NVIDIA releases CMP lineup and reduces hashing rates on GeForce cards

[porina]

1 hour ago, Craftyawesome said:

Nintendo couldn't patch the bug on existing switches. No one said they couldn't release new ones with a fix.

Without looking up the details, from memory they software patched it so that after the patch was installed it couldn't be exploited any more. You had to have a switch with software before that point to get around it, and never update the software again.

 

4 minutes ago, Moonzy said:

Edit: so the cards launches on the 25th, I'm wondering if people would crack the code before then

Presuming there are review samples out there with a pre-release driver. We'll have to wait and see.

[Craftyawesome]

45 minutes ago, porina said:

Without looking up the details, from memory they software patched it so that after the patch was installed it couldn't be exploited any more. You had to have a switch with software before that point to get around it, and never update the software again.

The RCM exploit is a bootrom exploit. This is only modifiable at the factory. If you have a switch from before summer 2018, you can hack it.

 

There are software exploits that got patched, but these weren't claimed to be unpatchable.

[Kisai]

On 2/18/2021 at 6:40 AM, itswillum said:

They're also using their drivers to limit the hash rate when mining-type algorithms are being run on the new RTX card. This is in an effort to differentiate between their gaming series of cards and cards meant for miners.

 

 

I called this happening three months ago. The only way they're going to "hurt" miners is by making the cards be much less efficient, which they can only do in the driver unless they're going to put it into the firmware. In which case who knows what other kinds of loads it might impair.

 

 

[Wh0_Am_1]

On 2/18/2021 at 8:51 AM, BuckGup said:

Aw yes, someone in a higher up corporate position thinks they know more than the average tech enthusiast or engineer lower than them. It'll probably take less than a week for a fix to come out and all it does is put more people in danger by using unofficial low level software/drivers. NVIDIA the way you're meant to be played

Don't even need to mention (but going to anyway) that since the cryptocurrencies that are eating GPUs are designed to be run on consumer hardware, versus ASICs that large companies use for Bitcoin so that no one entity has too much control over the network. All of these currencies are going to alter their algorithms to be more efficient on the new hardware.

[ewitte]

The latest I read is this only hurts ETH mining performance there are other algorithms that can still make $7/day.  Waste of time.  Too much money to be made that will be thrown at anything Nvidia comes up with.

[Wh0_Am_1]

On 2/18/2021 at 12:44 PM, Rym said:

A splendid solution. Furthermore, they should do hardware modifications for any new RTX gaming GPUs they'll manufacture to make it impossible to circumvent these changes.

 

You want a gaming GPU? Buy an RTX 3080.

 

You want a mining GPU? Please, this way sir.

 

It is how it's meant to be. An RTX 3080 was never designed for mining. But special mining GPUs can be created instead. Like an RTX 3080, but with zero gaming performance, all about that mining performance.

Yeah, um the mining algorithms are going to circumvent the alterations sooner or later, as Ethereum, and Monaro, two of the major ones that have traditionally been eating cards have switched algorithms to prevent ASICs from being used, and at that point mining GPUs will basically become ASICs. Why are they doing this? Simple, the goal of these currencies is to be as distributed and decentralized as possible. Which means that their mission is to prevent large entities from having too much power over the network, to prevent them from intentionally crushing the system on a whim. Nah in the end this is just Nvidia virtue signaling while knocking over a single tree blocking only one of the many roads that lead to Rome.

[Wh0_Am_1]

On 2/19/2021 at 5:21 AM, Moonzy said:

Yeap, Nvidia cards didn't need firmware update, but now that they implement this, idk what to expect

 

I guess the re-use, reduce and recycle gets thrown out the window this time, if mining GPU can't be re-used after they're not profitable

 

I really can't see anyone except gamers benefiting from this, and even then, how much they benefit from is questionable

For non-mining gamers, they might be able to get their GPUs, depending on how fast the drivers and firmware is broken (I suspect within a week, judging by the game cracking community)

For everyone else, they lose the ability to mine and potentially gets nerfed in other aspects as well

As for the used GPU market, the mining cards will be e-waste when they're not profitable any longer so ggwp

And Nvidia will be in hot waters when supply is still insufficient and gamers no longer have miners to blame, as I've mentioned before

 

Edit: I would like to add that Nvidia can't back down now, the backlash would be too great

I'll be watching from the side lines with my buttered popcorn

Inb4 they just slowly forget about a this and move on

 

Feels like Nvidia is kicking themselves quite often recently.

Now imagine what would happen if this lead to currencies finally switching to using Tensor cores. Nvidia would probably be livid.

