|
Paul MaudDib posted:TUF 3080 non-OC dropped this afternoon on Amazon... at $1199 Already gone too, unless I just couldn't find the correct listing.
|
#
?
Feb 21, 2021 00:15
|
|
|
|
| # ? Apr 28, 2024 22:37 |
|
|
repiv posted:intentionally tanking the performance of certain workloads for segmentation is nothing new, AMD and NV alike have been doing it forever Doesn't AMD have open source drivers though? It really sucks if the only two companies making GPUs have already been doing this. This is coming from someone who doesn't mine, btw. I don't care if miners are screwed over, and it's a good thing for everyone else (including me), but I've just seen enough closed source BS to know that doing such things is bad for consumers in the long run.
|
|
#
?
Feb 21, 2021 01:33
|
|
|
The Gadfly posted:Doesn't AMD have open source drivers though? They still have firmware signing, if they did what NVIDIA is rumored to be doing with the 3060 Ti/3070/3080/3090 and re-released their cards as 6850XT with a firmware mining brake installed (such that it broke compatibility with the older, non-limited VBIOS) then there's nothing that the open-source drivers could really do about it. Just like the open-source drivers can't really fix AMD's firmware gimping the FP64 performance on 290X. the effectiveness of this approach is really going to turn based on how good their heuristic is. If it's relatively trivial to rewrite mining kernels so that the heuristic doesn't pick them up, it's not going to do much. Maybe they are going to be expansive and poo poo on anything that has high utilization of the memory bus, but then have each specific application signed (or recognized by the driver) so that they manually remove the brake. Not sure how you validate that from the GPU itself since you can't depend on the drivers themselves being non-altered, but I guess there's ways to do that. Root of trust/trusted computing is a known problem. Intel SGX is broken though, and it's hard to trust any client-side secrets in this era of specter/meltdown. Paul MaudDib fucked around with this message at 01:50 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 01:41
|
|
|
Would you consider "you don't have DLSS nor ray-tracing unless you buy an RTX card" an example of deliberate segmentation? or alternatively, cards like the GT 710 or GT 1030 do not have NVENC - you need a GTX card for that, and you still get outliers like the GTX 1650 using the Volta/Pascal encoder while all the other cards (including the GTX 1650 Super) gets the Turing encoder EDIT: to be clear, my point is perhaps more that "locking of features has been going on for a while", but I don't intend that to mean this is necessarily good or consumer-friendly. gradenko_2000 fucked around with this message at 01:49 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 01:44
|
|
|
I don't know enough about the low-level hardware involved, but would locking out mining have any similarities to HDCP and only allowing certain content to display if certain variables are met? I know there are more distinct units of hardware interacting with HDCP, but it was a thought that occurred to me.
|
|
#
?
Feb 21, 2021 01:47
|
|
|
it's psychological but people do get unhappy about soft-locking of features - people on the ML side have been bitching about the 3090 being handicapped
|
|
#
?
