Nvidia Purposely Reducing Hash Rate for RTX 3060 GPUs, Creates Cryptocurrency Mining Processors (CMP)

Archaea

[H]F Junkie
Joined
Oct 19, 2004
Messages
11,159
There are a lot of big talkers in this thread who don’t actually know what they are talking about and have obviously never done any crypto mining themselves.

Nvidia firmware has not been hacked yet, not at least publicly. AMDs firmware editors abound and it’s a real thing to do BIOS modification to change clock and memory speeds and power target. I’ve done it myself on my former RX580 and made my own custom BIOS. Five versions of it actually which I shared with the community. Entire site ecosystems exist for this. (Well did, I see my favorite site for learning and sharing about it is now gone. Anorak.tech. )
There are now GUI based plug and play tools to modify AMD BIOS. Like this
https://www.igorslab.de/en/red-bios...en-more-stable-overclocking-navi-unlimited/2/
But nvidias BIOS have been successfully locked down since Pascal. The firmware has to be signed for Nvidia cards, and from what I can find it hasn’t been broken now for three generations of cards. There have been some workarounds made for pascal, but they have limited applicability. Like “Oh God a Company”’s ETH enlargement pill which reworked how the memory worked on a 1080, 1080ti, and Titan, but it requires the program be running in admin mode to affect hash rate and memory timings and it wasn’t directly applied to the card at a BIOS/firmware level. I’d assume more stuff like that exists in the private mining elite circles that isn’t shared or known about by the public — but if we could modify Nvidia firmware — that’d probably have been well known. So I’ll presume we still can’t.

Nvidia doesn’t sell ASICS.

Small computer shops aren’t universally mining and selling their cards used. My local computer shop owner, names Robby, I’ve talked to about mining on a few occasions and he asks me questions about it, even recently texted me asking for info, but has never attempted it. My single real world experience is no less valid than a invalid generalization that all are doing it with no valid source given.

Bitmain sells every antminer ASIC they make, they aren’t languishing around because no one wants them. In 2017/2018 timeframe Bitmain was temporarily valued at more than Nvidia. Saying nobody would buy purpose built mining equipment is nonsense. I’ve bought four antminers myself. Three of the four were profitable, though my siacoin miner Iost a few hundred bucks on because the token is a stinker. I’d buy purpose built mining cards as a hobby miner.

Nvidia easily sold every P10X mining card they previously made — so this mining cards won’t sell nonsense is .... well nonsense. There are even motherboards like the Asus B250 Mining Expert which I’m using right now that support 19 cards but six of them have to be dedicated mining cards from the Nvidia p10x line. (If you want to mine with all Nvidia cards) I don’t know how that requirement was arranged. If it was a smoke filled room handshake between asus and Nvidia - but it’s there and here three years later I can find no evidence on line that limitation has been removed by the community. So Nvidia or motherboard makers can enforce the use of mining cards with that type of practice. All in all, yes I would prefer typical GPUs, but in light of no available graphics cards options - like current, or better pricing on the mining versions as a good lure I’d personally buy the mining versions. (Price must be right though - about half as I outlined above would make me jump for Mining cards is reasonable - more than that and I’m not biting / others will though)

As to used up turds. I bought a dozen PNY XLR8 1070 cards from a miner, who has mined on them for about a six months to a year. I mined on them the next 2-3 years. Of those for an cards with 3-4 years of mining on them - they all passed the 3Dmark stability test for gamers, and worked perfectly fine with the exception of one fan on one card that had a squeak to it. You can buy new fans for less than $15 bucks on Amazon that are plugnplay - so no big loss there. I’ve purchased used mining cards before and I’d purchase them again. Just got to know and trust the seller — like almost anything else. Side note. I wouldn’t buy any Zotac refurb. I bought 15 1080ti direct from the Zotac refurb store, and 1-2 were dead right from the starting gate. 7-8 died within six months. Meanwhile I’d had a dozen + other 1080tis mining a year or two + without any issue whatsoever with the same settings. Meaning the Zotac cards were faulty to begin with, and not properly serviced before being sold (who knows what their original problem was that initiated their original return, but it wasn’t fixed. One card even arrived with janky fans. I no longer trust Zotac refurbs. I have mined with 60-80 cards since 2017. Lots of different makes and models. AMD and Nvidia. As far as failures I had one PowerColor AMD card fail and those refurb Zotac cards fail. Nothing else. Mining doesn’t kill cards, at least not if done correctly.
 
Last edited:

DukenukemX

Supreme [H]ardness
Joined
Jan 30, 2005
Messages
5,480
Just like the "unpickable locks" lock makers like to claim once in a while....

