There's 5.0 support on their AM5 MB (at least guaranteed for X670E) so I doubt RDNA3 won't feature 5.0.
Still surprised that there is no 5.0 even for AD102.
Still surprised that there is no 5.0 even for AD102.
Wonder what PCIe 5.0 does for power consumption which is already looking poor. Aside from the cost, power might be a good reason to "only" stick with PCIe 4.0.
That seems like an expensive murder weapon.
I'm not sure if "has blood on it" is a valid reason to RMA either, so I can't recommend throwing you 4090 at another human being.
Saylick
Diamond Member
- Sep 10, 2012
- 4,060
- 9,486
- 136
Funny enough, if you tried throwing a GPU at someone not even 6 months ago they'd gladly welcome it with open arms. Only so they could hopefully catch it and plug it into their PC to mine with it.
coercitiv
Diamond Member
- Jan 24, 2014
- 7,381
- 17,498
- 136
Says the transisitor count is over 75 B. That would be 2.6x+ the number of transistors.
Aapje
Golden Member
- Mar 21, 2022
- 1,530
- 2,106
- 106
But how efficiently are they used? If it scales as poorly as Ampere, that will increase the price way more than the performance.
If the die size is similar at around the 600mm² mark and from what we know about Samsung 8nm and TSCM 5nm, then x2.6 is surely a given.
Huge transistor increase but power doesn't look good at all.
GodisanAtheist
Diamond Member
- Nov 16, 2006
- 8,348
- 9,730
- 136
Wonder how much of that increased transistor budget is going into beefy caches vs compute hardware.
Not much point in compute hardware that you can't keep fed. We've already seen that bigger caches can be a smart move, so I think NVidia is wise to recognize a good idea when they see one.
The area trade off for what you'd need to spend on a wider memory bus isn't so unfavorable either. Even if SRAM doesn't scale as well as logic, it scales better than IO.
The area trade off for what you'd need to spend on a wider memory bus isn't so unfavorable either. Even if SRAM doesn't scale as well as logic, it scales better than IO.
NVIDIA GeForce RTX 4090 & RTX 4080 16GB/12GB max TGP and GPU clocks specs have been leaked - VideoCardz.com
NVIDIA RTX 4090 reportedly with maximum TGP at 660W Multiple sources have now confirmed upcoming desktop NVIDIA cards. We combined specs for upcoming RTX 40 GPUs from multiple sources of ours, but also reports from kopite7kimi and Chiphell who confirm some parts of them. What is new is the...
videocardz.com
Videocardz claims they have the specs for the 4090 and the 4080s and also claim that the 4080 12 GB and 16 GB will be released this year, sometime in November.
It is.... but in fairness, the full(ish) xx104 has normally been what the xx80 card is.
It's a $200 hike if you figure the 3080's real MSRP is $799. It is a much more expensive node.
But that's why there's a 4080 12 GB. People are going to complain about it and this is the "solution".
Don't worry, prices will soon be a lot lower than assumed. Unfortunately, it won't be for nice reasons.
How does a collapse happen? Slowly at first, then all at once.
It'll be okay. But these are likely going to be the only desktop cards released for some time.
It is tactical move by nVidia to rename supposedly RTX4070 to RTX4080 12GB especially using full die of AD104. Beside marketing reason, I believe the main reason is to target AMD Navi 32 which happens to have 7680 cores with 12GB GDDR6 192-bit memory support. nVidia knows about the costing disadvantage against RDNA3, thus naming AD104 as 4080 12GB hope to let people believe it is a good value card. My estimated price of RTX4080 12GB would be between $599 to $699, we will see...
Last edited:
4070Ti supposedly uses AD103 die with 256-bit memory and normally come much later than 4070 just like 30 series launching
Well, if the 4080 12 GB is AD104 that's obviously not happening. The speculation of the 4070 is a cut AD104, 7168 cores and 160 bit.
Well, that is what have been reported, we will know in a few days. It is indeed weird to use full die in the first wave of product launching thus my analysis.
Come to think about 4070Ti, nVidia could use AD103 with 12GB 23Gbps GDDR6X memory instead of 16GB in order to up sale 4080 16GB
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter