• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

2027/2028 Nvidia/AMD/Intel GPU speculation

M

Marees

Supreme [H]ardness
Joined
Sep 28, 2018
Messages
4,587
lets start

AMD RDNA 5 GPUs may launch after Nvidia's RTX 60 series in late 2027​

The next GPU cycle is slipping further away​


https://www.techspot.com/news/110882-amd-rdna-5-gpus-may-launch-after-nvidia.html


AMD’s Next-Gen RDNA 5 Gaming GPUs Might Launch After NVIDIA’s RTX 60 Series In 2027​


https://wccftech.com/amd-next-gen-r...ht-launch-after-nvidia-rtx-60-series-in-2027/
The memory shortage is being blamed for the rumored timeline of Nvidia's RTX 6000 cards launch. If, as many claim, the cards get here in the second half of 2027, it will mean a 30-month gap between generations – longer than any modern GPU-series release.

We used to have Nvidia's RTX Super line to hold gamers over until the next-gen architectures arrived, but the memory crisis means those cards have been delayed indefinitely. From Team Green's point of view, not only are memory costs too high to warrant a Super-series release, but there's going to be no direct competition from AMD until 2027, which leads us to the latest rumor.

According to Kepler_L2, a long-time leaker with a good track record, AMD will launch its RDNA 5-based consumer graphics cards after Nvidia's RTX 6000 in H2 2027.
Click to expand...

The reason behind AMD launching after NVIDIA is said to be NVIDIA's margins, and that despite AMD's best pricing strategy to counter their launch, NVIDIA can get away with a price drop. So, say AMD was to launch first at highly competitive prices, NVIDIA could launch their lineup at the same or even undercut a few products in the mainstream / high-end category and get away by taking the flagship crown, which it has been doing so since the GTX 980 Ti, leaving AMD with no desire to compete in the $1000 US+ segment.

RDNA5 has been 2027 for a while, and they can't launch before NVIDIA anyway.
NVIDIA margins are massive and they can counter almost anything with a price drop. What AMD needs is something that is in a difference performance bracket altogether, like the 9800X3D is.
via Kepler_L2 (Anandtech Forums)
Click to expand...
If AMD launches after NVIDIA, then they have an opportunity to tackle NVIDIA's pricing since the green time isn't known for changing prices after their products have reached retail. They just leave it for the market to decide. There have been a few cases where older GPUs have seen a price drop, but those are only after a year or when a new refresh is expected.

So, looking at the release time, AMD launching after NVIDIA would mean that RDNA 5 gaming GPUs also won't land until late 2027. And that means a long wait, considering AMD's first RDNA 4 GPU launched on 6th March (RX 9070 XT), and a 2H 2027 launch would indicate a 2.5 years' worth of wait or over 800 days. For NVIDIA, it's an even longer wait till the Rubin RTX 60 series since the first RTX 50 "Blackwell" graphics card launched on 30th January 2025.

h/t Daniel Owen
7:28 RDNA 5 release date leak/rumor analysis
 
just found an nvidia thread

pendragon1 said:
yup, already. rumors though...

NVIDIA's next-gen GeForce RTX 60 rumors begin: Rubin GPU, more VRAM, DLSS 5, TSMC 3nm node
PC gamers can't even buy NVIDIA's new GeForce RTX 5090 or RTX 5080, and now we've got the hype train for next-gen RTX 60 series GPUs.
https://www.tweaktown.com/news/1029...gpu-more-vram-dlss-5-tsmc-3nm-node/index.html

2026 Nvidia GPU hardware might launch six months early, but don’t panic yet​

https://www.pcgamesn.com/nvidia/rubin-gpu-early-release

https://wccftech.com/nvidia-next-gen-rubin-ai-architecture-now-rumored-to-launch-by-h2-2025/
Click to expand...
 
somewhat old news

Intel "Xe4" and AMD "GFX13" codenames surface for next-gen 'Druid' GPUs​

News
By Hassam Nasir published May 13, 2025
Both companies are already laying the groundwork for their future GPU releases.


