For years, AMD and Nvidia have traded blows as the sole manufacturers of consumer-grade desktop graphics cards. Although Intel's Arc series brought Team Blue to the fray, the Nvidia GeForce RTX 4000 series dominates our best GPU lists, and for good reason. Not only do the Ada Lovelace GPUs bring amazing ray-tracing capabilities to the table, but you also get to enjoy smoother frame rates thanks to a combination of DLSS 3.5 and Nvidia's proprietary frame generation technology.
Related
Which GPU should you buy in each $100 price bracket?
There's something here for everyone, regardless of whether you're a budget user or a gaming enthusiast
1
What's more, Team Green's upcoming series is expected to be even better than the current generation. Named “Blackwell” after the Mathematician David Blackwell, the new graphics cards from Nvidia could bring high-bandwidth GDDR7 memory, support for PCIe 5.0, and several other features to the table. In this article, we’ll recap all the rumors and leaks about the GeForce RTX 5000 family.
Nvidia GeForce RTX 5000 series: When we're hoping to see it and how much we want to pay
Nvidia tends to follow a 2-year release cycle for consumer graphics cards, with a refresh of the present-gen hardware around the one-year mark. Since Team Green debuted the Super variants of its current RTX 4000 GPUs in January 2024, it was believed that the Blackwell series will be unveiled by late 2024, though all signs now point to a CES 2025 announcement instead.
Earlier, HardwareLUXX, a German tech website, reported that we won’t see the “Ada Lovelace-Next” graphics cards before 2025. On the other hand, Youtuber Moore’s Law Is Deadclaimed that the newer cards could drop as early as Q4 2024. In an older X (previously called Twitter) thread, leaker kopite7kimi speculated that the new GPUs may be released "at the end of the year," before they noted that due to sanctions, the new GPUs could be pushed to 2025. Regardless, the leaker also stated that the RTX 5080 (code-named GB203) will be launched before the top-of-the-line RTX 5090.
Meanwhile, the pricing for the new graphics cards is anybody’s guess. The RTX series is notorious for being overpriced, and the exorbitant pricing trends are expected to continue with the Blackwell family. Assuming we’ll have an RTX 5090 instead of a 5090 Ti as the flagship GPU, it should bear the same (or slightly higher) launch price of $1,599 as the current-gen GeForce RTX 4090.
Specs | RTX 5090 | RTX 4090 |
---|---|---|
Code-name | GB202 | AD102-300 |
Interface | PCIe 5.0 x16 | PCIe 4.0 x16 |
Memory | 36GB or 24GB GDDR7 | 24GB GDDR6X |
CUDA cores | 24,576 | 16,384 |
L2 cache | 128MB | 72MB |
Boost clock | 2.9GHz | 2.52GHz |
Architecture | Blackwell | Ada Lovelace |
Memory bus width | 512-bit | 384-bit |
Memory bandwidth | 1,536 GB/s | 1,008 GB/s |
Nvidia GeForce RTX 5000 series: What we could see
Higher bandwidth GDDR7 memory on most GPUs
Many GPU enthusiasts were disappointed with the Ada Lovelace family because Nvidia decided to bring over the GDDR6X and GDDR6 memory from older GPUs. Luckily, the rumor mill claims that the flagship graphics card(s) from the GeForce RTX 5000 series will leverage the latest GDDR7 memory. For reference, the GeForce RTX 4090 had a memory bandwidth of 1,008 GB/s, which is the same as that of the last-gen GeForce RTX 3090 Ti. In contrast, the memory bandwidth of the RTX 5090 is rumored to be 1,536 GB/s, which is a staggering 52% upgrade!
There have also been plenty of rumors about the GeForce 5090's memory bus. After many contradictory statements about the RTX 5090, kopite7kimi revealed that crème-de-la-crème GPU of the Blackwell family would feature a 512-bit memory bus. In June, they released a chart containing the details about the memory bus on each GPU in the upcoming family. Interestingly, the RTX 5050 (GB207) is the only graphics card in the series to possess the outdated GDDR6 video memory.
While we're on the subject, aside from a few rumors claiming the RTX 5090 will have 36GB of GDDR7 memory, there haven't been too many updates on the VRAM size of the GPU. Nevertheless, the GPU is bound to have at least 24GB of VRAM, like the last-gen RTX 4090. As for the other high-end model, kopite7kimi earlier cited that the GB203 card may possess half as much memory as the GB202 variant. Assuming it's not a "multi-chip package," this means a theoretical RTX 5080 will feature anywhere between 12 to 18GB of memory.
Higher TDP alongside 16-pin connectors
Earlier in July, PSU manufacturer Seasonic accidentally added the TDP requirements of the next-gen Nvidia GPUs to its online wattage calculator. FIrst discovered by Twitter user Everest (@Olrak29_), the leaked specs were shortly taken down by Seasonic, but not before news outlets like VideoCardZ tracked the power consumption of all next-gen GPUs.