If there's a will, there's a way and somebody will figure out how to get around the limiter...
I've learned many things from watching the lockpickinglawyer, to the point where you don't even need to pick a lock to unlock it. Electronic locks can be opened with a strong magnet since they use magnetic solenoids to open. Most locks can be racked open in a matter of seconds. Some locks can be opened with a shim in a few seconds. Combination locks can usually be opened without knowing the code, as you can usually look into the mechanism to see the position that opens it.

But nvidias BIOS have been successfully locked down since Pascal. The firmware has to be signed for Nvidia cards, and from what I can find it hasn’t been broken now for three generations of cards.
My fake GTX 1050 Ti says otherwise. It's really a GT 450, which is not Pascal but the drivers that comes with the card have been properly hacked to fake a Pascal card. So much so that Windows 10 does genuinely believe it's a GTX 1050 Ti. Of course there's problems with this since it also believes it has 4GB of VRAM when it actually has 1GB, which will crash the card if it goes beyond the 1GB of VRAM. Also I gotta use these hacked drivers otherwise it won't work. BTW, I'm still looking for a working bios for my fake GT 450 so if anyone knows how to change the firmware on it that would be helpful.

If anybody hacks Nvidia's drivers you won't be downloading them off Guru3D. Good chance those drivers will be sold for a high price. Also there's Linux which anybody willing to contribute to Nouveau can make any changes they want. Too bad the Nouveau drivers suck ass but the code is there and open source, so anybody can modify it to crypto mine.
 

Archaea

[H]F Junkie
Joined
Oct 19, 2004
Messages
11,159
My fake GTX 1050 Ti says otherwise. It's really a GT 450, which is not Pascal but the drivers that comes with the card have been properly hacked to fake a Pascal card. So much so that Windows 10 does genuinely believe it's a GTX 1050 Ti. Of course there's problems with this since it also believes it has 4GB of VRAM when it actually has 1GB, which will crash the card if it goes beyond the 1GB of VRAM. Also I gotta use these hacked drivers otherwise it won't work. BTW, I'm still looking for a working bios for my fake GT 450 so if anyone knows how to change the firmware on it that would be helpful.
GT450 hardware is less < Pascal
The hardware is locked down since Pascal.

edit: DukenukemX

Here is a GT450 BIOS
https://www.techpowerup.com/vgabios/143074/143074

edit edit — ahh I see you’ve been down this path already.
 
Last edited:

sover

Limp Gawd
Joined
Jul 19, 2016
Messages
371
Just like the "unpickable locks" lock makers like to claim once in a while....

If there's a will, there's a way and somebody will figure out how to get around the limiter...
If they made their chips correctly (and there's no reason to think they didn't) then it's not a particularly hard nut to crack thanks to code signing. Just hardcode the chips to only use the boot values for all core and memory speeds (which would be very low) if the core part of the firmware they provide to AIBs isn't signed by nVidia, in the firmware only allow the speeds the card's mining mode has for all unsigned drivers, then make sure your code signing cert used for signing drivers doesn't get stolen. They could easily implement something like this going forward for all nVidia cards.
 

LeftSide

Gawd
Joined
Nov 14, 2003
Messages
540
https://mobile.twitter.com/bdelrizzo/status/1362619264423747590

Ryan at Anandtech got a few more details from Nvidia. Looks like the silicon, bios, and drivers are all signed, so it won’t be easy to just use a hacked driver or hacked bios.
Also, it looks like it won’t affect folding@home.
As long as you know what your getting into and it won’t mine as well as it games, I think this is a great move for gamers! It’s probably still going to be hard to find, but easier than the 3080 for sure.
 

sover

Limp Gawd
Joined
Jul 19, 2016
Messages
371
Shame these places lack such a mountain of money to overcome this 'only a difficulty'.......

For all intents and purposes you can put this issue to bed. It's quite likely it will take years for them to find a way around this without stealing the signing certificate. And even if they do find a way around it in only a few months the next version from nVidia will fix that vulnerability.

Really, if this solution even holds the miners off for 6 months then it did its job, miners might try buying a bunch of stock and holding it for a little while, but once it's clear this isn't something that's getting changed in the short term they'll just return them to the market.

The only surprising part of all of this is that nVidia didn't try to enforce this kind of market segmentation from the start. If there's money to be made nVidia is going to work hard to make it.
 

LeftSide

Gawd
Joined
Nov 14, 2003
Messages
540
Shame these places lack such a mountain of money to overcome this 'only a difficulty'.......
Maybe I should say it in a different way.
Without a solar winds style hack, where Nvidias signing keys are stolen or bypassed. This solution should hold up for a long time.
 

staknhalo

[H]ard|Gawd
Joined
Jun 11, 2007
Messages
1,480
I remain skeptical and think "THEY DID IT, THIS WILL SURELY WORK!" is overly optimistic

As said, where there's a will, there's a way

Money, and the more of it, just hastens that all

IMO mining firms are just doing the little Neo from the Matrix kung-fu 'bring it' gesture to this news.
 

