back to top
Tuesday, April 23, 2024
Cryptonoyz, crypto news
HomePC HardwareNVIDIA RTX 5000: all the information about Blackwell GPUs

NVIDIA RTX 5000: all the information about Blackwell GPUs

If you hesitate between buying an RTX 40 or waiting for RTX 5000, we tell you everything about the graphics cards that NVIDIA I would like to launch in 2024.

After Ampere and Ada Lovelace, it is Blackwell’s turn as graphic architecture that could be a revolution. At the doorstep of the NVIDIA RTX 40 Super, we must say that there is already information about the RTX 5000 and its possible improvements compared to the RTX 4000. Let’s compile it as God intended, and see what is expected.

Everything about the NVIDIA RTX 50: chronological order

We searched the newspaper archive to talk about the next NVIDIA GPUs and organize all the information that has swarmed Professional Review.

Ambiguity in performance, how big will the jump be?

The first news about the RTX 50 was published by Gustavo Gamarra in PR, echoing a very reliable source: Red Gaming Tech. At this point, it was stated that the chiplet design was on the table and NVIDIA reserved an ace up its sleeve (a monster GPU) in case RDNA 3 did damage.

With the MCM architecture at the forefront, it seemed that they had redesigned the structure of the Stream Multiprocessor, as well as the bus links that interconnect the SMs. Although they thought the RTX 50 would bring a manual denoiser accelerator part to Ray Tracing, this feature was released with the latest DLSS 3 update: Ray Reconstruction.

However, Blackwell was introduced at the end of March 2024 with the motto of “the engine of the new industrial revolution.” In fact, they combined Blackwell with Grace to create the GB200 Superchip. Jensen stated that Blackwell is not a chip, but the name of a platform, and took the opportunity to compare the new chips against Hopper (the professional GPUs from 2 years ago).

nvidia rtx 5000

At the moment, NVIDIA has not given hardware figures at Blackwell so that we can get an idea of ​​what may be coming for GeForce.

The NVIDIA RTX 5000 will use the TSMC 4NP (5 nm) node

nvidia blackwell

NVIDIA seems to do the same in each generation, initially hesitating whether to opt for MCM or a monolithic design. There was consensus between Moore’s Law is Dead, TechTubers and Red Gaming Tech about the 3nm process in the NVIDIA RTX 50, but the latest information points to a 5nm process.

The GB202 would be the reference RTX 5090, and would share the same node as the B100, TSMC’s 4NP which is a 5 nm variant. According to what Kopite stated on his Twitter, it could increase the density of its transistors by 30%, but something doesn’t add up here.

Additionally, saying that there was already talk about the GPUs of each model, being the flagship of GeForce GB102: 144 SM, 384-bit, GDDR7, 96 MB L2 and PCIe 5.0 x16 support. For those for whom this data sounds “Chinese”, the RTX 4090 comes with 128 SM, 72 MB L2 and 384-bit memory bus. Everything indicates that the flagship of the RTX 5000 would be the RTX 5090.

In summer, they would change from GB102 to GB202. Here see the entire news about RTX 50 Blackwell.

The newspaper DigiTimes warned that the 3nm node by TSMC could not arrive until 2025, and this information for me is true. The RTX 50 with 3 nm depended on the Taiwanese manufacturer, and the news emerged in April 2023, so there was room for maneuver.

Even so, tell you that TSMC does not appear to manufacture 4nm node, while Samsung does. The Taiwanese range from 5 nm to 3 nm, so I can tell you that this information is not entirely reliable, despite coming from DigiTimes!

There could be an oscillation in the business focus, prioritizing the AI, supercomputers and HPC segment; In fact, there were leakers who said that we could not rule out the 4nm process.

The RTX 5090 would increase CUDA Cores by 50%

nvidia rtx 50

You already know that the CUDA Cores They are the general-purpose cores that NVIDIA has in its graphics cards, popularly called “shaders.” Well, information from Chiphell would affirm several data about the NVIDIA RTX 5090:

  • +50% CUDA cores.
  • +52% memory bandwidth (1.5 TB/s).
  • +78% increase in L2 cache memory (128 MB).
  • +15% increase in clock speeds (boost).
  • +70 approximate performance.

Although it came from Chiphell, Twitter leakers approved this information, so Blackwell was beginning to get his point. Until now, all the information fits, except increasing L2 cachepreviously having 96 MB L2 by Red Gaming Tech.

