Probably. Though there's always the chance that China will copy/invent a better/equal AI engine with another "Deepseek" moment and the house of cards will come crashing down
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Nvidia RTX 6000 series (codename Rubin)
- Thread starter Grim5
- Start date
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,712
I think the 6090 will be less than the 5090 at release.
This time around we're getting a new (smaller, cheaper) manufacturing process, new architecture. This will allow the die to be (much) smaller.
They won't have to release a 600W complicated heavy card. They can release a ~350W card with 16/24GB that will slightly outperform the 5090 in traditional rasterization and significantly outperform it in overall DLSS gaming due to software locked new DLSS modes etc.
Enhanced memory compression tech/software will also mean 16GB/24GB will be adequate for gaming and will drive costs down.
At least that's what I hope. I don't want to see a £3k 6090FE
That will be a 6080. Naturally, it should be cheaper.
When do we guess these will launch? Q1 2027, maybe Q2 at the latest? Or does the RAM/AI stuff mean we're looking at 3 years now for a new gen?
I really want to upgrade but there is nothing to upgrade to outside of the 5090 (Don't really want a 600W card either lol) or a used 4090. 5080 is like a side-grade for a grand.
I really want to upgrade but there is nothing to upgrade to outside of the 5090 (Don't really want a 600W card either lol) or a used 4090. 5080 is like a side-grade for a grand.
I can't see mid-range happening until late 2027 and even then I wouldn't be surprised if it was 2028.
If you're not after a high power halo product like the 6090 is likely to be, you might see a 6080 mid next year but I'd not hold my breath.
Hold in mind the above isn't even accounting for price or parts volumes, even if they launch when I expect or earlier they could very well be like hens teeth.
Soldato
- Joined
- 4 Feb 2011
- Posts
- 2,715
I'll go with Q4 2027 earliest, most likely Q1 2028.
I can't see there being any significant, or even worthwile GPU upgrades for gaming over current offerrings for some considerable time, other than in the area's of AI compute. For the next two years at least there is limited resources, wafer capacity and fab time to say the least. The investors and pension funds ploughing their wealth into AI buildouts expect a return - they need us to spend our money on services and not our own hardware.
AMD are focused on concqering the enterprice CPU and console GPU spaces. Intel continue to dominate the mainstream CPU market and with their close ties to Nvidia now, all Nvidia needs to do is work on more energy efficient versions of existing IP while maintaining their lead in the AI hardware and software ecosystem.
Global energy is the biggest issue. We were supposed to all be driving electric cars within the next decade despite it being impossible to recharge them all - and that was before the data centre buildouts are now after the energy we use to power our homes.
We now have PSU's on the limit of what American domestic consumers can typically power, and just this last week I bought from here a 2.2KW PSU to simply do what I did in 2012 with a 750W unit - feed a pair of GPU's. This can't continue.
Nvidia and co have to start selling far more energy efficient products to reduce our domestic consumption to free up more electricity for data centres so they can run even more Nvidia hardware on top of their own efficiency gains.
Last year Nvidia cut a few SKU's to make more product with fewer GDDR7 IC's. Arguably this was initially because the Vera CPU element of their IP consumed DDR 7, but Vera has now been replaced (other than for pre-orders) for a CPU IP using on die SRAM.
From within 1 year, going from not having enough GDDR7 to sell so many VRAM rich GPU's, to the 'apparent' other extreme of not having enough GPU's, reigniting production of a two generation old RTX 3060, I believe supports my assumption;
Significant memory reducing efficiencies have made impressive gains, which is not just about reducing the memory capacity requirement to render a frame - but the therefore reduced memory bandwidth to render it. GPU memory controllers consume a lot of power and the 'revamped' 3060 using higher capacity 3GB memory dies, uses fewer memory controllers. Combine this with the high yielding 3060 dies which are essentially binned rejects, on a smaller fine tuned node and we get far fewer wafers to produce far more mainstream AAA capable GPU's for far les power consumption.