LeftSide

Gawd
Joined
Nov 14, 2003
Messages
540
Sure, console gamers.

If you can't do what you want to do with it, it's not a PC, it's a console.
If you know about it before hand it’s fine.
Remember when nvidia released the 660 and said 512mb of the memory was slower than the rest, and everyone was fine with it. Then they released the 970, and we found out 1gig of the memory was slower from testing, and everyone lost their minds?
As long as they are upfront about what the card is capable of, I have no issue with it.
 

sover

Limp Gawd
Joined
Jul 19, 2016
Messages
371
I'm sure it will be right there on the box.
If they don't misrepresent the card's performance in crypto-currency hashing then what exactly will anyone be able to bring suit over? nVidia's marketing of these parts is exclusively in relation to their performance in games, not in crypto-hashing.
 

GoodBoy

[H]ard|Gawd
Joined
Nov 29, 2004
Messages
2,042
This whole thing is a PR ploy to get some of the miner associated heat off of them. Nothing less, nothing more.
I think it is because there is profit to be made.

if after 6 months of production there are some chips with failures that affect display output but not compute, it makes sense that if they can be made into another product and sold, to do so.

As far as the driver being hacked, sure it probably will. The casual miner is the only one really affected by the driver. The professional miners don't care either way, they don't need the driver. But hopefully they would be persuaded to buy the specific mining part vs a card with display outputs in any event.
The miners will preferentially buy the parts made for that purpose. They should have better availability as only miners would be buying them. Which means less miners buying gaming cards which means more of those should get used for actual gaming.

It's not going to help the supply issues all that much (supply will still be short), but will make some dent if they are using chips that had flaws rendering them unusable for gaming. in this way it generates more profit by using chips otherwise destined to the scrap heap, and increases the supply of gaming cards that gamers can actually buy. Again, supply will stay short, but more gamers who want to game and not mine should hopefully get cards than would otherwise be so. I know some gamers do both, and using mining profits to pay for the card upgrades. Most can still do this as the other models are not affected (as of yet) by driver changes. The part that does seem like "PR" is the driver announcement, but will have some small impact on which card some casual miners decide to buy. I doubt they would nerf mining across all cards, there would be both gamers and casual miners ticked about that.

For everyone throwing a fit and professing to buy only AMD, great! It makes no sense to me, but I am still trying to get a 3090 so the less of you I have to complete with to find one the better. I mean what's with the rage? Miners buying cards isn't something nvidia can control. About all they can do is redirect chips to that market and hopefully stave off hardcore miners buying as many gaming cards. Purposely nerf hash rate? Not sure how, or even if that is really a good idea. Plenty of gamers are also casual miners to help pay for their pc upgrades. What you do with a piece of hardware after you buy it should be your business and your business alone.
 

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
52,020
I think it is because there is profit to be made.

if after 6 months of production there are some chips with failures that affect display output but not compute, it makes sense that if they can be made into another product and sold, to do so.

As far as the driver being hacked, sure it probably will. The casual miner is the only one really affected by the driver. The professional miners don't care either way, they don't need the driver. But hopefully they would be persuaded to buy the specific mining part vs a card with display outputs in any event.
The miners will preferentially buy the parts made for that purpose. They should have better availability as only miners would be buying them. Which means less miners buying gaming cards which means more of those should get used for actual gaming.

It's not going to help the supply issues all that much (supply will still be short), but will make some dent if they are using chips that had flaws rendering them unusable for gaming. in this way it generates more profit by using chips otherwise destined to the scrap heap, and increases the supply of gaming cards that gamers can actually buy. Again, supply will stay short, but more gamers who want to game and not mine should hopefully get cards than would otherwise be so. I know some gamers do both, and using mining profits to pay for the card upgrades. Most can still do this as the other models are not affected (as of yet) by driver changes. The part that does seem like "PR" is the driver announcement, but will have some small impact on which card some casual miners decide to buy. I doubt they would nerf mining across all cards, there would be both gamers and casual miners ticked about that.

For everyone throwing a fit and professing to buy only AMD, great! It makes no sense to me, but I am still trying to get a 3090 so the less of you I have to complete with to find one the better. I mean what's with the rage? Miners buying cards isn't something nvidia can control. About all they can do is redirect chips to that market and hopefully stave off hardcore miners buying as many gaming cards. Purposely nerf hash rate? Not sure how, or even if that is really a good idea. Plenty of gamers are also casual miners to help pay for their pc upgrades. What you do with a piece of hardware after you buy it should be your business and your business alone.
Do you think that NV is just now selling directly to miners? This has been going on literally for years.
 

staknhalo

[H]ard|Gawd
Joined
Jun 11, 2007
Messages
1,480
Do you think that NV is just now selling directly to miners? This has been going on literally for years.

