Latest TSMC Node GeForce RTX 5000 Series
Nvidia’s GeForce RTX 4000 series GPUs are already among the best graphics cards on the market.
Nevertheless, rumors of next-generation GeForce RTX 5000 series (Blackwell) GPUs are already emerging, even if we don’t seem too keen on this upgrade path.
Our friendly and close leaker kopite7kimi, who has a proven track record, has provided us with details about the successor to Ada lovelace.
According to this leaker, the GeForce RTX 5000 series graphics cards will reportedly use TSMC’s 3nm process node.
The current GeForce RTX 4000 series GPUs use a 5nm process node, so the logical next step would be 3nm.
At the 3nm process node, TSMC promises up to 15% performance improvement compared to the 5nm process node at the same power level and transistor count.
In terms of power savings, the foundry claims that the 3nm process node can reduce power consumption by as much as 30%.
But what’s impressive about TSMC’s 3nm process node is that the die size is estimated to be about 42% smaller than the 5nm process node.
So it would be interesting to see this improvement reflected in Blackwell’s silicon.
For example, the AD102 that powers the flagship GeForce RTX 4090 is a massive 609 mm² die.
The GA102 that Nvidia used for the previous GeForce RTX 3090 Ti was Samsung’s 8nm process node and measured 628 mm².
However, this time, Nvidia is said to be sticking to TSMC, making the transition from 5nm to 3nm even more interesting.
GeForce RTX 5000 series graphics cards are said to support DisplayPort 2.1.
Nvidia is late to the game, as AMD has already adopted DisplayPort 2.1 in the chipmaker’s existing Radeon RX 7000 series products like the Radeon RX 7900 XTX.
Nvidia’s current GeForce RTX 4000 series graphics cards only have DisplayPort 1.4a connectivity.
AMD’s implementation differs between its consumer and workstation lineups.
The Radeon RX 7000 series supports UHBR13.5, giving you access to 54 Gbit/s of bandwidth.
Meanwhile, the W7000 series fully supports UHBR20, with a maximum bandwidth of 80 Gbit/s. Unfortunately, kopite7kimi’s comment is vague and does not specify which standards Nvidia can use with Blackwell.
Other rumors suggest that GeForce RTX 5000 series products will feature a PCIe 5.0 interface, which is overkill for mainstream graphics cards.
What’s more likely is that Blackwell will continue to use the 16-pin power connector, but perhaps rely on a revised 12V-2×6 version.
This 12V-2×6 standard could result in more secure interconnects, with Nvidia remaining largely silent on the 12VHPWR meltdown fracas at the expense of consumers, other than generally blaming user error. There are high expectations.
Contrary to Nvidia’s conclusion and the opinions of several others in the industry, one of the many professional repair technicians who receives several damaged RTX 4090s per month believes that the main root cause is the connector; He stated that third-party adapters are important contributors and emphasized that user error is one possibility, if not another.
It’s always a good idea to be aware of industry-wide oversights like this. It would be great if Nvidia and other companies adopting the standard emphasized the ATX 5.0 specification and made third-party cables, expansions, and adapters safe.
At this point, nothing is certain about Blackwell’s SKUs.
Of course, there are rumors about more cores and GDDR7 memory for the GeForce RTX 5000 series, but for now, take them with a grain of salt.
Explanation:
Next-generation Geforce, RTX5000, Blackwell information
The auxiliary power standard will continue to be 12VHPWR, but its revised version, 12V-2×6, is expected to be adopted.
It seems that it has already been secretly adopted in current GPUs, but we apologize to those who purchased it without knowing it.
Looking at what various people have said on X, there are no problems reported in Japan, so it seems that there will be no problem even if the secret standards are changed.
That was good (black lol)
For those of you who bought it without knowing it, how do you feel about it? (bitter smile
Okay, enough venom, let’s look at some more information.
It is certain that TSMC3nm will be adopted, and if the number of transistors is the same, it will require 42% less die area and improve performance by 15%.
GA102・・・10,752AD102・・・18,432GB202・・・24,576
The number of CUDA cores is shown above.
AD102-GA102=7,680GB102-AD102=6,324
Unfortunately, the increase in the number of CUDA cores is smaller than with Ada Lovelace.
If AD102 is set as 1 as a percentage,
GB202：AD102：GA102=1.333：1：0.583
and both the number and percentage of increases are increasing from Ampere to Ada Lovelace.
This is still considered the biggest performance leap in nVIDIA history, so DLSS4 will likely bring about a significant performance improvement.
Since the number of transistors is not specified, it is difficult to guess, but it may be possible to make a lot of changes internally.
The reason why nVIDIA is so bullish is probably because the AI/ML software improves performance more than the performance improvement of hardware.
Rival company AMD has just finally implemented frame generation with FSR3, but there is a possibility that they will be able to differentiate themselves again with DLSS4.
Intel has not yet implemented frame generation.
In the future, it seems likely that the overall completeness of the product, including drivers and AI/ML-related software, will determine the outcome of the product.
If this happens, Radeon, which is weak in software, will be in a lot of trouble, and Intel, which is still a low-quality product and is not at the starting line with AMD and nVIDIA products, will also be in a lot of trouble.
Also, it is said that AMD failed in the design of Navi41/42 with the next generation RDNA4 and can only release Navi43 (= RX8600 equivalent).
The next generation RTX5090, RTX5080, and RTX5070 may increase in price significantly.
Since there is no competition, there is no problem with nVIDIA no matter how much it costs.
AMD’s GPU Radeon series
Radeon 7000 series
SAPPHIRE
¥288,800 (As of 2023/11/18 10:15:15 Amazon research – details)
SAPPHIRE
¥148,322 (Amazon research as of 2023/11/18 10:16:31 – details)
SAPPHIRE
¥42,500 (As of 2023/11/18 10:18:48 Amazon research – details)
Radeon RX 6X50 series
SAPPHIRE
¥83,333 (as of 2023/11/18 15:32:02 Amazon research – details)
Radeon RX 6000 series
SAPPHIRE
¥35,000 (As of 2023/11/18 13:16:13 Amazon research – details)
SAPPHIRE
¥29,800 (As of 2023/11/18 09:45:19 Amazon research – details)
SAPPHIRE
¥27,878 (As of 2023/11/18 15:32:03 Amazon research – details)
* SAPPHIRE is a manufacturer specializing in AMD Radeon and is a reference manufacturer for Radeon.
