-and-mlcc-configuration.html

 

I have 3080 FE I have been experimenting with. The card was very unstable until I switched my power supply (Corsair HX1000) from multi-rail 12V to single-rail 12V, so the very high short term power draw seems to be an issue. Max boost at stock looks like it is 2100 mhz (as reported by nvidia-smi), but I have never seen over 2055 reported. Capping max boost at below 2000 mhz seems to be stable. Also, locking the card at max P state seems to be stable (using nvidia-smi -lgc 2100). Idle power draw (TBP) goes from about 28 watts to 59 watts when locking the max P state.

 

So I suppose this means Asus models can't be ruled out either, and instability reports for them are real.

Also indicates Igor's analysis was spot on, and there are voltage drops due to noise when switching frequencies around peak load scenario.

Which also means that the specific part list on the different models only makes a different with regard to how high the "peak load" may go before triggering the issue, but not a single model on the market is safe.

And the only solution is going to be to put an actual hard cap on the boost clock, below the point where the noise from frequency switching couldn't be filtered. Even if the loss of the top boost steps results in a measurable performance loss.

 

Fully stable after switching to single-rail or did you still need to lower clocks/mess with P-states?

I'm also running a 3080 FE, seeing the same 2055MHz max as you and fully stable so far.

 

I thought 12V multi-rail power supplies had died out with the dinosaurs.

 

So oddly enough I still have some instability when using the recommended setup - two separate 8-pin PCIe power cables going to the adapter, but has been 100% stable with a single 8-pin PCIe daisy chain power cable going to the adapter. I'm not sure if this means it an issue with the power supply or if there is an issue with power balancing on the 3080. The power supply has been fine with everything else I have thrown at it.

The power supply is not really multi-rail, but can apply current limits per rail - hence the switch.

 

How many amp the 12v rail delivers ?

Right now, it's not 100% clear what is causing the instability, so don't change your PS right now imo...

 

At this point it seems like to have the least headaches one should return their RTX 30x0 video cards since its still that return window? Otherwise you're at the mercy of whatever the manufacturers decide to do later on with recalls?

 

Its always wise to await brand new products a little while, its always they reach the masses some trouble can come to light. Also, maybe good to see what AMD has to offer. If they can give me navi21 with its ballpark 25TF at a cheaper then 3080 price, il get that. Somehow i feel thats gonna happen.

 

Could RTX 3080 based upon GA102 have been a placeholder for GA103?:

 

Not just the issue of recalls which may not happen, but also the risk of a driver side patch slashing performance by an unknown amount. Which will apply even to those units which are running (mostly) stable.

Plus, as the current shortage has likely been caused by some AIBs not catching the issue pre-production but having to rebuilt their launch day inventory, prices are likely to normalize by the time the cards become properly available at all.

May end up even cheaper if AMD delivers, or same price if they don't. With AMD and NVidia sourcing different production lines, effective availability is improving either way.

If your haven't gotten your card yet, cancelling the order is realistically the only sane option.

Ifyou already got your card, it depends on how stable it runs for you. If the card doesn't run with your power supply, refund is also a better choice than to buy a new power supply, given how prices for these have spiked as well.

All in all, only makes sense to keep if running perfectly well for you. The massive power draw is also something to consider for the next summer season.

And on a plus side, every refund is also going to hurt the scalpers who have ridden the launch.

 

Surely any GA103 based anything is at least 6 months away.

 

Lots of users reporting reduced boost clock, some report reduced default voltage, consistent reports of less stability issues. Looks like boost clocks of 2.1 Ghz have been disabled (for cards with stock clocks), and the cards are in general boosting far more conservatively. Looks like 2.0 GHz boost steps have also been disabled by application profiles (?), but not entirely.

Still some reports for crashes in games utilizing RT though.

Take the positive bias with a grain of salt though, some reports of ongoing crashes have been deleted from that thread by moderators of that subreddit.

https://videocardz.com/newz/msi-qui...-gaming-x-trio-design-amid-stability-concerns
Meanwhile MSI has also changed their part list.

EDIT:
Originally stated reports of "increased performance", all proper measurements taken indicate the opposite, it came with an on average 1-2% performance loss.

 

Too late, I already already swapped in a different model 1000W PSU. It seems stable so far. The previous PSU is rated for 40A per 12V plug (480W) with over current protection enabled. I was seeing some resets that I am pretty sure were the OCP kicking in, but only when the GPU was boosting from a low power state to a high power state, not under sustained load. Unfortunately, I don't have the equipment to measure the peak power draw, but the evidence leads me to believe it is higher on the 3080 than any previous GPU I have run on the PSU. Igor's Lab has some peak power draw measurements in their 3080 review, but it looks like they only measured under a sustained load (peak was 486W for <1ms).

It is hard to say if this generalizes to the issues other people are having, but peak power draw when boosting from a low power state to a high power state could certainly be a factor.

 

Have you tried GPU-Z to measure the power draw?
https://forum.beyond3d.com/posts/2160762/

 

Let me guess, it'll require a 50 gigawatt power supply?

 

I think you got that backwards, a GA-103 would probably be a smaller die than GA-102 and hopefully more power efficient all the while providing the same number of SMs as the scaled down part used in 3080 now.

 

I would be curious to see how a 64SM, 8GPC, 128 ROP card stacks up to the 3080.