No I think he's just saying Nvidia was acquiring more and more defective chips through all this and realized the could make a mining card with them, along with all this PR hubub about pRoTeCtInG gAmErZ (while still selling directly to miners all the while, cause money talks) - and make a few extra dollars with some nice current-event-topical marketing. Which I find plausible, at the very least.
 

BassTek

Supreme [H]ardness
Joined
Jul 13, 2002
Messages
5,880
I wonder what happens if you run two applications at the same time with different workloads, although I'm sure they have though of that scenario already.
 

kju1

2[H]4U
Joined
Mar 27, 2002
Messages
3,453
No I think he's just saying Nvidia was acquiring more and more defective chips through all this and realized the could make a mining card with them, along with all this PR hubub about pRoTeCtInG gAmErZ (while still selling directly to miners all the while, cause money talks) - and make a few extra dollars with some nice current-event-topical marketing. Which I find plausible, at the very least.

Yeah "defective" thats it. But still good enough to sell to miners because its more profitable. I dont believe for an instant that yields are that bad.
 

staknhalo

[H]ard|Gawd
Joined
Jun 11, 2007
Messages
1,480
Yeah "defective" thats it. But still good enough to sell to miners because its more profitable. I dont believe for an instant that yields are that bad.

Wouldn't shock me either. Be NVIDIA, see for whatever reasons on your end - whether the reasons are or aren't all within your control - higher income and profit stream from miners ATM than gamers, just start 'repackaging' your public-on-the-shelf formerly-gaming products as miner products.
 

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
52,020
Yeah "defective" thats it. But still good enough to sell to miners because its more profitable. I dont believe for an instant that yields are that bad.
I do not either. No where close. This is another NVIDIA PR ploy.
 
  • Like
Reactions: kju1
like this

Sycraft

Supreme [H]ardness
Joined
Nov 9, 2006
Messages
4,931
But nvidias BIOS have been successfully locked down since Pascal. The firmware has to be signed for Nvidia cards, and from what I can find it hasn’t been broken now for three generations of cards.
What many people don't know or appreciate is how public key cryptography can be used to make stuff like this pretty well impossible to break. A company can enforce signing and little can be done about it. So a somewhat simplified explanation for people:

Public key cryptography aka asymmetric cryptography is one where you have two keys: A private key and a public key derived from that private key. Anything done with one can only be undone with the other. So if something is encrypted with a public key, it can only be decrypted with the private key, not with the public key, and vice versa. Also while the public key is easy to mathematically derive from the private key, the private key is infeasible to the point of impossibility to derive from the public key. What this means is you have two keys: One that you hold on to, and one that you can distribute to the world. Lots of stuff uses this, like SSL/TLS.

Ok so how would we use this to make firmware you can't mess with? Well what we do is we take the public key and actually bake it in to the hardware of the chip. Make it so that it is physically wired in there, no way to change or remove it. Then all firmware is either signed with the private key, or even completely encrypted with it. The chips are then designed so that they will only run the firmware if they can verify the signature or do the decryption using their public key.

What this then means is that even though a person online can get the firmware to reverse engineer or change, they can't sign/encrypt it as needed. They only have the public key and they need the private key. Only way to get that is either to break in and steal it, and it is presumably well protected, or to invent a new field of mathematics that has somehow eluded all the smartest mathematicians and cryptographers in the world. Not happening.

So realistically the only thing that could happen is if there was a flaw in their implementation. If somehow they implemented the signature check on the chip weakly or wrong and you could modify the firmware to bypass the check. But if it is done well, you are screwed. You'd have to get the private key to sign it, and that you realistically can't do.

Cryptography is not like physical locks, where time and knowledge can let you pick any tumbler. There are so many combinations for keys, like more than there are atoms in the universe, that it is effectively "unpickable".
 

staknhalo

[H]ard|Gawd
Joined
Jun 11, 2007
Messages
1,480
What many people don't know or appreciate is how public key cryptography can be used to make stuff like this pretty well impossible to break. A company can enforce signing and little can be done about it. So a somewhat simplified explanation for people:

Public key cryptography aka asymmetric cryptography is one where you have two keys: A private key and a public key derived from that private key. Anything done with one can only be undone with the other. So if something is encrypted with a public key, it can only be decrypted with the private key, not with the public key, and vice versa. Also while the public key is easy to mathematically derive from the private key, the private key is infeasible to the point of impossibility to derive from the public key. What this means is you have two keys: One that you hold on to, and one that you can distribute to the world. Lots of stuff uses this, like SSL/TLS.