The internal codenames for these GPUs, Intel's Xe4 and AMD's GFX13, have been spotted by Kepler and x86isdeadandback at X, as noted by VideoCardz. While this does not allude to the specifications, it shows that both teams are actively pouring resources into the development of their future graphics products.

Intel officially confirmed its fourth-generation Arc GPUs would be codenamed Druid, employing the Xe4 architecture, a while back.
Tom Peterson's comments support this, ... that the hardware teams have moved on to the next project, Druid (Xe4). Based on commits to the Dawn repository, developers are starting to integrate support for Xe4, which should fall under Intel's Gen15 umbrella.

https://www.tomshardware.com/pc-com...x13-codenames-surface-for-next-gen-druid-gpus
 
work graphs can be expected to be significantly important for all next gen GPUs

Lakados said:
https://overclock3d.net/news/gpu-di...mance-gains-using-work-graphs-with-radeon-gpu
https://developer.nvidia.com/blog/advancing-gpu-driven-rendering-with-work-graphs-in-direct3d-12/
https://developer.nvidia.com/blog/work-graphs-in-direct3d-12-a-case-study-of-deferred-shading/
https://github.com/NVIDIAGameWorks/donut_examples/tree/main/examples/work_graphs
https://overclock3d.net/news/softwa...ure-could-make-future-games-less-cpu-limited/

With the proliferation of GPU-driven rendering techniques – such as Nanite in Unreal Engine 5 – the role of the CPU is trending towards primarily resource management and hazard tracking, with only a fraction of time spent generating GPU commands. Prior to D3D12 Work Graphs, it was difficult to perform fine-grained memory management on the GPU, which meant it was practically impossible to support algorithms with dynamic work expansion. Even simple long chains of sequential compute work could result in a significant synchronization and memory overhead.

GPU-driven rendering was accomplished by the CPU having to guess what temporary allocations were needed by the GPU, often over-allocating to the worst case, and using previous frame readback for refinement. Any workloads with dynamic expansion either meant issuing worst case dispatches from the CPU, having the GPU early out of unnecessary work, or non-portable techniques were used, like persistent threads.

With Work Graphs, complex pipelines that are highly variable in terms of overall “shape” can now run efficiently on the GPU, with the scheduler taking care of synchronization and data flow. This is especially important for producer-consumer pipelines, which are very common in rendering algorithms. The programming model also becomes significantly simpler for developers, as complex resource and barrier management code is moved from the application into the Work Graph runtime.

We have been advocating for something like this for a number of years, and it is very exciting to finally see the release of Work Graphs.

DX12 with the latest Agility update supports them but it will be a good while before games start using them as they have to be built with this rendering method in mind.
Click to expand...
 

Microsoft Unveils DirectX Raytracing 1.2 With Huge Performance & Visual Improvements, Next-Gen Neural Rendering, Partnerships With NVIDIA, AMD & Intel​

DirectX Raytracing 1.2 (DXR) & Neural Rendering Announced at GDC, Support Across AMD, Intel & NVIDIA Hardware​


Starting with DXR 1.2 (DirectX Raytracing), Microsoft is introducing two new technologies which include OMM (opacity micromaps) and SER (shader execution reordering), which will help deliver huge performance boosts.

https://wccftech.com/microsoft-dire...s-next-gen-neural-rendering-nvidia-amd-intel/

Another major aspect that is being introduced by Microsoft is support for cooperative vectors, which is a brand-new programming feature coming to Shader Model 6.9, very soon. With these cooperative vectors, developers can leverage new hardware acceleration engines for vector and matrix operations, allowing them to integrate neural rendering techniques directly within the graphics pipeline

All major hardware vendors, including NVIDIA, AMD, & Intel, are on board the Cooperative Vectors and Neural Rendering train:

 
mostly based on pure speculation/feelings.

If Rubin desktop is that much later than Rubin R100/200, maybe that like A100 vs Ampere they will not be on the same node, could see Rubin desktop going on TSMC 2, keep Rubin on TSMC 3 and have the Rubin 300 ultra or its datacenter successor skip 2N and go directly to A16 after that, so they do not compete correctly for wafer.

a 2 nodes gain in a single gen would make the longer time look natural and could make an somewhat impressive gen gain easier with less effort to put in them.

