Predictions on when GPUs will be back at RRP.

[Rob_B]

Has it finally done it? If so then that's great news for various reasons!

[Percy1983]

Has it finally done it? If so then that's great news for various reasons!

2 days 19 hours 8 minutes to go. (estimated on current hash rate)

Past the getting used GPUS, it will cut the electricity used massively which from any point of view is amazing.

Hopefully this will be the end of GPU mining.

[DanceswithUnix]

2 days 19 hours 8 minutes to go. (estimated on current hash rate)

Past the getting used GPUS, it will cut the electricity used massively which from any point of view is amazing.

Hopefully this will be the end of GPU mining.

I see Google have an estimator running with a pair of pandas trying to merge:

https://www.google.com/search?q=ethereum+merge

and if you check out the total hashrate (click "All-time") : https://2miners.com/eth-network-hashrate

there's been a bit of a drop but it looks like there are a lot of GPUs out there still waiting to find a new home. I give it at least a week after merge before the owners realise they can't use them profitably on any other coins, and ebay goes nuts with old mining gear.

[kompukare]

I see Google have an estimator running with a pair of pandas trying to merge:

https://www.google.com/search?q=ethereum+merge

and if you check out the total hashrate (click "All-time") : https://2miners.com/eth-network-hashrate

there's been a bit of a drop but it looks like there are a lot of GPUs out there still waiting to find a new home. I give it at least a week after merge before the owners realise they can't use them profitably on any other coins, and ebay goes nuts with old mining gear.

I've tried in vain to argue with those rose-tinted glasses miners and their "we'll just find another coin to mine". Even graphs and screenshots of total network hashrate simply do not convince them. A lot irrationality with miners - about the only ones more deluded are those who bought some future crypto coin with cash or borrowings!

[Percy1983]

I look forward to the race to the bottom as they realise, I am not to worried about a mining card if cheap enough and I am more than happy to clean and re paste if needed.

The big hits will be the warehouses full as they may still have commitments to rent and no income over night with thousands of cards to sell.

[DanceswithUnix]

The big hits will be the warehouses full as they may still have commitments to rent and no income over night with thousands of cards to sell.

It was pretty obvious what was going to happen for some months now, if not years. I can understand someone running their mining rigs right up to the last block to squeeze every last drop out of them, but if they sold up a few months ago the GPUs would have had way more second hand value.

The total hashrate graph is interesting, I suspect the recent drop during May and June is down to European energy prices going through the roof. That would imply that many cards around here are already offline. But even if I'm right, cards dumped on the market in other countries would impact prices here after a short delay if there is a global oversupply.

A bit early in the day for thinking I know but... AMD and Nvidia must know this. I wonder if they are held off on their next gen GPU releases until after the merge. The new GPUs will be on a new process, with perhaps some new architectural toys thrown in. If they can make the new cards twice the performance of the old cards, then people may still want to buy them new over cheap second hand kit. That would really hose people with a warehouse full of mining rigs.

[Saracen999]

It was pretty obvious what was going to happen for some months now, if not years. I can understand someone running their mining rigs right up to the last block to squeeze every last drop out of them, but if they sold up a few months ago the GPUs would have had way more second hand value.

...

Maybe, but ever had some shares that were going up and, despite knowing you ought to sell before they peak, held off just a tad too long? I guess the simple term for it is greed. The same psychology may apply to getting out of mining - the smart more is obviously before it peaks and declines but it means foregoing some profit so .... tomorrow? Next week?

...

A bit early in the day for thinking I know but... AMD and Nvidia must know this. I wonder if they are held off on their next gen GPU releases until after the merge. The new GPUs will be on a new process, with perhaps some new architectural toys thrown in. If they can make the new cards twice the performance of the old cards, then people may still want to buy them new over cheap second hand kit. That would really hose people with a warehouse full of mining rigs.

The rationale I've seen is more to do with over-committment on fab contracts, because they could basically sell (at daft prices) anything they could make and then .... mining collapse, energy prices etc. That (reportedly) led to serious over-committment to RTX3000's.

Jayz did a video a few days ago on nvidia's latest report to investors where they pretty much stated that, unlike the usual process, they weren't dumping RTX3000 prior to releasing RTX 4000, but rather, at least for two quarters, were going to run RTX3000 and 4000 lines in parallel.

The logic seems to have been to clear large stocks of high-end 3000s but keep low-end. That may reflect how many they have of what, and what those purchase contracts were for. It also sort-of ties in with fairly high-end 4000s coming out first (as usual) but may well imply low end 4000s are mid-'23 or later.