Ok so how would we use this to make firmware you can't mess with? Well what we do is we take the public key and actually bake it in to the hardware of the chip. Make it so that it is physically wired in there, no way to change or remove it. Then all firmware is either signed with the private key, or even completely encrypted with it. The chips are then designed so that they will only run the firmware if they can verify the signature or do the decryption using their public key.

What this then means is that even though a person online can get the firmware to reverse engineer or change, they can't sign/encrypt it as needed. They only have the public key and they need the private key. Only way to get that is either to break in and steal it, and it is presumably well protected, or to invent a new field of mathematics that has somehow eluded all the smartest mathematicians and cryptographers in the world. Not happening.

So realistically the only thing that could happen is if there was a flaw in their implementation. If somehow they implemented the signature check on the chip weakly or wrong and you could modify the firmware to bypass the check. But if it is done well, you are screwed. You'd have to get the private key to sign it, and that you realistically can't do.

Cryptography is not like physical locks, where time and knowledge can let you pick any tumbler. There are so many combinations for keys, like more than there are atoms in the universe, that it is effectively "unpickable".

Good thing governments the size of the US, or China (hosting and bankrolling a lot of mining firms), are not known to stockpile unknown exploits for themselves and never do things like break into companies (either digitally with remote access or delivered pre-compromised HW - or even with actual personnel/spies) to acquire the data they want, regardless of laws. I mean God forbid governments, companies don't ever do that stuff either.
 
Last edited:

Bassman99

Gawd
Joined
Feb 3, 2021
Messages
672
There are a lot of big talkers in this thread who don’t actually know what they are talking about and have obviously never done any crypto mining themselves.

Nvidia firmware has not been hacked yet, not at least publicly. AMDs firmware editors abound and it’s a real thing to do BIOS modification to change clock and memory speeds and power target. I’ve done it myself on my former RX580 and made my own custom BIOS. Five versions of it actually which I shared with the community. Entire site ecosystems exist for this. (Well did, I see my favorite site for learning and sharing about it is now gone. Anorak.tech. )
There are now GUI based plug and play tools to modify AMD BIOS. Like this
https://www.igorslab.de/en/red-bios...en-more-stable-overclocking-navi-unlimited/2/
But nvidias BIOS have been successfully locked down since Pascal. The firmware has to be signed for Nvidia cards, and from what I can find it hasn’t been broken now for three generations of cards. There have been some workarounds made for pascal, but they have limited applicability. Like “Oh God a Company”’s ETH enlargement pill which reworked how the memory worked on a 1080, 1080ti, and Titan, but it requires the program be running in admin mode to affect hash rate and memory timings and it wasn’t directly applied to the card at a BIOS/firmware level. I’d assume more stuff like that exists in the private mining elite circles that isn’t shared or known about by the public — but if we could modify Nvidia firmware — that’d probably have been well known. So I’ll presume we still can’t.

Nvidia doesn’t sell ASICS.

Small computer shops aren’t universally mining and selling their cards used. My local computer shop owner, names Robby, I’ve talked to about mining on a few occasions and he asks me questions about it, even recently texted me asking for info, but has never attempted it. My single real world experience is no less valid than a invalid generalization that all are doing it with no valid source given.

Bitmain sells every antminer ASIC they make, they aren’t languishing around because no one wants them. In 2017/2018 timeframe Bitmain was temporarily valued at more than Nvidia. Saying nobody would buy purpose built mining equipment is nonsense. I’ve bought four antminers myself. Three of the four were profitable, though my siacoin miner Iost a few hundred bucks on because the token is a stinker. I’d buy purpose built mining cards as a hobby miner.

Nvidia easily sold every P10X mining card they previously made — so this mining cards won’t sell nonsense is .... well nonsense. There are even motherboards like the Asus B250 Mining Expert which I’m using right now that support 19 cards but six of them have to be dedicated mining cards from the Nvidia p10x line. (If you want to mine with all Nvidia cards) I don’t know how that requirement was arranged. If it was a smoke filled room handshake between asus and Nvidia - but it’s there and here three years later I can find no evidence on line that limitation has been removed by the community. So Nvidia or motherboard makers can enforce the use of mining cards with that type of practice. All in all, yes I would prefer typical GPUs, but in light of no available graphics cards options - like current, or better pricing on the mining versions as a good lure I’d personally buy the mining versions. (Price must be right though - about half as I outlined above would make me jump for Mining cards is reasonable - more than that and I’m not biting / others will though)