Feb 21, 2021 01:47
|
|
|
gradenko_2000 posted:Would you consider "you don't have DLSS nor ray-tracing unless you buy an RTX card" an example of deliberate segmentation? Turing Minor physically did not have RTX cores though, and GT cards physically do not have NVENC blocks. The analogous thing is more like the FP64 limiting that took place on AMD Hawaii or NVIDIA GK110 chips, which was purely a firmware thing (or sometimes was determined by resistors physically placed on the card). Or the way gaming cards get slowed down by the driver when they detect you using some OpenGL calls that are specific to CAD applications. Doesn't really change the overall point though, segmentation has been taking place for a long time. People are indeed unusually sensitive about it when it's purely software-based though. Like people reacted extremely badly to the idea of being able to unlock hyperthreading after purchase, they viewed that as "processor DLC". Seemingly it is preferable to have that market segmentation burned into the hardware and to have to sell your hardware and buy a new one if you want to upgrade, people dislike the feeling of being nickeled-and-dimed so strongly that they reject even having the option presented to them, it just rubs their face into it too much even if it's purely advantageous for them to have the option. It's not like you're paying up front for the option, or not paying much, but people would rather literally not have it at all because it just drives home "the hardware could do this but you won't let it". (obviously it would be better if Intel would sell you an i7 at pentium prices, or AMD would sell you a 3800X at 3300X prices, but market segmentation is just a reality, there aren't enough "broken" processors to fill the higher levels of demand for low-end products, if there was no market segmentation then there would be no lower-end SKUs, or they would be much scarcer.) Paul MaudDib fucked around with this message at 02:04 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 01:51
|
|
|
You guys are implying that it's irrational to dislike softlocking. It's not. You're buying a hardware product physically capable of x, y, and z. If the only reason that product doesn't do z is because you haven't paid additional money, then it's anti-consumer. No one's going to be upset if the product they bought is just not physically capable of something, like if a fan with a certain motor isn't able to spin at a certain rpm due to physics. However, it is fine to be upset if the only reason the fan is spinning slower is because there a counterweight attached to the fan that you need to buy a code to unlock.
|
|
#
?
Feb 21, 2021 03:55
|
|
|
right now you pay say, $500 for a gaming card while pros spend $3000 for the same silicon but without the artificial caps if they stop doing artificial segmentation then they'd need to charge everyone something like $600-700 to get the same revenue overall are you willing to pay extra on principle for features you'll never use?
|
|
#
?
Feb 21, 2021 04:09
|
|
|
repiv posted:right now you pay say, $500 for a gaming card while pros spend $3000 for the same silicon but without the artificial caps it would probably end up being a lot more than that, because enterprise uses a lot more hardware than consumers. Amazon plus every other hyperscaler plus everyone with their own datacenter means there are literally multiple other computers (on average) just devoted to serving your local PC. Every professional workstation and office PC, etc. Business uses a shitload of hardware overall. so the larger effect is that if you don't do market segmentation, you are really cutting prices for your largest and most profitable customers, it is going to take a large increase to the "average" price to make up for that. It is more like "are you OK with paying $1000-1500 for that $500 gaming card just on principle". and obviously low-end CPUs and such are segmented by features too, those would have to go up significantly as well to account for ECC support, Vpro support, AVX support, etc that is used to funnel business users towards more expensive platforms. It is fine to be mad when specific market segmentation stuff is obnoxious and overly restrictive (segmenting ECC is pretty obnoxious). But as a whole market segmentation benefits consumers greatly, we are getting cheap hardware and business is forced to pay a lot more, that subsidizes the R&D for our cheap hardware and lets us get it a lot closer to production cost. Paul MaudDib fucked around with this message at 04:26 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 04:21
|
|
|
the reality of how silicon production works means artificial segmentation is far more efficient than making physically different products for each sector too pro nvidia cards can encode 10 videos simultaneously while consumer ones are artificially capped to 3. could nvidia make a distinct die for consumers with a smaller nvenc block that's only physically capable of encoding 3 videos, rather than cutting down the big 10 video block? sure, but the cost of taping out an entirely separate die means you'd pay $X extra for a product that behaves exactly the same.
|
|
#
?
Feb 21, 2021 04:30
|
|
|
If this makes it easier for me to get a new card in a month or two, it's good. If it doesn't, it's bad. I don't really give a poo poo about the effects of segmentation other than that.
|
|
#
?
Feb 21, 2021 04:33
|
|
|
you can't legally use geforce cards in data centers so there's no soft (or hard) lock required nvlink and the physical size, fan config of the cards means you can't stack the geforce cards either for entry-level enterprise/prosumer level use "think of the company's profit margins" is a pretty dumb take
|
|
#
?
Feb 21, 2021 04:44
|
|
|
How much above retail do you think I'd need to pay a goon to send a 3090 FE to Australia? I don't know what 3090 FE availability is like right now
|
|
#
?