The logic went on to suggest that recent RTX3000 price cuts weren't about dumping stock to clear ready for 4000, but more about resetting (or, put less charitably, fixing) the price differentials across the range to suit the above strategy of releasing high-end 4000s but running much of the 3000 series in parallel, at least for two quarters.

It took a bit of translating investor-speak into English but that seemed to be what nVidia's CEO was saying had been, and was going to, happen.

Price 'fixing' illegal? Not really. It depends how you do it. By formal agreements with other companies to fix prices? Oh, yeah. But by just deciding on how much of your own stock to release to resellers, and at wha price? Nope. Arguably unethical, and certainly a bit manipulative but not illegal. Of course, it only works because, at least for a good chunk of the GPU market they are not only dominant but a monopoly supplier.

All told though, only very senior nVidia management actually know and everybody else, including "reviewers" are just guessing. Some may guess right, but some, not so much.

[Percy1983]

Maybe, but ever had some shares that were going up and, despite knowing you ought to sell before they peak, held off just a tad too long? I guess the simple term for it is greed. The same psychology may apply to getting out of mining - the smart more is obviously before it peaks and declines but it means foregoing some profit so .... tomorrow? Next week?

The rationale I've seen is more to do with over-committment on fab contracts, because they could basically sell (at daft prices) anything they could make and then .... mining collapse, energy prices etc. That (reportedly) led to serious over-committment to RTX3000's.

Jayz did a video a few days ago on nvidia's latest report to investors where they pretty much stated that, unlike the usual process, they weren't dumping RTX3000 prior to releasing RTX 4000, but rather, at least for two quarters, were going to run RTX3000 and 4000 lines in parallel.

The logic seems to have been to clear large stocks of high-end 3000s but keep low-end. That may reflect how many they have of what, and what those purchase contracts were for. It also sort-of ties in with fairly high-end 4000s coming out first (as usual) but may well imply low end 4000s are mid-'23 or later.

The logic went on to suggest that recent RTX3000 price cuts weren't about dumping stock to clear ready for 4000, but more about resetting (or, put less charitably, fixing) the price differentials across the range to suit the above strategy of releasing high-end 4000s but running much of the 3000 series in parallel, at least for two quarters.

It took a bit of translating investor-speak into English but that seemed to be what nVidia's CEO was saying had been, and was going to, happen.

Price 'fixing' illegal? Not really. It depends how you do it. By formal agreements with other companies to fix prices? Oh, yeah. But by just deciding on how much of your own stock to release to resellers, and at wha price? Nope. Arguably unethical, and certainly a bit manipulative but not illegal. Of course, it only works because, at least for a good chunk of the GPU market they are not only dominant but a monopoly supplier.

All told though, only very senior nVidia management actually know and everybody else, including "reviewers" are just guessing. Some may guess right, but some, not so much.

I guess they will release the top end 4000 series and sell an artificially restricted stock 3000 series has the mid to bottom end so everything can have a top end price without falling over other models, why release say a 4070 which makes all the previous top end look silly and devalues it all.

I genuinely hope a flood of used cards messes this up for them, as yes they can manipulate the price of new cards but if used is so much cheaper many will accept mining cards at a quarter of the price of new.

[cptwhite_uk]

If I can find a RTX 3080 for sub-£500 in the next month, I may bite....I've had incremental upgrades recently...... GTX1080 (£420 used) > RX 5700XT (£280 new) > RTX 2070 Super (£300 used) > GTX 1080 TI (£260 used) > RTX 3060 Ti (£369 new, sold for £475) > RX 6800 (£575 used)

RX 3080 would be another +20% performance bump.

[DanceswithUnix]

The logic went on to suggest that recent RTX3000 price cuts weren't about dumping stock to clear ready for 4000, but more about resetting (or, put less charitably, fixing) the price differentials across the range to suit the above strategy of releasing high-end 4000s but running much of the 3000 series in parallel, at least for two quarters.

That does make some sense for Nvidia.

If the 4090 is indeed twice as fast as a 3090 for the same power as a 3090 ti, then there isn't an overlap. A quick look on Overclockers and it looks like stock of 3090 there at least is quite low, which makes sense as people who want the fastest and/or latest aren't going to want to buy a 3090 once the 4090 is out. Similarly, what budget builder will go for a 3090?

Also, the 3000 series silicon is made by Samsung, and 4000 at TSMC so both can be made in parallel with improved availability.