As to used up turds. I bought a dozen PNY XLR8 1070 cards from a miner, who has mined on them for about a six months to a year. I mined on them the next 2-3 years. Of those for an cards with 3-4 years of mining on them - they all passed the 3Dmark stability test for gamers, and worked perfectly fine with the exception of one fan on one card that had a squeak to it. You can buy new fans for less than $15 bucks on Amazon that are plugnplay - so no big loss there. I’ve purchased used mining cards before and I’d purchase them again. Just got to know and trust the seller — like almost anything else. Side note. I wouldn’t buy any Zotac refurb. I bought 15 1080ti direct from the Zotac refurb store, and 1-2 were dead right from the starting gate. 7-8 died within six months. Meanwhile I’d had a dozen + other 1080tis mining a year or two + without any issue whatsoever with the same settings. Meaning the Zotac cards were faulty to begin with, and not properly serviced before being sold (who knows what their original problem was that initiated their original return, but it wasn’t fixed. One card even arrived with janky fans. I no longer trust Zotac refurbs. I have mined with 60-80 cards since 2017. Lots of different makes and models. AMD and Nvidia. As far as failures I had one PowerColor AMD card fail and those refurb Zotac cards fail. Nothing else. Mining doesn’t kill cards, at least not if done correctly.
Brief reply to a long post.
Half price seems right to me since its got only half the usage potential.
 

Flogger23m

[H]F Junkie
Joined
Jun 19, 2009
Messages
11,510
There are likely a number of factors at play here (and tbh we dont know that they dont do this). First off they would be eating into GPU, err I mean CMP, sales which is likely more profitable for them than mining is. And then there is the fact that mining is volatile and regulations around it are still firming up.

IMO it doesn't make sense - eventually bitcoin will be fully mined out. No the far better play is to sell the equipment miners need and let them be the ones holding the bag when all the coins are mined.

And I think if they essentially had a monopoly on it regulations will come much quicker.

Like the gold rush, most of the people that ended up rich in the long term were the suppliers selling equipment and lodging to the miners.
 

kju1

2[H]4U
Joined
Mar 27, 2002
Messages
3,453
ROFL...this was amusing to me.
 

Attachments

  • Screen Shot 2021-02-19 at 3.55.03 PM.png
    Screen Shot 2021-02-19 at 3.55.03 PM.png
    499.2 KB · Views: 0

Sycraft

Supreme [H]ardness
Joined
Nov 9, 2006
Messages
4,931
Good thing governments the size of the US, or China (hosting and bankrolling a lot of mining firms), are not known to stockpile unknown exploits for themselves and never do things like break into companies (either digitally with remote access or delivered pre-compromised HW - or even with actual personnel/spies) to acquire the data they want, regardless of laws. I mean God forbid governments, companies don't ever do that stuff either.
If governments want to mine crypto, they have big computers to do it on. They also can just contract with nVidia directly. This has no bearing on anything.
 

staknhalo

[H]ard|Gawd
Joined
Jun 11, 2007
Messages
1,480
If governments want to mine crypto, they have big computers to do it on. They also can just contract with nVidia directly. This has no bearing on anything.

Point being there is no thing that is full proof secure.

The only determining factor in how secure something is, is how enticing of a target it is.

There are ways and methods around anything and everything.
 

Archaea

[H]F Junkie
Joined
Oct 19, 2004
Messages
11,159
Point being there is no thing that is full proof secure.

The only determining factor in how secure something is, is how enticing of a target it is.

There are ways and methods around anything and everything.
nonsense

AES256 encryption - how do you break it? The way it's encrypted is plainly documented and clear to anyone to view, study and digest. But the complexities of reversing it are impossible with modern technology.
There are plenty of ciphers that use multiple salts and multiple rounds of encryption to make them uncrackable anytime into the foreseeable future.

I get a kick out of people saying the government has little secret cracks to these things and nothing is safe. Have you ever worked for the government - you are giving them too much credit. My former job of nine years (before I moved to application support) was information security and forensics in a company of 25,000 people that is quasi government. I know pretty well the capabilities against good strong security measures. They aren't as robust as people assume. They rely on voltage hacks, and brute force against image clones and snapshots, instead of black box voodoo math that isn't supposed to exist. You guys have watched too much Swordfish.
 

Sycraft

Supreme [H]ardness
Joined
Nov 9, 2006
Messages
4,931
Point being there is no thing that is full proof secure.
No, but things can be pretty effectively secure. Like AES. Do you know how to break AES? In the 23 years it has been out there, as the most studied and attacked cryptosystem in history, has anyone ever broken it (hint: no)? When used properly, good crypto can be really secure, to the point that we don't see breaks. So thinking that nVidia's firmware signing would be easy to bypass is wishful thinking. There may be a way, there is no way to prove there isn't. However that doesn't mean that there is and that people can just do it if they want to. It is not like a lock, where even a properly designed one can be defeated with just time and effort. It is one where there must be a flaw.