Feb 21, 2021 04:45
|
|
|
The main factor in bringing costs down for the consumer is competition. Just because a company makes more profit from businesses through softlocks, doesn't mean that they pass the profits on to the typical retail consumer out of the goodness of their hearts. They just pocket the excess profit for themselves, unless their competitors force them to lower prices. The reason that prices are the way they are, is that the business did a market analysis on what sale price would optimize their profit.
|
|
#
?
Feb 21, 2021 05:35
|
|
|
Yeah if there's an example of that it's the GPU market, where rational market analysis has resulted in no third parties being unable to capture market surplus through pricing, and the CEO of the largest company isn't known for setting prices minutes before going on stage at the annoucement keynote
|
|
#
?
Feb 21, 2021 05:48
|
|
|
BurritoJustice posted:How much above retail do you think I'd need to pay a goon to send a 3090 FE to Australia? I don't know what 3090 FE availability is like right now The biggest impediment is that BB releases the cards in groups and also only unlocks 'x' number of cards per geographic region. You can get the card into your cart only to find out there aren't any stores within "250 miles" that'll let you pick it up.
|
|
#
?
Feb 21, 2021 05:48
|
|
|
Doctor Nutt posted:Already gone too, unless I just couldn't find the correct listing. Every card is sold out within literal seconds. By the time someone makes the post here, you refresh the page and read it, and go to Amazon, they're gone. Just give up like I did and embrace your 1060 for another six months or a year or forever
|
|
#
?
Feb 21, 2021 06:19
|
|
|
Gonna fit a Morpheus cooler to this 2060, least I won't be able to hear it.
|
|
#
?
Feb 21, 2021 06:26
|
|
|
FuturePastNow posted:If this makes it easier for me to get a new card in a month or two, it's good. If it doesn't, it's bad. I don't really give a poo poo about the effects of segmentation other than that. E: I hate that guy's mug, so here's a link instead. Linus's take: 1) This might gently caress over new card buyers if Nvidia prioritizes miner GPU production over gamer GPU production. 2) This might gently caress over used card buyers, especially if Nvidia makes miner GPUs sans video outputs, which brings us to... 3) This will only make more e-waste once those output-less miner GPUs get chucked, especially if no one finds a secondary purpose for them. At least that's what I'm getting out of it. Any way you look at it, Nvidia gets to make bank. I wouldn't be surprised if they even jacked up the MSRPs of their next-gen cards just to recoup whatever profits they lost out to the scalpers.
|
|
#
?
Feb 21, 2021 07:57
|
|
|
The Gadfly posted:The main factor in bringing costs down for the consumer is competition. Just because a company makes more profit from businesses through softlocks, doesn't mean that they pass the profits on to the typical retail consumer out of the goodness of their hearts. They just pocket the excess profit for themselves, unless their competitors force them to lower prices. The reason that prices are the way they are, is that the business did a market analysis on what sale price would optimize their profit. Why is it that the GPU thread is the most libertarian hangout on the forums these days? There are all kinds of examples of objects that could theoretically do a lot more being designed in a way so they don't. Basically all of the white goods in your kitchen are an example of this. Those are for safety reasons, obviously, but there are often considerations like "if we limit the actual capabilities of X, then we're able to provide support for it a lot better, which will bring down our running costs". A company has a responsibility to sell you the thing that can do what it is advertised to do, not sell you the thing that meets the limits of the physical components contained in a product (especially not a product like a processor). You also get legal regulations in countries etc that mean a lot of products are artificially capped so they don't fall into certain tax brackets, especially when it comes to power consumption. As another poster said, its really rational to feel somewhat hard done by if you know a limit is being placed on a product. But there's no indication they'll be doing this retroactively with existing cards - and I think when someone offers to sell you something they have every right to do it entirely on their terms. Nobody's forcing you to buy the GPU to can't mine well, just like nobody's forcing you to buy the DRM'ed video game or the album locked to one ITunes account or whatever. Same ballpark. I do agree with Linus that it is a huge shame that it doesn't look like there'll be any way those miner-specific cards can be repurposed down the road. We don't need more e-waste at this point. Although (not that it excuses NVidia), 'bad for the planet' is basically crypto's whole MO. The Grumbles fucked around with this message at 11:36 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 11:29
|
|
|
The mining GPUs could be repurposed if NVidia allowed the pass-through hack that lets you run video out through the motherboard, but if they wanted to allow that they probably would have just stuck on at least one video out on the cards on the first place
|
|
#
?