They apparently and engineer working for month pre-acquisition with the grog teams, it could be really tempting to have on desktop gpu have sram AI decoding

-DLSS is a particularly perfect use case for this, could make it near instant, very low power heat and let the gpu/gpu vram unnafected, frame generation, part of denoising/ray reconstruction maybe has well
- Using old node (lot of TSMC 10-14 which is really good for this and eventually a giant amount of TSMC 4 that was doing Lovelace-Blackwell for them, will become still great node that Rubin does not necessarily use for logic-compute chips but for the sram one would be great one
- packaging has got better and by 2027/2028 "old" packaging tech will be the greatest one of 2024 that they used for spark and B200
- leaving the vram alone for some stuff could get more interesting has well
- and maybe the most important use case, vram compression/ decompression just on use to save space could become a very big deal for amd-nvidia and sram decoding for an near instant / low power decompression part would be great.
 
Last edited:

20+ Ray Tracing, Scheduling, and other AMD patent grants and filings, hinting at future GPUs and gaming technologies​


https://www.reddit.com/r/GamingLeak...&utm_medium=post_embed&utm_content=action_bar

https://www.neogaf.com/threads/20-r...-future-gpus-and-gaming-technologies.1692390/



  • AMD has filed and been granted 20+ patents focused on next-generation GPU technologies, strongly emphasizing ray tracing performance, efficiency, and smarter workload scheduling.
    • These patents suggest major architectural upgrades aimed at future RDNA or UDNA GPUs and possibly next-gen consoles.
  • Ray tracing and geometry improvements
    • Advanced ray tracing techniques including payload sorting, deferred any-hit shaders, and fine-grained context saving.
    • New geometry systems like Dense Geometry Format, Displaced Micro Meshes, and improved tessellation for higher visual detail at lower cost.
  • BVH and intersection optimization
    • Multiple patents target faster ray traversal using improved BVH structures.
    • Techniques include mass culling of BVH nodes, low-precision or integer-based intersection testing, and pre-filtering to reduce expensive calculations.
  • Smarter GPU scheduling
    • Heavy focus on Work Graphs–centric scheduling.
    • GPUs gain more autonomous control over task dispatch, load balancing, and shader execution, reducing CPU overhead.
Click to expand...



#1 Patent list - Ray tracing/RT, execution and payload sorting

  1. https://patentscope.wipo.int/search/en/detail.jsf?docId=US464433976 (Fine grained context saves)
  2. https://patentscope.wipo.int/search/en/detail.jsf?docId=US464434024 (Shader Engine HW payload coalescer)
  3. https://patentscope.wipo.int/search/en/detail.jsf?docId=US464434039 (Enhanced resource barriers)
  4. https://patentscope.wipo.int/search/en/detail.jsf?docId=US464435335 (Animated curved surface patches)
  5. https://patentscope.wipo.int/search/en/detail.jsf?docId=US464435300 (Deferred any hit shaders)
  6. https://patentscope.wipo.int/search/en/detail.jsf?docId=US464101855 (High quality animated Displaced Micro Meshes/DMMs)
Sourced from

#2 Patent list - BVH related

  1. https://patentscope.wipo.int/search/en/detail.jsf?docId=US470469751 (Interpolated geometry + Dense geometry format/DGF)
  2. https://patentscope.wipo.int/search/en/detail.jsf?docId=US470603545 (DGF + DMMs/subdivision surfaces)
  3. https://patentscope.wipo.int/search/en/detail.jsf?docId=US470603544 (DGF without indices (implied))
  4. https://patentscope.wipo.int/search/en/detail.jsf?docId=US470469793 (Dual use BVH for RT and collision detection)
  5. https://patentscope.wipo.int/search/en/detail.jsf?docId=US470469794 (Mass culling/discard of BVH nodes)
Sourced from

#3 Patent list - DGF + OBBs + Pre-filtering (low precision/INT-based parallel intersection testing)