You will see variations between now and the end of 2024 because there can be many changes of plans. Of course, whatever comes out in the summer of next year… is going to go to mass.

Up to 24,576 CUDA cores in the most powerful RTX 50

rtx 50

Called GB202 or RTX 5090, a new information from @kopite7kimi claimed that we would see around 24,576 CUDA cores in Blackwell’s most powerful model. We came from an upgrade to 144 SM, but this new information would point to 192 SMwhose count would be equivalent to the mentioned CUDA Cores.

Tell you that the RTX 4090 comes with 16,384 CUDA Cores, so we are talking about multiply that figure by x1.5. There was also talk of 512-bit bus, a figure that has been repeated throughout 2023.

New Tensor Cores and RT Cores along with AI accelerators

nvidia rtx 5000

Yes, you conceived of Tensor Cores as those “AI accelerators”, but NVIDIA may segment its cores even more. New accelerators for data processing have been introduced at Blackwell, so we could see cores focused on exclusive AI tasks, and I’m not talking about DLSS.

There is no information about the RT Cores yet, but rest assured that, if they present new Tensor Cores, they will do the same with the cores for Ray Tracing.

NVIDIA RTX 5090 would reach 3 GHz and with a 512-bit bus

nvidia rtx 5090

These numbers could be given taking into account that they are increasing the frequency more and more (look at the RTX 4000 vs 3000) and that GDDR7 memory would make that 512-bit bus viable. In addition, the information that I myself published about the NVIDIA RTX 5090 in May 2023 was related to Samsung confirming GDDR7 memories.

He Boost Clock of the RTX 4090 can reach 2520 MHzso we are talking about an additional 400-500 MHz on the RTX 5090.

GDDR7 memory in the NVIDIA RTX 5000

samsung gddr7

This news occurred in March 2023, generating a quarter jump between one leak and another. It was also published by Gustavo, and included AMD’s RTX 50 and RX 8000: GDDR7 for all.

This way, GDDR6X memory is abandoned (or so it seems) to give way to GDDR7 with 36 Gbps speed as a starter in theory, but In practice we would see 28 or 32 Gbps. This upgrade is necessary, and the truth is that we can say that Samsung and Micron would have the memories ready at the end of 2024. That is, the launch of the GDDR7 memories coincides with the supposed launch of Blackwell and RDNA 4.

And finishing with the technical data, we could see bandwidths that exceed 1TB/s and memory buses up to 512 bit. The memories that we see inside each RTX 5000 GPU have all the earmarks of being from Samsung, as noted in a report by Industry Growth Insights.

In addition, 2 GB modules are planned as has been done, while 3 GB modules are on the roadmap.

Lots of news in AI

nvidia b100

This has not been discussed until the end of 2023, and NVIDIA is going to implement many improvements in artificial intelligence for its next RTX 50 with Blackwell architecture. They are seeing the business it entails, and DLSS 3 It is more than refutable proof: we must invest in artificial intelligence.

Therefore, not only is it possible that we will see an acceleration due to on-chip AI, but DLSS 4 It could be a blast after seeing the potential of version 3.0. All this was known from the information on the NVIDIA B100.

Wait until 2025?

nvidia rtx 50 launch

Information from HardwareLuxx indicated that there would be no NVIDIA RTX 50 until 2025, illustrating this information with a roadmap leaked by NVIDIA. In this image we saw «Ada Lovelace-Next“, where reference is clearly made to the RTX 5000 and Blackwell architecture.

If you want to know the launch of the NVIDIA RTX 5000, go to the end of the post where we give you our “truncheon”.

Launch of the NVIDIA RTX 5000

As we have stated in Professional Review, the NVIDIA RTX 5000 release date appears to be linear with previous GeForce families:

  • RTX 5080 and 5090 at the end of 2024.
  • RTX 5070, 5060 and others from the 1st quarter of 2025.

This is neither official, nor is it information that comes from Asia or from the TSMC factory itself. We rely on our experience and NVIDIA’s modus operandi in its presentation policy.

We recommend the best graphics cards on the market

Do you think they will arrive next year?

Mark Tyson
Mark Tyson
Freelance News Writer. Always interested in the way in which technology can change people's lives, and that is why I also advise individuals and companies when it comes to adopting all the advances in Apple devices and services.


Please enter your comment!
Please enter your name here