Feb 21, 2021 11:33
|
|
|
Maybe when the bubble bursts the miners will have a crisis of conscience, start folding proteins with all those GPUs, and we'll cure cancer
|
|
#
?
Feb 21, 2021 11:37
|
|
|
It's funny when you think about it. Nvidia's wasting precious resources to sell a purpose built, highly sophisticated device that does nothing (except burn energy).
|
|
#
?
Feb 21, 2021 13:49
|
|
|
Rinkles posted:It's funny when you think about it. Nvidia's wasting precious resources to sell a purpose built, highly sophisticated device that does nothing (except burn energy). Just like mining in general. Apropos
|
|
#
?
Feb 21, 2021 14:46
|
|
|
Y'all are veering off into non-market, non-tech related crypto talk. Please keep it on topic.
|
|
#
?
Feb 21, 2021 17:10
|
|
|
So whens the 3080 Ti hitting shelves?
|
|
#
?
Feb 21, 2021 17:43
|
|
|
spunkshui posted:So whens the 3080 Ti hitting shelves? “Shelves” Who TF knows at this point
|
|
#
?
Feb 21, 2021 17:50
|
|
|
spunkshui posted:So whens the 3080 Ti hitting shelves? I see three possibilities: 1. NVidia spins off new 3090 / 3080 / 3070 / 3060Ti SKUs, but it's JUST the mining nerf, BUT they also don't call it a full refresh because it's the same specs with just the mining nerf applied. Say they call it the 3095 / 3085 / 3075 / 3065Ti. Those new SKUs could be coming out within this year, but without any performance improvements I don't know if it'd really matter to anyone waiting for a 3080Ti for performance reasons. Perhaps the real performance improving refresh comes later, using the SUPER suffix instead of Ti. 2. NVidia spins off new SKUs, and it's just the mining nerf, and they already start calling it the Ti, but that would end up kinda misleading if a Ti doesn't actually carry any performance improvements. Perhaps the real performance improving refresh comes later, using the SUPER suffix instead of Ti. 3. NVidia combines applying the mining nerf to the older SKUs with the performance upgrades. Since we've already had persistent rumors of a 3080Ti that they just keep putting off because the current models are already flying off the shelves anyway, this might STILL mean that such a thing would come out this year regardless, since they'd "just" combine the existing 3080Ti spec with the mining nerf, and utilize the mining nerf strategy as a justification to come out with a new SKU sooner. Especially since a 3080Ti would probably have such slim performance gains relative to existing SKUs with the exception of more VRAM.
|
|
#
?
Feb 21, 2021 17:57
|
|
|
My assumption is that the 3080Ti release is mostly dependent on G6X supply loosening up, since presumably it's going to be a 20GB part. If G6X isn't the constraint, why wouldn't they already be die-harvesting a 3080Ti with a $1200 MSRP or something? Especially if they throw the 3060 mining handicap onto it, they could really push the MSRP high while keeping good street price positioning relative to the 3080/3090. I haven't actually seen any memory news lately, so maybe I'm off base, but that just seems likely.
|
|
#
?
Feb 21, 2021 18:04
|
|
|
Actually, the new rumor is the 80Ti will be a 12GB G6X part. The amount of buffer the 80 should've had originally. -_-
|
|
#
?
Feb 21, 2021 18:10
|
|
|
12GB would mean the full 384bit bus is unlocked too so it would have a bandwidth advantage over a 20GB one with a double stacked 320bit bus
|
|
#
?