GPU | Code-name | TDP |
---|---|---|
RTX 5090 | GB202 | 500W (50W more than the RTX 4090) |
RTX 5080 | GB203 | 350W (30W more than the RTX 4080) |
RTX 5070 | GB205 | 220W (20W more than the RTX 4070) |
RTX 5060 | GB206 | 170W (55W more than the RTX 4060) |
RTX 5050 | GB207 | 100W |
Clearly, the Blackwell family has some power-hungry graphics cards, with the RTX 5090 requiring an additional 50W more than its last gen counterpart. While the RTX 5080 and 5070 surpass their predecessor's power consumption by 30W and 20W respectively, the RTX 5060 is quite the behemoth. With a massive TDP of 170W, the RTX 5060 needs 55W more than the RTX 4060 and 5W more than the 16GB variant of the RTX 4060 Ti. What's more, all GPUs will reportedly feature 16-pin connectors, which is quite surprising, considering there's even a RTX 5050 in the mix. Regardless, you might need to upgrade your power supply if you want to outfit your PC with the upcoming Team Green GPUs.
Potential MCM design
Unlike AMD, Nvidia has stuck to monolithic dies for its graphics cards. A monolithic chip has all the circuitry on the same die, which makes it difficult to shrink the transistors without increasing the overall chip size. In contrast, a Multi-Chip Module (MCM) approach involves adding separate chips (called chiplets) connected via interposers onto a single substrate. This increases the scalability of MCM-based chips and allows manufacturers to squeeze in better specs.
According to leaker kopite7kimi, Nvidia’s GB100 and GB102 data center chips will feature an MCM design instead of the traditional monolithic architecture. Likewise, the high-end GB202 chip is expected to bring the MCM architecture to consumer GPUs. However, this new design probably won’t be included in the GB203, GB205, GB206, and GB207 chips as Nvidia will go with the same old monolithic architecture for its budget and mid-range GPUs.
New TSMC 4N manufacturing process (with a higher core count)
Nvidia has continued to utilize newer manufacturing processes over the last couple of GPU generations, and judging by the rumors, the Blackwell series will be no different. Team Green’s upcoming GPU family will be based on the TSMC 4N process node, based on the TSMC 5 manufacturing process. Recently, kopite7kimi also confimred that the flagship GB202 chip will have the same manufacturing process as GB100, the chip meant for data center GPUs. However, the leaker noted that there would be a 30% increase in chip density of the GB202 SKU.
In an older tweet, kopite7kimi revealed that there won’t be a significant increase in the number of Graphics Processing Clusters (GPCs) and Texture Processor Clusters (TPCs) on the new GPUs. Additionally, panzerlied claimed that besides a 15% increase in frequency and 78% more cache, the GeForce RTX 5090 could see a 50% improvement in scale, which could translate to a higher 24,576 CUDA core count.
Support for PCIe 5.0 and DisplayPort 2.1
Although the Ada Lovelace GPUs will work with PCIe 5.0 motherboards, they aren’t able to fully utilize the high-speed interface standard. However, the Blackwell GPUs are expected to be able to leverage the lightning-fast data transfer speeds provided by PCI Express Gen 5.0.
Likewise, the GeForce RTX 5000 graphics cards are rumored to be a step up from their predecessors in terms of display capabilities. According to kopite7kimi, the new GPUs will be compatible with the DisplayPort 2.1 standard, which supports much higher bandwidth, refresh rate, and resolution than the last-gen DisplayPort 1.4.
Nvidia GeForce RTX 5000 series rumors: To believe or not to believe
Although many of these leaks were disclosed by trustworthy sources, you should take them with a grain of salt. Not only have the rumors surrounding these GPUs changed time and again, but many of the leaks surrounding older Nvidia graphics cards were refuted soon after the products made their way to the market. For example, there was quite an uproar from the community when the flagship RTX 4090 from the Ada Lovelace family was believed to have a max TDP of 800W. Thankfully, the card debuted with a 450W TDP instead.
While there are plenty of reasons to believe that Nvidia will debut the Blackwell family in 2025, the company has kept its lips sealed for now. Either way, we're bound to hear more about the upcoming GPU family before long. But if you're unable to wait for the Blackwell series, the RTX 40-series GPUs are still worth looking into. In particular, the GeForce RTX 4070 remains our favorite GPU that provides solid performance without costing a fortune.
Related
MSI GeForce RTX 4070 Ventus 3X review: Great value 1440p gaming
An affordable RTX 4070 performance with impressive 1440p results
- GPU
- Nvidia
- GPU
Your changes have been saved
Email is sent
Email has already been sent
Please verify your email address.
You’ve reached your account maximum for followed topics.
Manage Your List
Follow
Followed
Follow with Notifications
Follow
Unfollow
Readers like you help support XDA. When you make a purchase using links on our site, we may earn an affiliate commission. Read More.