  1. https://www.patents-review.com/a/20250111586-pre-filtering-nodes-bounding-volume-hierarchy.html (DGF BVH fallback + Pre-filtering)
  2. *https://www.patents-review.com/a/20250131640-intersection-testing-dense-geometry-data-triangle.html (DGF + Pre-filtering)
  3. *https://www.patents-review.com/a/20250131639-dense-geometry-format.html (DGF description)
  4. https://www.patents-review.com/a/20...w-precision-ray-intersection-accelerated.html (Quantized BVH data for Pre-filtering)
  5. https://www.patents-review.com/a/20250209723-system-method-low-precision-ray-tests.html (Pre-filtering)
  6. https://www.patents-review.com/a/20...s-oriented-bounding-boxes-based-platonic.html (Quantized/INT-based OBBs)
Sourced from

* = #2 and #3 mentioned previously by Kepler_L2 here

#4 Patent list - Work Graphs-centric scheduling

  1. https://patents.google.com/patent/US20240111574A1 (Shader Engine autonomous scheduling (WGS) and dispatch (ADC))
  2. https://patents.google.com/patent/US12153957B2 (Hierarchical work scheduling and load balancing)
  3. https://patents.google.com/patent/US20240111575A1 (WGS to Command Processor communication scheme)
  4. https://patents.google.com/patent/US20240202003A1 (Fixed-function/FF + Work Graphs)
  5. https://patents.google.com/patent/US12436767B2 (Two FF pipeline modes)
  6. https://patents.google.com/patent/US20250217195A1 (WGP-level autonomous scheduling and dispatch)
Sourced from
 
new video by RGT dropped on RDNA 5

in short a GPU with die size that is extremely competitive because of reduced L2 cache & removal of infinity cache.

in other words, ideal for consoles. this thing is built for consoles

shared L1 cache eliminates data replication across GPU cores, improving collective hit rate & efficiency at the cost of increased latency for certain workloads (addressed by workgraphs & others), & die area.
Basically, net improvement in peak throughput.


View: https://youtube.com/watch?v=WBKKrh2DmAA

https://x.com/MrMPFR/status/2015553754884362286?s=20

https://x.com/MrMPFR/status/2015571566130614629?s=20
 
while i have a lot of confidence in the upcoming monolithic amd Xbox soc and its most direct equivalent in the discrete gpu scene (could be a second RDNA 2 moment with the amount of effort they are probably putting in a new generation of console gpus), the big version with a dual compute die (GMD) I imagine it is a bit rolling the dice if it will work well for gaming.

Gideon said:
My speculation, is no one will care because prices will be sky high and availability will be poor unless the AI bubble bursts.
Click to expand...
Will see on that one, my current speculation is that they will be on accessible Playstation 6 (for rdna 5 gpus at least), AMD packaging is not the recent fancy one usually used in datacenter, PS6 went small-monolitic and seem thinking about pricing a lot, my guess would be people caring enough for PS6 physical wait line and F5/bot race at the launch....

Could be hard for Intel to have sky high gpu prices has well
 
LukeTbk said:
Will see on that one, my current speculation is that they will be on accessible Playstation 6 (for rdna 5 gpus at least), AMD packaging is not the recent fancy one usually used in datacenter, PS6 went small-monolitic and seem thinking about pricing a lot, my guess would be people caring enough for PS6 physical wait line and F5/bot race at the launch....

Could be hard for Intel to have sky high gpu prices has well
Click to expand...

Intel would have to build something that at least was as fast or faster than the current generation out now for most to care, that would be the challenge for Intel.
 
Not a speculation but a yet to be confirmed fact: 2027/2028 GPUs will be severely overpriced and have way too little memory with abysmal actual availability.
 
