Gráfica nVidia RTX 4000 séries "Ada" (Modelos SUPER lançados)

Se esse artigo estiver proximo da verdade, o salto entre AD103 e AD102 é enorme. Muito superior entre o GA103 e o GA102 actual.
Estou curioso se a (suposta) RTX4080 será um AD103 e guardam o canhão para os preços ridiculos.

É que actualmente não justifica muito pagar o extra entre uma RTX3080 e uma RTX3090 por exemplo, mas nesse cenário a diferença seria muito relevante.
 
Se esse artigo estiver proximo da verdade, o salto entre AD103 e AD102 é enorme. Muito superior entre o GA103 e o GA102 actual.
Estou curioso se a (suposta) RTX4080 será um AD103 e guardam o canhão para os preços ridiculos.

É que actualmente não justifica muito pagar o extra entre uma RTX3080 e uma RTX3090 por exemplo, mas nesse cenário a diferença seria muito relevante.
Nestas rtx 3000 comparando a gtx 1080/80ti e rtx2080/80ti , 3080/80ti/90 diferença/preço foi um fiasco total , vamos ver se na rtx 4000 a diferença de performance da 4080/80ti/90 já compensa como as gerações anteriores .
Mas na mesma as rtx 3090/80ti venderam que nem pães apesar de não justificar preço/performance face a 3080.
 
Na perspectiva do consumidor essa diferença acaba por não ser má, porque o que acontece atualmente é quando o produto mais caro vende a NVIDIA acaba por fazer poucas unidades do produto mais barato, se existir uma diferença grande na quantidade de silicio já não será bem assim.
Mas ao fazer esta segmentação acho que demonstra claramente que esperam competição feroz por parte da AMD pelo topo.
 
(Rumores)
For starters, the flagship AD102 die set to power the top-tier RTX 4080, RTX 4080 Ti, and RTX 4090 GPUs will pack a total of 12 GPCs (Graphics Processing Clusters). These will be further divided into 6 TPCs (Texture Processor Clusters) per GPC, resulting in an overall count of 72 TPCs. This totals to 144 SMs (Streaming Multiprocessors) or 18,432 FP32 shaders (half as many INT32). Like Ampere, each SM will pack 128 FP32 and, 64 INT32 cores, in addition to a handful of Tensor and RT cores. The GeForce RTX 4090 will be paired with 24GB of GDDR6X memory via a 384-bit bus, while the RTX 4080 will trim it to 16GB and a 320-bit bus. Finally, the RTX 4080 Ti will likely combine 18-20GB GDDR6X memory with a 384-bit bus. These GPUs will have a TBP of 500-600W and should launch in August or September.

The most interesting part of this leak is the enlargement of the L2 cache. The RTX 4090, 4080, and 4080 Ti will pack up to 96MB of L2 cache which is a massive upgrade over the 6MB featured in Ampere and Turing. Much like AMD’s Infinity Cache (L3), this should improve hit rates, and boost internal bandwidth.

Down the stack, we have the AD103 and AD104 dies. These will power the RTX 4070 Ti, and RTX 4070 with 10,752, and 7,680 FP32 shaders respectively. Both the SKUs should come with 16GB of GDDR6X memory across a 256 or 192-bit bus. The L2 cache will be trimmed to 64MB for the RTX 4070 Ti (AD103) and 48MB for the RTX 4070 (AD104). These GPUs will have a TBP of 300-250W, with a release in late 2022.

Finally, we have the RTX 4060 and 4050 powered by the AD106 and the AD107, respectively. The former will pack a total of 4,608 cores across 36 SMs, and the latter will cut it down to 3,072 cores across 24 SMs. The RTX 4060 will likely be paired with 12GB of GDDR6 memory across a 128-bit bus while the RTX 4050 will combine 8GB across a 128-bit bus. These SKUs should have a TBP of 250W, and 200W, respectively, with a launch slated for early 2023.

https://www.hardwaretimes.com/leake...gain-over-rtx-3090-18432-cores-96mb-l2-cache/
 
Também me custa a acreditar nesse rumor.

A RTX 3070 tem 5888 cores e 256 Bits de Bus.

A RTX 4060 teria 4608 cores e 128 Bits de Bus. Como é que conseguem 12GB de memoria a 128 Bits? Não será 192 Bits?

Mesmo que a 4060 consiga chegar à performance da 3070, o que significa que teria de ser mais eficiente, como é que a 3070 tem um TDP de 220W e a 4060 teria um TDP de 250W?

O processo de fabrico altera de Samsung 8nm para TSMC 5nm, e para a mesma performance o consumo energético ainda aumenta?
 
Sim, com TSMC 5nm era de esperar um ganho absurdo de eficiencia. Mesmo se pegasse as Ampere e portasse para 5nm, ias ter a 3090 a consumir 250w-270w, 3080 com 240-250, 3070 a sub 200, a 3050 a 75w, etc...
 
Caches enormes eliminam a necessidade de barramentos largos e como tal aumentam bastante a eficiência energética e também eliminam a necessidade de colocação de módulos de memória mais rápidos. GDDR7 não deve estar pronta para consumo tão cedo e não me parece que alguém queira colocar HBM neste segmento de mercado.
 
Isso é para regar com sal do himalaia!

450W é loucura, mas não duvido dado que as 3090 chegam as 400w algumas. 650/850 deve ser coisa de server / MCM, tais consumos entraria no dominio do LN2.
 
