Nvidia RTX 50 Series: News, rumors & more

Looking for the latest on the Nvidia RTX 50 series graphics cards? While the GPUs remain elusive and unannounced, leaks and rumors point to another generational leap in performance.

The RTX 50 series of GPUs is inevitably coming. Now that the launch of the 40 Super refresh has come and gone, with the RTX 4070 Super, 4070 Ti Super and 4080 Super officially released, all eyes are on the next big thing on the horizon for Nvidia: The 50 series ‘Blackwell’ GPUs.

Much of the information remains speculative, since it relies on leaks provided by industry insiders Kopite7Kimi and Moore’s Law is Dead. Though these two have had a relatively good accuracy rate in the past. But, until Team Green officially releases the specifications of the new GPUs, all of this must be taken with a pinch of salt.

RTX 50 series specifications speculation

The Blackwell architecture has been officially announced by Nvidia at GTC 2024. Though the chips are not the same as commercial gaming GPUs, the specs do at least provide some helpful information. When combined with the latest leaks from Kopite7Kimi, it’s suggested that the flagship gaming GPU chip, the GB202, will use the same process node as the AI-focussed GB101 chip.

Nvidia

It is also expected that Nvidia will offer significant advances in performance, with the flagship GB202 gaming chip having considerable improvements compared to the Ada Lovelace AD102 chips. In particular, the L1 cache will see some large gains in performance.

Kopite7Kimi indicated (via WCCFTech) that the chip would offer 12 GPCs, with 8 TPCs each, providing a total of 96 TPCs. Using the Ada Lovelace structure as a template would give us 192 SMs, which equates to a massive 24,567 CUDA cores. That is an increase of 33% over the Ada equivalent AD102 chip.

GPU chip	GB202 (Blackwell)	AD102 (Ada Lovelace)
CUDA cores	24,567	18,432
Memory cap	48 GB	24 GB
Max memory bus	512-bit	384-bit
Memory technology	GDDR7	GDDR6X

Earlier leaks from Kopite7Kimi pointed towards Nvidia reusing the 384-bit memory bus from the Ada chips for the Blackwell chips, instead of upgrading to 512-bit. However, more recent information says Nvidia will be upgrading to a 512-bit memory bus, at least on its flagship chips.

According to Moore’s Law is Dead, the uplift in performance from the 40 series to the 50 series will not be as impressive as other generations.

Blackwell GPU die breakdown speculation

Die	Potential GPU
GB202	RTX 5090
GB203	RTX 5080
GB205	RTX 5070 Ti / RTX 5070
GB206	RTX 5060 Ti / RTX 5060 / RTX 5050 Ti
GB207	RTX 5050 / RTX 5030

The chips are expected to be cut down into five different dies: GB202, GB203, GB205, GB206, and GB207. The GB202 is the flagship, and the GB207 is reserved for the budget end of the range. Curiously, the GB204 is missing from the lineup. Historically, 204 chips have powered the popular 70-suffix cards.

According to Kopite7Kimi, the GB203 chip will be ‘half of GB202’, which would match the difference in performance between the RTX 4090 and the RTX 4080. Based on this, and following the trends set by the 40 series.

It is unknown if Nvidia plans to keep the monolithic structure for its next generation of gaming GPUs or utilize the MCM or multi-chip-module structure. AMD is reportedly already experimenting with chiplet configurations with its Radeon graphics cards, and many analysts expect that this is the direction GPU development needs to head to maintain improvements in successive generations.

As usual, until the specifications are officially revealed, all this needs to be taken with a healthy dose of skepticism.

RTX 50 series features speculation

Thanks to information once again courtesy of persistent leaker Kopite7Kimi, we can expect that the RTX 50 series of graphics cards will feature support for Display Port 2.1, along with HDMI 2.1. An update compared to the Ada Lovelace series, which did not feature DisplayPort 2.1 support.

The power connector has proved to be something of a contentious subject. After reports of the 12VHPWR connector found on the RTX 4090 and the 40 Super cards melting under some circumstances, many consumers were hoping a change was forthcoming from Team Green.

Moore’s Law Is Dead recently revealed that Nvidia plans to switch to an entirely new connector type for its next-generation cards. The leak suggests that Nvidia is working on an all-new 16-pin connector, but this has not yet been substantiated by other sources.

If Nvidia sticks with the 12VHPWR connector then the power draw would be limited to 600W, though rumors that Nvidia plans on touting the efficiency of its new cards suggest it will be considerably under this limit.

RTX 50 series release date speculation

Once again, leaks courtesy of Moore’s Law is Dead suggest that the first card in the RTX 50 series could be launched as soon as Q4 of 2024. If these rumours are accurate the first of the Blackwell gaming GPUs to see retail will be the RTX 5090.

The unnamed source told Moore’s Law is Dead that the launch date will depend on the sales of the existing 40 series cards, particularly the refreshed 40 Super cards. The source goes on to state that Nvidia plans to promote the improved efficiency of the 50 series at CES 2025, with a launch to coincide.

RTX 50 series price rumours

Very little information is available on even tentative pricing for the Blackwell 50 series at this point. The best estimate for pricing is that the RTX 5090 will retail with an MSRP of $1699 at launch.

Nvidia is said to have spent an eye-watering $10 billion on developing the Blackwell platform, though it is expected that the company will attempt to recoup much of that cost through sales of its dedicated AI chips.

It does leave many analysts and consumers worried about what price structure Team Green will use for the 50 series. Especially since high demand sent the RTX 4090 much higher than its launch price of $1,600 and far into the $2,000 range.

However, the 40 Super refresh cards adopted a fairly aggressive price strategy in an effort to keep pace with AMD. Only time will tell which stance Nvidia will adopt.