Also turns out companies are getting better at this stuff, they learn from past mistakes. For example: The Xbox One is 8 years old. Go find me a mod kit for it... You can't. The boot security has yet to be broken on it. Does that mean it never will be? No, but it is clearly pretty hard, it lasted the whole useful design life of the platform.

The point is that right now, nVidia firmware modding isn't happening. They use cryptography to secure it, and it is so far effective. It isn't something like using crypto to try and keep you from copying a movie, that is where they don't want you to have the data, but still have to give you the data. No way to design that secure, it is fundamentally something cryptography doesn't do, you can't give someone a decryption key and yet not have them able to decrypt. This though, is something crypto is perfectly capable of doing, and the kind of thing we have a lot of experience with. Secure signing and verification is done a lot, in a lot of systems, and so we've found out a lot about how to avoid implementation problems. Thus not too hard to think that they might have an implementation that isn't going to be broken. Particularly if the only benefit is being able to mine crypto currency.
 

ChadD

Supreme [H]ardness
Joined
Feb 8, 2016
Messages
5,150
I think it will be more likely that they figure out a way to fool the workload detection rather than disable it via hacked firmware. It will be interesting to see how this plays out.
Ya more then one way to attack this problem. They don't want to cripple all compute just specific compute ? Ya someone will figure a way to get around this... IF its profitable to do so anyway. This still feels like a PR stunt more then anything... had they talked about this before 3080 and 70s where launched I would perhaps believe they where serious. I know stock wasn't a massive issue at that point... but still I come back to the real solution, enough stock for everyone that wants one. At that point Nvidia could even sell mining cards potentially as resell values would drop significantly enough to perhaps change the practice of mining for 6 months and being able to sell the cards for so close to their original pricing.
 

Bassman99

Gawd
Joined
Feb 3, 2021
Messages
672
No, but things can be pretty effectively secure. Like AES. Do you know how to break AES? In the 23 years it has been out there, as the most studied and attacked cryptosystem in history, has anyone ever broken it (hint: no)? When used properly, good crypto can be really secure, to the point that we don't see breaks. So thinking that nVidia's firmware signing would be easy to bypass is wishful thinking. There may be a way, there is no way to prove there isn't. However that doesn't mean that there is and that people can just do it if they want to. It is not like a lock, where even a properly designed one can be defeated with just time and effort. It is one where there must be a flaw.

Also turns out companies are getting better at this stuff, they learn from past mistakes. For example: The Xbox One is 8 years old. Go find me a mod kit for it... You can't. The boot security has yet to be broken on it. Does that mean it never will be? No, but it is clearly pretty hard, it lasted the whole useful design life of the platform.

The point is that right now, nVidia firmware modding isn't happening. They use cryptography to secure it, and it is so far effective. It isn't something like using crypto to try and keep you from copying a movie, that is where they don't want you to have the data, but still have to give you the data. No way to design that secure, it is fundamentally something cryptography doesn't do, you can't give someone a decryption key and yet not have them able to decrypt. This though, is something crypto is perfectly capable of doing, and the kind of thing we have a lot of experience with. Secure signing and verification is done a lot, in a lot of systems, and so we've found out a lot about how to avoid implementation problems. Thus not too hard to think that they might have an implementation that isn't going to be broken. Particularly if the only benefit is being able to mine crypto currency.
I know little about bios crypto and security protocols in general.
And might not be much incentive to do it, and might not even be possible.
But there are people out there that hack for fun like the lock picking lawyer.
Its like a crossword puzzle or logic problem for the OCD guy with time on his hands.
 

DanNeely

2[H]4U
Joined
Aug 26, 2005
Messages
3,863
Ya more then one way to attack this problem. They don't want to cripple all compute just specific compute ? Ya someone will figure a way to get around this... IF its profitable to do so anyway. This still feels like a PR stunt more then anything... had they talked about this before 3080 and 70s where launched I would perhaps believe they where serious. I know stock wasn't a massive issue at that point... but still I come back to the real solution, enough stock for everyone that wants one. At that point Nvidia could even sell mining cards potentially as resell values would drop significantly enough to perhaps change the practice of mining for 6 months and being able to sell the cards for so close to their original pricing.

The workload is defined by the mining algorithm itself. The only way they could make it look less like etherium mining is to insert junk operations into the middle of the computations slowing the mining down. And if I was designing the implementation for NVIDIA I'd do >50% of operations are ether mining means apply the 50% slowdown nerf; meaning any such spoofing sufficient to avoid detection would also result in a >50% slowdown.