XoR_ said:
Not a speculation but a yet to be confirmed fact: 2027/2028 GPUs will be severely overpriced and have way too little memory with abysmal actual availability.
Click to expand...
overpiced & delayed yes

but I expect memory to be sufficient because of new gddr7 tech for premium end & also lpddr5x/lpddr6 at the budget end
 
Marees said:
overpiced & delayed yes

but I expect memory to be sufficient because of new gddr7 tech for premium end & also lpddr5x/lpddr6 at the budget end
Click to expand...
aren't we forgetting something important about RAM?
our future AI overlords can gobble any amount and faster than you can say "oh flock"
 
With neural rendering and super ML texture compression, all you need will be 4gb cards :D Just older games will suck. Well games now, this year and most next and before to clarify.
 
noko said:
With neural rendering and super ML texture compression, all you need will be 4gb cards :D Just older games will suck. Well games now, this year and most next and before to clarify.
Click to expand...
To be honest today's memory requirements come mostly because of bad development practices.
We were supposed to have texture streaming to fetch closeup textures just-in-time and not kept in the GPU all the time...
...and instead we have issue like shader compilation because devs are too lazy to write short function to recompile all shaders before launching actual game.
We have shaders optimized by tons of if/else statement instead of branchless programming - which could avoid need for TAA completely.
We have TAA which is terribly tuned for both visuals and performance.
We have terrible overuse of various frame and other buffers which do way too little in single pass/buffer wasting tons of memory and memory bandwidth.
We have most games use bad lighting models which don't even run faster than more realistic models.

ML texture compression won't really solve anything. If anything dev will find ways to just write even worse code.
Write shader that executes in 0.1ms... nah, just throw this artifacting ML/AI model to process the same effect making game look inconsistent to itself...

And with GPUs I stand firm in my belief that GPUs will be overpriced and everyone will be mad they don't come with more memory.
I guess Jensen is irritated they pushed for 24GB and then 32GB in high-end because today they would prefer it to be at most 16GB. Unlikely we will see more than 32GB anytime soon.
 
XoR_ said:
To be honest today's memory requirements come mostly because of bad development practices.
We were supposed to have texture streaming to fetch closeup textures just-in-time and not kept in the GPU all the time...
...and instead we have issue like shader compilation because devs are too lazy to write short function to recompile all shaders before launching actual game.
We have shaders optimized by tons of if/else statement instead of branchless programming - which could avoid need for TAA completely.
We have TAA which is terribly tuned for both visuals and performance.
We have terrible overuse of various frame and other buffers which do way too little in single pass/buffer wasting tons of memory and memory bandwidth.
We have most games use bad lighting models which don't even run faster than more realistic models.

ML texture compression won't really solve anything. If anything dev will find ways to just write even worse code.
Write shader that executes in 0.1ms... nah, just throw this artifacting ML/AI model to process the same effect making game look inconsistent to itself...

And with GPUs I stand firm in my belief that GPUs will be overpriced and everyone will be mad they don't come with more memory.
I guess Jensen is irritated they pushed for 24GB and then 32GB in high-end because today they would prefer it to be at most 16GB. Unlikely we will see more than 32GB anytime soon.
Click to expand...
The newer tech coming definitely does not solve the current tech and current games and games in development. So yep, memory will be a bitch with the price for entry.

RT plus BVH consumes 10 to 20% more vram memory, hybrid RT. PT adds even much more due to buffers needed for every ray, all lighting while hybrid just needs it for the effects being done, as in reflections.

BVH is a CPU hog, each frame object movement has to be updated and sent to the GPU, to then add a continued stream of textures which never come alone but with shaders for thousands of objects, you start hitting CPU bottlenecks, the bandwidth of PCIe is way sufficient, the bandwidth of system memory is not, CPU cache makes a huge difference. Some games perform better with Intel processors with very fast system ram, that is an example when CPU cache is exceeded and system ram has to be used more.

Don’t know the solution, newer tech will probably start swinging when next gen consoles hit. How well those games will play on current tech is another thing.
 
XoR_ said:
...and instead we have issue like shader compilation because devs are too lazy to write short function to recompile all shaders before launching actual game.
We have shaders optimized by tons of if/else statement instead of branchless programming - which could avoid need for TAA completely.
Click to expand...
we have compilation stuters because they try to avoid those if/else to a maximum ;) (compiler tend to be good at unrolling them and people put a lot of #ifdef type instead of if) or at least you see the tension between the volume of compilation needed at the first launch (would people accept 3 hours, 100GB disk usage?) and if/else avoidance. Shaders compilation are not necessarily a sign of lazyness, but a sign a lot of work being made to do branchless programming.