Se trocares os pads da 3090 FE consegues dissipar uns 500w na boa, não me admirava nada que com algumas afinações não consigam melhorar a eficiência da transferência termica. A EVGA já andou a brincar com BIOS de 500w nesta geração, não me admirava nada que aumentem isso mas isso terá tudo um custo. E os 5nm da TSMC devem trazer clocks absurdos mesmo com a quantidade de cuda cores que se fala.
 

Nvidia in Talks to Use Intel as a Foundry to Manufacture Chips​

As inconceivable as it may sound, your next Nvidia GPU could be manufactured by Intel. Nvidia CEO Jensen Huang held a question and answer session with the press today, and the topic quickly turned to Intel's Foundry Services (IFS) initiative that will see Intel making chips for other companies as part of its IDM 2.0 initiative. Surprisingly, Huang confirmed that his company is considering using Intel's foundry to possibly make some of its chips.
"Our strategy is to expand our supply base with diversity and redundancy at every single layer. At the chip layer, at the substrate layer, the system layer, at every single layer. We've diversified the number of nodes, we've diversified the number of foundries, and Intel is an excellent partner of ours[…]. They're interested in us using their foundries, and we're very interested in exploring it,"
"A service-oriented company that dances with your operations. TSMC is a services company that dances with, what, 300 companies worldwide? Our own operations are quite an orchestra, and yet they dance with us. And then there's another orchestra they dance with. So the ability to dance with all of these operations teams and supply chains teams is not for the faint of heart, and TSMC does it just beautifully. Its management, its culture, and its core values, and you put that on top of technology and products."
"I am encouraged by the work that is done at Intel, I think this is a direction they have to go, and we're interested in looking at their process technology. Our relationship with Intel is quite long; we work with them across a whole lot of different areas, every single PC, every single laptop, every single PC, supercomputer, we collaborate."
"Foundry discussions take a long time, and it's not just about desire. We have to align technology, the business models have to be aligned, the capacity has to be aligned, the operations process and the nature of the two companies have to be aligned. It takes a fair amount of time and a lot of deep, deep discussion – we're not buying milk here. This is really about the integration of the supply chains. Our partnerships with TSMC and Samsung in the last several years are something that took years to cultivate. So we are very open-minded to considering Intel, and I'm delighted by the efforts that they're making."
"We have been working closely with Intel, sharing with them our roadmap long before we share it with the public, for years. Intel has known our secrets for years. AMD has known our secrets for years. We are sophisticated and mature enough to realize that we have to collaborate.[...] We share roadmaps, of course, under confidentiality and a very selective channel of communications. The industry has just learned how to work in that way."
"On one hand, we compete with many companies, but we also partner deeply with them and rely on them. As I mentioned, if not for AMD CPUs in DGX, we couldn't ship DGX. If not for Intel CPUs and all of the hyperscalers connected to our HGX, we wouldn't be able to ship HGX. If not for Intel's CPUs in our Omniverse computers that are coming up, we wouldn’t be able to do the digital twin simulations that rely so deeply on the single-threaded performance that they're really good at."
https://www.tomshardware.com/news/n...el-foundry-intel-and-amd-know-all-our-secrets

Nesta altura, não é assim tão surpreendente.
 
Nestas rtx 3000 comparando a gtx 1080/80ti e rtx2080/80ti , 3080/80ti/90 diferença/preço foi um fiasco total , vamos ver se na rtx 4000 a diferença de performance da 4080/80ti/90 já compensa como as gerações anteriores .
Mas na mesma as rtx 3090/80ti venderam que nem pães apesar de não justificar preço/performance face a 3080.
Ainda me lembro de me dizerem a 2070 Super e boa mas 400€ por uma grafica e muita guita , 400€ e uma RTX 3050 agr lol...

Acredito bem nisso a Intel tb esta a investir bastante em novas fabricas. A TSMC precisa de concorrência e "precisamos" a nivel mundial de aumentar a produção chips que estamos a fazelo , quem prodiuz as maquinas e que nao consegue acompanhar a demanda para tantas novas fabs.
 
Pelos vistos é bem provavelmente que existam SKUs com 600w na próxima geração. 24 fases + 4 para a memória. E a pcb da 3090 Ti FE supostamente é compativel com a nova geração no numero de pinos, o que a ser verdade, facilita o processo de engenharia.

NVIDIA GeForce RTX 4090/4080 AD102 PCB to support up to 24GB of GDDR6X memory, 600W TDP very likely​

https://videocardz.com/newz/nvidia-...to-24gb-of-gddr6x-memory-600w-tdp-very-likely

Artigo original do Igors Lab:

https://www.igorslab.de/en/new-deta...b-gddr6x-and-a-clever-voltage-converter-orgy/
 
images

eu a olhar para a minha fonte de 850W....
 
However, what we see at first glance are a whopping 24 voltage converters for the GPU! Since NVIDIA (this has also been leaked in the meantime) will again rely on a uP9512 that can only generate 8 phases (but is readily available), a total of three voltage converters will most likely be driven together per phase. The four dark rectangles near the drill holes for the cooler should be the VRM for the memory voltage converters. This brings us to four phases for the GDDR6X.
Speaking of cooling, some sources report that NVIDIA will rely on a tripple-slot air cooler for the Reference or FE, while almost all board partners are likely to go with a 3.5-slot design. This means that the weight class above 2 kilograms will certainly also have a lot of new competitors.
:blaugh:
 
Back
Topo