NVIDIA's behaviour is nothing short of baffling (and frustrating). They seem intent on destroying their own reputation. This generation has been an absolute disgrace and yet there seems to be no sign of them relenting. Yes, they are set to release the 4070 at a lower price than anticipated but by all descriptions, it is a miserable piece of hardware and can't even begin to justify the lower price. I mean what the hell are NVIDIA thinking? Are they really this stupid and out of touch?
I am one of the people you talk about. I have always bought the 70 or 80, and do so every few years, usually in response to a new game that's come out that my old GPU struggles with!! For many years that was £300 ish. Now, sure, there is inflation, but £1000+ has just pushed this in to a realm where I am unhappy to go. Especially since NVIDIA keep messing about, deliberately crippling their products with too little memory. I mean if I am to spend £1000 on a card it has to last a lot longer than two years.
I am just so unhappy with the state of NVIDIA products right now. I have no idea where I will go next.
Because Nvidia spun that all the huge sales they got during the pandemic was definitely not mining. Now it's quite clear mining was a real factor in it all,and have sort of promised more of the same revenue/margin increases.
The reality is that Ada Lovelace is a huge generational improvement tier to tier,and even RDNA3 is pretty solid,especially if you think AMD has had less of a process node jump. So if we look at the latest TPU review:
ASUS x Noctua is back! This time the GeForce RTX 4080 gets the low-noise treatment, with impressive results. Testing in our review confirms that even at full load, running at 4K with more than 60 FPS, the card is nearly inaudible! We measured less than 24 dBA, which is quieter than the room...
www.techpowerup.com
So:
1.)GA102(RTX3090TI) to AD102(RTX4090) - 46% faster at 4K for the RTX3090TI or 65% for the RTX3090
2.)GA104(RTX3070TI) to AD103(RTX4080) - 75% faster at qHD
3.)GA106(RTX3060) to AD104(RTX4070TI) - 76% faster at qHD
4.)Navi 22(RX6700XT) to Navi 31(RX7900XTX) - 2.13X faster at qHD
5.)Navi 21(RX6900XT) to Navi 31(RX7900XTX) - 41% faster at qHD
This comes from nearly a two node jump from Samsung 8NM to TSMC 4N 5NM. Except Nvidia has tried to rejig the tier of dGPU used in each product to pad their margins and profits upwards. Even if the node costs more,things like VRAM costs and other parts have gone down a lot in price too.Why I am having two figures for Navi 31,is AMD is also doing some renaming too. Since the memory chiplets are essentially made on a slightly improved 7NM process,it's more a halfway house between Navi 21 and Navi 22,if it were on monolithic die. It certainly would be under the 520MM2 die size of Navi 21 on TSMC 5NM.
But the worst thing is they could have still pushed the dGPUs used one tier up and still ended up with a good price/performance improvement. An example:
1.)RTX3090/RTX3090TI(GA102) 24GB to RTX4090(AD102) 24GB
2.)RTX3080TI(GA102) 12GB to RTX4080TI(AD102) 20GB at current RTX4080 pricing
3.)RTX3080(GA102) 10GB to RTX4080(AD103) 16GB at £700
4.)RTX3070TI(GA104) 8GB to RTX4070TI 12GB(AD104) at £600
5.)RTX3070(GA104) 8GB to RTX4070 12GB(AD104) at £500
6.)RTX3060TI(GA104) 8GB to RTX4060TI 12GB(AD104) at £400
7.)RTX3060(GA106) 12GB to RTX4060(AD106) at £300.
In the above tiers,essentially the dGPUs have all been pushed up one level compared to Ampere and increased in price. Most people might put up with it,if performance was increasing by around 50% at similar pricing compared to the last generation. Nvidia would still be making as much or money as before IMHO.They did this to a lesser degree during Kepler(all of us warning people about what the Titan class would eventually lead to happened),and some other generations for brief periods and it sort of worked. But it seems they feel confident to just to repeat Turing but in a worse way. Turing also happened after a mining boom too,and it was when Nvidia twigged they could clear older inventory at minimum price cuts.
You mean corporate welfare?
Only if they are about to go belly up due some excess mis-speculation, then the state picks of the tab!
As for GPU coolers and motherboards getting heavier: while I have been impressed with how heavy my B550 board is (and I had to RMA the Mortar before I got the B550M Asus so got to feel them both), I wonder whether some isn't just to make them seem fancier. Lead core to add a few extra grams!
Those bargain boards from way back when (Skt775 mATX board which could take a Core 2 Celeron or Pentium to twice its rates) were £35 or so. Okay, those did not have many features but they did the job. Still unsure whether PCIe 5.0 is actually a good feature on consumer boards or not - although I have listened to CAT's concern that newer lower end GPUs (at mid-range prices, of course) will be lane restricted. My unsureness was more that PCIe 5.0 SSDs are far too power hungry. Never used to even thing about PCIe's power requirements until the X570 came with the fan - yes they were able to respin it eventually and it was a repositioned chipset - but currently it seems to be Moore's Law is dead (the economic one, not the scaling at any price) and instead we will overclock everything to the max out the factory gate. Smaller dies clocked to absolute max with zero concerns for efficiency. If companies don't rein this in soon, the EU or similar will have to.
But the corporate welfare happens even when they are doing well. Then on top of that instead of keeping money away for a rainy day,when they screw up the taxpayer picks up even more of the tab. Trickle down economics my arse,especially as many try and get away from paying enough tax too.
The platform PCI-E 5.0 segmentation isn't being picked up on reviewers at all,or many enthusiasts. Instead you had enthusiasts on here defending the artificial segmentation of PCI-E 5.0 on cheaper motherboards,which are being sold high end prices. Then at the same time,defending low VRAM amounts on these same low end dGPU sold as "mid range" dGPUs. AMD ironically was the first to do this with the RX6500/RX6600 series. Now Nvidia has followed suit by pushing rubbish like the RTX4060/RTX4060TI with only 8GB of VRAM,and apparently only PCI-E 8X links. Just imagine if this trend continues when the RX9600XT 12GB/RTX6060 12GB only have PCI-E 8X links,and starting to have issues on PCI-E 4.0 motherboards? People are too emotionally trusting of these companies - PCMR really need to stop thinking criticism of these companies is some sort of dishonour which needs to be defended. They need to get into their minds that we need to criticise all these companies for their games.
Have people forgotten,that what Intel and Nvidia did in the last decade was helped by apathy? The same is starting to happen with AMD. They might be the lesser of two evils,but relativism does not work too when the situation is already dire.
Not anytime soon, if the trend continues there will be tough decisions to make on writing off massive amounts of stock or selling it off and "undervaluing" everything they are trying to acheive.
Many of these companies are happy to stomach the downward trend atm after the record profits from the last few years.
It's already happening - you can get prebuilt systems and laptops with previous generation dGPUs for less than building one yourself.
Except all the PCMR lot on the internet/social media don't realise Nvidia does not want to reduce prices to enthusiasts at all. Yet enthusiasts/gamers still try and defend them.
www.youtube.com
Plus the more power efficient card.