[Percy1983]

Since miners became a thing I have been buying selling at peaks:

Bought a 290x for £190 and sold it later for £220.00
got an rx580 for £210, sold it for £230
got the Vega56 for £205, sold for £475

So since my 290x I am £320 up so if I buy one for £320 I will net zero spend on graphics cards since 2014.

I will say this has been my longest time between cards and luckily had a R9 290 spare to get me through.

[ik9000]

That does make some sense for Nvidia.

If the 4090 is indeed twice as fast as a 3090 for the same power as a 3090 ti, then there isn't an overlap. A quick look on Overclockers and it looks like stock of 3090 there at least is quite low, which makes sense as people who want the fastest and/or latest aren't going to want to buy a 3090 once the 4090 is out. Similarly, what budget builder will go for a 3090?

Also, the 3000 series silicon is made by Samsung, and 4000 at TSMC so both can be made in parallel with improved availability.

Yeah it's a weird one. Say you have £1k to spend (a nice dream granted) do you go 3090 now for 24G and 350W power and no need to change PSU, or wait to see where the 4080 16G /4090 24G land price-wise with a 450W-800W* peak rumoured power draw and new PSU required for most people? I can't see the 4090 being £1k or less and i'd be surprised if the 4080 is much lower than 1k, and possibly more given the stated intent to keep the 3000 series retailing for the near future. If we knew the new pricing RRP it would be a lot easier to make a call.

*There are rumours the stated massive performance gains are for boosted cards running at >600W OC and that the boosts vs 3000 series won't be so high for cards sat at "standard" 450W (though how you can justify 450w standard for a GPU seems weird. My old P4 system only had a 500W PSU back in the FX5000/6000 days.

[kalniel]

Partner board 3090s probably pulled quite a lot more than 350W, especially in transient. I don't think the 4080 is going to be hugely different in terms of real PSU headroom required.

[DanceswithUnix]

*There are rumours the stated massive performance gains are for boosted cards running at >600W OC and that the boosts vs 3000 series won't be so high for cards sat at "standard" 450W

That's quite possible. The process change should mean easily double the transistors for the same size silicon and GPUs scale well (unless you are Intel ) so double the performance (or more) is certainly on the table. The problem is taming the heat from twice the fets toggling away, as the process change no longer gives you that the way it did in 199x years.

I'm sure in the middle of the stack you could, for instance, get the performance of a 3070 with less silicon and less heat making for a cheaper and better GPU. The problem is what they do at the top end. Things may well be heat limited, in which case if (number from thin air guess) the new process gets Nvidia 30% less heat for the same clocks then they could end up making a 3090 replacement that's only 30% bigger die size to keep the same overall thermals with a corresponding 30% performance bump. Or they can max the reticule size, add a ton of shaders and then underclock to get temperatures down. There's a few knobs to tweak there.

With Intel bullishly gunning for high end ARC cards a while back, Nvidia may have gone for a big die to try and stay way out of Intel's reach. Basic specs like die size were probably made a year ago after all.

But to get back on thread, if it is a massive die and we predict that the RRP for a 4090 will be £2000, then I'm not sure I much care when it is RRP as I'm not in that market!

[Percy1983]

Less than 24 hours to merge.

Been tracking 6700xt cards on ebay and they are falling under my £320 budget. More interestingly on each listing I clicked other listings and they were all selling more than one card, so of course this is the smaller miners who see the writing on the wall and are getting out quick.

I will wait a little longer.

[ik9000]

That's quite possible. The process change should mean easily double the transistors for the same size silicon and GPUs scale well (unless you are Intel ) so double the performance (or more) is certainly on the table. The problem is taming the heat from twice the fets toggling away, as the process change no longer gives you that the way it did in 199x years.

I'm sure in the middle of the stack you could, for instance, get the performance of a 3070 with less silicon and less heat making for a cheaper and better GPU. The problem is what they do at the top end. Things may well be heat limited, in which case if (number from thin air guess) the new process gets Nvidia 30% less heat for the same clocks then they could end up making a 3090 replacement that's only 30% bigger die size to keep the same overall thermals with a corresponding 30% performance bump. Or they can max the reticule size, add a ton of shaders and then underclock to get temperatures down. There's a few knobs to tweak there.

With Intel bullishly gunning for high end ARC cards a while back, Nvidia may have gone for a big die to try and stay way out of Intel's reach. Basic specs like die size were probably made a year ago after all.

But to get back on thread, if it is a massive die and we predict that the RRP for a 4090 will be £2000, then I'm not sure I much care when it is RRP as I'm not in that market!

it looks like it will be BIIIIIG. 3 slots min, 4 on AIB like the zotac factory images, and longer than the 3090 (which is already huge).