Feb 21, 2021 18:17
|
|
|
i'm confused, is nvidia making mining-specific cards or are they making cards that can't be used for mining, alongside cards that could be used for mining or gaming?
|
|
#
?
Feb 21, 2021 18:33
|
|
|
They are making mining-specific cards (the CMP series), and also releasing a singular card (the 3060) which is supposedly designed to be bad at mining The 3070/3080/3090 are still good for either gaming or mining but there's rumours they might be phased out and replaced with new variants that have mining throttled like the 3060
|
|
#
?
Feb 21, 2021 18:37
|
|
|
wyoak posted:i'm confused, is nvidia making mining-specific cards or are they making cards that can't be used for mining, alongside cards that could be used for mining or gaming? yes You have CMP which is for mining only, no display outputs. Cant be used for gaming at all. Then you have the upcoming 3060 which has special "unhackable" drivers to make it bad at mining ETH while still being fantastic at mining shitcoins Going forward those drivers will be applied to the rest of the 30xx newly produced gpus. Already sold GPUs are grandfathered in and can mine ETH and game without restriction. Then you have AMD which is doing none of it but it doesnt matter because there is almost no supply of anything 6xxx Fauxtool fucked around with this message at 18:47 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 18:41
|
|
|
wyoak posted:i'm confused, is nvidia making mining-specific cards or are they making cards that can't be used for mining, alongside cards that could be used for mining or gaming? Yes, nVidia is making mining-specific cards. The "nerf" they're putting in only affects Ethereum mining (which to my admittedly limited of knowledge of all this crypto horseshit has no simple ASICs), supposedly mining any other coins on a 3060 works just fine. For now. This whole thing stinks from all angles, whether you're upwind or down. nVidia is trying to spin this as an anti-miner measure because they wanna help DA GAMERZ, but what it really seems to be is their attempt to become the arbiters of the decentralized cryptocurrency people are mining on their products. I kinda hope this becomes the bad decision that takes the piss out of their stock price, but I doubt it will be. EDIT: And yeah, the fact that this creates "pre-nerf" 30-series cards is also peak  .
BIG HEADLINE fucked around with this message at 18:51 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 18:49
|
|
|
gradenko_2000 posted:The mining GPUs could be repurposed if NVidia allowed the pass-through hack that lets you run video out through the motherboard, but if they wanted to allow that they probably would have just stuck on at least one video out on the cards on the first place Used high-end ex-mining cards have competed against new midrange gaming cards a fair amount over the past few years. Lots of people don’t take that option, but there’s some reduction in demand for newer hardware thanks to cheap used stuff that is still acceptably fast. By putting out mining-specific hardware that can’t be used for games, they’re helping ensure better future sales for their gaming hardware. Nvidia’s segmentation goes both ways here: they would really prefer to have multiple product lines that don’t overlap at all.
|
|
#
?
Feb 21, 2021 18:50
|
|
|
They can also reserve more silicon than they need for just regular gpus and keep it out of the competition's hands. They dont have to worry about marketing weaker chips and selling lower profit skus when they can just toss it at CMP where it will sell easily at what is assumed to be big markup. They could even remove the in between GPUs like 3060ti from the lineup entirely and really get rid of overlap Still dont like it and think they should just open everything and let the hardware speak for itself, quadros included. Fauxtool fucked around with this message at 18:57 on Feb 21, 2021 |
|
#
?
Feb 21, 2021 18:54
|
|
|
So what is the bet for Nvidia will do with non-3060 cards, because they're probably too close to legal trouble if they don't rename the cards. A Super refresh is probably a while away (2080 Super was in July by the same schedule), so what kind of naming do you go with?
|
|
#
?
Feb 21, 2021 18:57
|
|
|
|
| # ? Apr 28, 2024 22:37 |
|
|
So does this mean I can resell/trade in my non-locked 3060 ti, which will retain or go up in value since it can be used for gaming or mining, for a mining-locked 3070 super/ti/whatever when it drops? Seems like a pretty good deal imo
|
|
#
?
Feb 21, 2021 19:02
|
|
