Blackwell gpus

[Jono8]

The scene has moved on, there are now FPGA bitstreams and ASIC units that blow away dGPU cards for efficiency and profits. Its no longer feasible to buy Ada cards for example as nvidia has reserved the scalper margins and just now charge an extra 20%+ from the Ampere era. Electricity being twice as expensive as it was has crippled most from bothering because for a dGPU to now ROI its years instead of the days from four years ago.

Apple have introduced their own M chips, as @CAT-THE-FIFTH touched on above there are "AI" cpu's already breaking in with laptops (much like phones) so it wont be long before segments of the pie establish and the gaming GPU is pushed aside to where it was a couple of decades ago.

Yes I know.

I was just stating why you couldnt really get one.

Because they were essentially free.

At that time, 3000 series GPU's worked out free if you stuck them on nicehash for a few months.


That sort of insanely rabid demand wont be repeated.

 

[scubes]

Hopefully we will see some leaks soon ..

 

[chaparral]

How much longer are we going to get big performance improvements when buying the next gen cards ?

Surely there got come a time when even nvidia will struggle to bring out faster & faster GPU's

 

[gpuerrilla]

I was just stating why you couldnt really get one.

Because they were essentially free.

Oh I know, its why my current card ended up being an earner over the lifespan, still use it for that but the demand from miners till now will never be as bad as 2020/21 for the reasons I mentioned earlier. The only long term issue is the retail price has stayed high even an enthusiast welps at the current market prices.

 

[brostradamus]

How much longer are we going to get big performance improvements when buying the next gen cards ?

Surely there got come a time when even nvidia will struggle to bring out faster & faster GPU's

tsmc's working towards a goal of 1 trillion transistors by 2034, assuming its nothing more than a slogan we have some headroom, a lowball estimate of 250 bn by 2030 is still going to be a good result
there also are reports suggesting that the next wave of lithography tech is going to be prohibitively expensive, but its not going to be a concern till the mid 2030's

 

[Jay85]

tsmc's working towards a goal of 1 trillion transistors by 2034, assuming its nothing more than a slogan we have some headroom, a lowball estimate of 250 bn by 2030 is still going to be a good result
there also are reports suggesting that the next wave of lithography tech is going to be prohibitively expensive, but its not going to be a concern till the mid 2030's

Cant imagine the power requirement for that... There's been a steady trend of increasing wattage per generous.