Modern gpu are better branch predictor than 15 years ago has well, not fully unrolled loop can exist and so on.
 
LukeTbk said:
we have compilation stuters because they try to avoid those if/else to a maximum ;) (compiler tend to be good at unrolling them and people put a lot of #ifdef type instead of if) or at least you see the tension between the volume of compilation needed at the first launch (would people accept 3 hours, 100GB disk usage?) and if/else avoidance. Shaders compilation are not necessarily a sign of lazyness, but a sign a lot of work being made to do branchless programming.

Modern gpu are better branch predictor than 15 years ago has well, not fully unrolled loop can exist and so on.
Click to expand...
Would it be that hard to add option in the menu "compile all shaders"?

Stutters are not sign of branchless programming as much as sign of shaders not being compiled before they are needed.
 
I know it won't happen at this point but I would love to see nvidia allow AIBs to go back to using the old 8-pin power connectors if they want to.
Really hoping that the xx80 cards have 24gb next gen, and hopefully at least 20gb on the xx70 Ti cards.
The 6080 has got to be faster than the rtx 4090. Sad that it's taken so many gens for it to pass the older xx90 card.
Hoping that the top card rtx 6090 isn't 600w anymore. I don't want that much heat dumped in my room, though I know I can lower power limit or UV.
The prices have gotten out of control. I really hope we don't see another msrp increase for each tier next gen.
Availability is another concern. A paper launch is highly likely and then who knows when things will actually improve.
I'm really looking forward to the new process node.
 
  • Like
Reactions: ncjoe
like this
Celcius said:
I know it won't happen at this point but I would love to see nvidia allow AIBs to go back to using the old 8-pin power connectors if they want to.
Really hoping that the xx80 cards have 24gb next gen, and hopefully at least 20gb on the xx70 Ti cards.
The 6080 has got to be faster than the rtx 4090. Sad that it's taken so many gens for it to pass the older xx90 card.
Hoping that the top card rtx 6090 isn't 600w anymore. I don't want that much heat dumped in my room, though I know I can lower power limit or UV.
The prices have gotten out of control. I really hope we don't see another msrp increase for each tier next gen.
Availability is another concern. A paper launch is highly likely and then who knows when things will actually improve.
I'm really looking forward to the new process node.
Click to expand...

Far more likely to design a whole new connector prompting you to buy a whole new power supply
 
THRESHIN said:
My prediction is AI will continue to ruin our lives one GPU at a time.
Click to expand...
Yep i don't even think they will make a next gen at this rate.

Also anyone else feel all these data centers they are building for AI is really just to track everything and everyone in real time 24/7?
 
I predict the Dark Ages of PC gaming 2025-2029 I doubt prices will ever go down in my lifetime. Those Data Centers SSDs will wear out and they still will have to replace them along with other components.
 
littlegreenmenfromhome said:
I am happy about eg. DLSS and so a large portion of gamers, but do cry on about "evil technology".
Click to expand...
While we might get some good form it AI is going to make us all want to jump off a bridge. There's going to be cameras on every single street corner and everything we do will be tracked in real time minority report style. The people developing this junk don't have our best interest in mind so yeah it is evil. We are literally watching every freedom we have disappear and data centers eating all the power it can to do what? Replace the common man so the elites can have robots do everything for them and no people to deal with. Sounds like a amazing future to me.
 
tunatime said:
While we might get some good form it AI is going to make us all want to jump off a bridge. There's going to be cameras on every single street corner and everything we do will be tracked in real time minority report style. The people developing this junk don't have our best interest in mind so yeah it is evil. We are literally watching every freedom we have disappear and data centers eating all the power it can to do what? Replace the common man so the elites can have robots do everything for them and no people to deal with. Sounds like a amazing future to me.
Click to expand...
You should seek help, you sound like suffering for some sort of paranoia and debate the future with delusion will only go south.
 
You must log in or register to reply here.
Back
Top