[Moonzy]

1 minute ago, Wh0_Am_1 said:

Nvidia would probably be livid.

on the outside, sure

pretty sure they're happy when checking their balance books, but cant show it to the world or the vocal audience will spoil their image

[Wh0_Am_1]

On 2/19/2021 at 7:29 AM, williamcll said:

Why not just develop a NPU based hasher so people would go by AI accelerators instead of graphic cards.

Cryptocurrencies would avoid the switch due to their goal of being decentralized currencies, Ethereum and others have already switched algorithms 2 years ago to prevent ASICs from getting in on the game, they would do so again unless the processer type was common place in mainstream systems.

[Wh0_Am_1]

1 minute ago, Moonzy said:

on the outside, sure

pretty sure they're happy when checking their balance books, but cant show it to the world or the vocal audience will spoil their image

Then again they may end up with the same issue that nearly killed Prestige Ameritech after H1N1, and cost Nvidia dearly after the last rush. Building up too much production, only for the market to suddenly swing to a point where it is over saturated with product still in crates and on warehouse shelves. Honestly the only thing I recommend Nvidia does next time is increase the production lead times before their next major launch with significant power efficiency and computational power improvements.

[Wh0_Am_1]

On 2/19/2021 at 6:09 AM, porina said:

There are many use cases where GPUs can be used without display output. I was looking forward to a flood of cheap GPUs after the last mining bubble but it never happened. We'll have to see if it happens this time.

It was there, in the form of the 1060 3GB, I have purchased several of these GPUs for only ~=$60USD before the rush.

 

[porina]

5 minutes ago, Wh0_Am_1 said:

It was there, in the form of the 1060 3GB, I have purchased several of these GPUs for only ~=$60USD before the rush.

When did you got those? From your writing of "before the rush" I assume last year? That pricing seems about ball park for general market value, and not what I'm talking about. The bubble in 2017-2018 that bought cryptocurrency to the mainstream before it went away is the nearest parallel we have to current bubble. I was hoping for more GPUs than you can shake a stick at, at low pricing. That was predicted if mining was unprofitable and miners sold off their cards. Yes, there were some small sales but nowhere near the flood. I think big miners are actually more about the long game so they just kept mining even if there were dips in pricing making it appear unprofitable. They were in essence storing for the near wave.

[Wh0_Am_1]

1 minute ago, porina said:

When did you got those? From your writing of "before the rush" I assume last year? That pricing seems about ball park for general market value, and not what 

Indeed.

2 minutes ago, porina said:

The bubble in 2017-2018 that bought cryptocurrency to the mainstream before it went away is the nearest parallel we have to current bubble. I was hoping for more GPUs than you can shake a stick at, at low pricing. That was predicted if mining was unprofitable and miners sold off their cards. Yes, there were some small sales but nowhere near the flood. I think big miners are actually more about the long game so they just kept mining even if there were dips in pricing making it appear unprofitable. They were in essence storing for the near wave.

Ok, I get your point. Yes, that did not happen. Rather instead we saw much of the large supply of no longer very profitable 1060 3GBs eased onto the market over an extended period of time greatly reducing their price. But nothing near what much of the industry was expecting. No Miners are in for the long haul as they have learned their lessons, and they will continue to be for the foreseeable future.

[Zodiark1593]

What I’d probably do instead is to alter the driver to reduce wear on cards in mining workloads. Specifically, an option (enabled by default) to scale back voltages, and maybe a bit on clockspeeds depending on temperature threshold. 
 

The result is that cards are in better condition for resale, and leas technically included miners may actually profit more from reduced power consumption. If they wish to go full speed, the option is there as well. 
 

Silicon production is finite, so enabling the cards to better sustain the workloads will provide a boon later down the road for gamers that but second hand. A price hike in the near term would probably be necessitated to account for the glut in the market during a mining crash, but is preferable to neutering the GPUs. 

[Moonzy]

So Linus pretty much said what I said...

 

Though he didn't mention that gamers can mine to recoup cost as well, which some other channels mentioned

[suicidalfranco]

1 hour ago, Moonzy said:

gamers can mine to recoup cost as well

things some people in this forum simply don't want to understand 

meanwhile 3 weeks in, i've already recoup a quarter of the price of my card.

Could have just let it collect dust because work and football took away all my time for gaming . 

[Moonzy]

12 minutes ago, suicidalfranco said:

things some people in this forum simply don't want to understand 

i understand why some people are against it, and i respect their decision to do whatever they wish with their life

 

the only valid reason i can agree with is the power consumption

and that's a very important point, so i can see why linus doesn't support it.