Trying to bury the mining operations in junk code's unlikely to work unless future generations apply far more aggressive nerfing on consumer cards. If they eventually try to nerf it like they did FP64 on consumer cards we might have reason to worry.
 

cybereality

Supreme [H]ardness
Joined
Mar 22, 2008
Messages
7,036
No, the hackers won't crack AES or the private key, but there may be a way to bypass detection.

It is unknown how Nvidia is doing this, but if the detection is based on software running on a local machine, then it is up for grabs.

As a basic example, a hacker could rename "nicehash.exe" to "crysis.exe" and make it look like a game (I know Nvidia is not this dumb, but a basic example).

Or if the detection is driver based, you could hook into the vtable of the driver and redirect the "check_eth_hash()" function and have it always return "false". That would be fairly easy and cracked within a day.

But if it's baked into the firmware, then it would be much harder. Maybe not impossible (nothing is impossible or unhackable) but it could be infeasible to crack within the lifetime of the product, which would achieve the goal.

So yeah, if it takes hackers 6 or 12 months to hack, that would be a success because it would stop the miners from using it for the lifetime of the launch (in a year, new cards will be out and we can start all over again).
 

ChadD

Supreme [H]ardness
Joined
Feb 8, 2016
Messages
5,150
The workload is defined by the mining algorithm itself. The only way they could make it look less like etherium mining is to insert junk operations into the middle of the computations slowing the mining down. And if I was designing the implementation for NVIDIA I'd do >50% of operations are ether mining means apply the 50% slowdown nerf; meaning any such spoofing sufficient to avoid detection would also result in a >50% slowdown.

Trying to bury the mining operations in junk code's unlikely to work unless future generations apply far more aggressive nerfing on consumer cards. If they eventually try to nerf it like they did FP64 on consumer cards we might have reason to worry.
I can't imagine they are really all that serious about this move in general. But ya Nvidia seems pretty committed to selling compute cards... and also too cheap to design 2 different chips as AMD plans too. (which to be fair hasn't been proven to be a great strategy quite yet either.) Nvidia has always shot their higher end cards in the foot a little by using basically the exact same hunk of silicon in both lines. They have always cannibalized their own pro type card sales with their gaming cards. To be extremely cynical perhaps this is all just a test so they can figure out how best to handicap their gaming cards to sell quadros down the road. Perhaps when the 4000s hit... they really will blow for compute completely. (but as you say this is a targeted handicap supposedly)

Ya we'll see how this all goes... my gut tells me this is basically the last we really hear of the Nvidia mining card push. They will make a token number and then go back to doing business as normal. I know I am pretty cynical on the entire situation. Hey if people are able to buy 3060s that is great. I'm sure there are lots of people ready to buy pretty much anything they can get... although imo heck unless your old 1080s or whatever have died. What games are really all that demanding right now anyway. Cyberpunk seems to have turned into a disappointment... not many people still playing it. If you really are forced to make your 2 year old card go another 6 months at this point its probably not the end of the world we may have thought while the CP hype train was still chugging. :)
 

warmon6

n00b
Joined
Nov 5, 2011
Messages
46
I've learned many things from watching the lockpickinglawyer, to the point where you don't even need to pick a lock to unlock it. Electronic locks can be opened with a strong magnet since they use magnetic solenoids to open. Most locks can be racked open in a matter of seconds. Some locks can be opened with a shim in a few seconds. Combination locks can usually be opened without knowing the code, as you can usually look into the mechanism to see the position that opens it.

Funny enough... i was going to use his favorite phrase "click on one, nothing on two, three is binding" then thought against it... lol

No, the hackers won't crack AES or the private key, but there may be a way to bypass detection.

It is unknown how Nvidia is doing this, but if the detection is based on software running on a local machine, then it is up for grabs.

As a basic example, a hacker could rename "nicehash.exe" to "crysis.exe" and make it look like a game (I know Nvidia is not this dumb, but a basic example).

Or if the detection is driver based, you could hook into the vtable of the driver and redirect the "check_eth_hash()" function and have it always return "false". That would be fairly easy and cracked within a day.

But if it's baked into the firmware, then it would be much harder. Maybe not impossible (nothing is impossible or unhackable) but it could be infeasible to crack within the lifetime of the product, which would achieve the goal.

So yeah, if it takes hackers 6 or 12 months to hack, that would be a success because it would stop the miners from using it for the lifetime of the launch (in a year, new cards will be out and we can start all over again).

Just like what happen with the Nintendo 64. Many years, there were ways to bypass the PIF-CIC chip for bootleg games but the chip itself was not cracked till 2015 (might be off by a year or two).

almost 20 years is certainly well past the life time off the product (as far as Nintendo is concerned)

 
Top