 

but if your local power is from green sources, or you simply dont care about global warming (you should), i don't see a reason why not to mine

[ewitte]

6 hours ago, suicidalfranco said:

things some people in this forum simply don't want to understand 

meanwhile 3 weeks in, i've already recoup a quarter of the price of my card.

Could have just let it collect dust because work and football took away all my time for gaming . 

I got my 3080 at like 780 and sold the 2080 for 550 so it only took 2-3 weeks for that to be a free card...  The 2080 was mostly funded from previous mining as well.  

[leadeater]

12 hours ago, Moonzy said:

So Linus pretty much said what I said...

It also sounded like he wasn't aware that there is in silicon and firmware measures too, sounded like he only thinks it's software/drivers. Though not sure how that changes much as it's really not clear how they are only identifying "Eth" mining and nothing else, sounds like that side of the detection can only be software/driver.

[leadeater]

3 hours ago, ThatOneDunce said:

Mining is only bad when I can't do it.

I can do it sort of, getting it back to real local currency or any currency I can actually spend is exceedingly difficult in this country. If it weren't for that well... I would have "a few" GPUs 

[Moonzy]

2 hours ago, leadeater said:

It also sounded like he wasn't aware that there is in silicon and firmware measures too, sounded like he only thinks it's software/drivers. Though not sure how that changes much as it's really not clear how they are only identifying "Eth" mining and nothing else, sounds like that side of the detection can only be software/driver.

From the email(?) that was shown in the video, Nvidia says that eth uses VRAM in a special way, so they limit the hash rate when they detect that

 

But again, firmware or not, people bios modded AMD cards before, to make it mine better, they will do it again

And there's also other algorithm

[leadeater]

44 minutes ago, Moonzy said:

From the email(?) that was shown in the video, Nvidia says that eth uses VRAM in a special way, so they limit the hash rate when they detect that

I doubt it's anything special and to that issue if Nvidia's statement is that it will only affect Eth then memory pattern monitoring will 100% result in something being improperly impacted. Of course if Nvidia actually told us how they were detecting it then it would make it easier to bypass, however Nvidia not telling means we have no way to verify their claim it will only affect Eth.

 

I don't really have a good reason to just trust Nvidia on this one.

[Moonzy]

1 minute ago, leadeater said:

I doubt it's anything special and to that issue if Nvidia's statement is that it will only affect Eth then memory pattern monitoring will 100% result in something being improperly impacted. Of course if Nvidia actually told how how they were detecting it then it would make it easier to bypass, however Nvidia not telling means we have no way to verify their claim it will only affect Eth.

 

I don't really have a good reason to just trust Nvidia on this one.

I was half expecting:

if(program==eth.exe)
   P_state_max=poopoo;

Just saying

[leadeater]

11 minutes ago, James Evens said:

Security by obscurity never works out if millions of dollar are involved.

It may not work but they still do it, generally they are just hoping to delay than anything else. The issue here is the unknown, because how are you actually going to verify if the workload is being incorrectly detected or not and then also why so. There's a hell of a lot of projects under BOINC with all sorts of workloads, make me rather doubtful this is going to all work out "perfectly".

 

Not that it changes the above quote, if Nvidia gives over those details it will be bypassed sooner than without that information. Hell I bet there would be a theoretical workaround ready before the cards actually reach people if so. 

[porina]

Random thoughts: "Solutions" don't have to be perfect. They just have to work well enough for long enough to have an impact. Too often some here seem to think if the solution isn't 100% then it isn't worth doing at all.

 

if there is something added in silicon to verify if firmware is signed before running, that would prevent trivial replacement with custom bios.

 

To my understanding ethereum and related mining is highly dependant on memory performance. We know it has a certain size footprint, and bandwidth/latency plays a big role. Typical basic optimisation is to increase mem clocks, reduce core clocks. I think but not 100% that bios mod is to implement vram optimisation beyond what is possible through driver.

 

If we take a step back, do miners want 3060s in particular, or only because it is just any GPU than can mine? I assume the latter. The nerf will be more work to get around regardless, and overall profitability will still be the optimisation if they look at others more different than ethereum. I don't keep up with all the coins, are there other GPU-friendly (so not ASIC level) ones that are substantially profitable? 

 

On possible false positive detections for other non-mining workloads, I think it is simple to detect. Implicitly it will be a bigger risk to those workloads that are memory heavy. Expected performance for the 3060 should be extrapolated from already released cards. If there is a significant delta from that, the mining nerf would be an obvious suspect. I think I saw somewhere that on a pre-release sample mining was observed to run at full rate for 30 secs before slowing. That pattern could be a good tell if that is how it actually works.