If this years 3060 experient is successful and profitable, then I think pretty much all we're going to get over the next few years is more energy efficient versions of 5080's and 90's on mature processes with enough tweaks to AI capabilities that developers can remake their 40 year games back catalogues for a new audience and those that missed them the first time round. The latest DLSS relevations show that even supermario cart could effectively be rendered into a photorealistic remake in real time before long.
I don't even hold hope that the datacentre GPU buyouts will later flood the market in any useful way to gamers; The HBM populating those substrates not only is poorer in latency, but the interconnects used to stack the IC's is engineered for sustained activity and temperature. The variances arising from gaming on and off leads to premature failure, and I do wonder if the stacked cache of the X3D Ryzens burnouts is caused by the same temperature swing effect when overvolted into full throttle.
Anyway, I can't see prices increasing for the next GPU generation, other than typical early adopter premiums and typical supply/demand larks (like when most of us feared missing out on loo roll during the pandemic). When the cost of fashion spikes, it's savy to buy last season. Unfortunately, I doubt we will see a DDR3 / Xeon highlight out of all of this either.
AMD are focused on concqering the enterprice CPU and console GPU spaces. Intel continue to dominate the mainstream CPU market and with their close ties to Nvidia now, all Nvidia needs to do is work on more energy efficient versions of existing IP while maintaining their lead in the AI hardware and software ecosystem.
Global energy is the biggest issue. We were supposed to all be driving electric cars within the next decade despite it being impossible to recharge them all - and that was before the data centre buildouts are now after the energy we use to power our homes.
We now have PSU's on the limit of what American domestic consumers can typically power, and just this last week I bought from here a 2.2KW PSU to simply do what I did in 2012 with a 750W unit - feed a pair of GPU's. This can't continue.
Nvidia and co have to start selling far more energy efficient products to reduce our domestic consumption to free up more electricity for data centres so they can run even more Nvidia hardware on top of their own efficiency gains.
Last year Nvidia cut a few SKU's to make more product with fewer GDDR7 IC's. Arguably this was initially because the Vera CPU element of their IP consumed DDR 7, but Vera has now been replaced (other than for pre-orders) for a CPU IP using on die SRAM.
From within 1 year, going from not having enough GDDR7 to sell so many VRAM rich GPU's, to the 'apparent' other extreme of not having enough GPU's, reigniting production of a two generation old RTX 3060, I believe supports my assumption;
Significant memory reducing efficiencies have made impressive gains, which is not just about reducing the memory capacity requirement to render a frame - but the therefore reduced memory bandwidth to render it. GPU memory controllers consume a lot of power and the 'revamped' 3060 using higher capacity 3GB memory dies, uses fewer memory controllers. Combine this with the high yielding 3060 dies which are essentially binned rejects, on a smaller fine tuned node and we get far fewer wafers to produce far more mainstream AAA capable GPU's for far les power consumption.
If this years 3060 experient is successful and profitable, then I think pretty much all we're going to get over the next few years is more energy efficient versions of 5080's and 90's on mature processes with enough tweaks to AI capabilities that developers can remake their 40 year games back catalogues for a new audience and those that missed them the first time round. The latest DLSS relevations show that even supermario cart could effectively be rendered into a photorealistic remake in real time before long.
I don't even hold hope that the datacentre GPU buyouts will later flood the market in any useful way to gamers; The HBM populating those substrates not only is poorer in latency, but the interconnects used to stack the IC's is engineered for sustained activity and temperature. The variances arising from gaming on and off leads to premature failure, and I do wonder if the stacked cache of the X3D Ryzens burnouts is caused by the same temperature swing effect when overvolted into full throttle.
Anyway, I can't see prices increasing for the next GPU generation, other than typical early adopter premiums and typical supply/demand larks (like when most of us feared missing out on loo roll during the pandemic). When the cost of fashion spikes, it's savy to buy last season. Unfortunately, I doubt we will see a DDR3 / Xeon highlight out of all of this either.
Soldato
- Joined
- 19 Feb 2007
- Posts
- 15,785
- Location
- Sol
All current info points to around Q3'ish next year.
Sounds about right, Even at the silly inflated prices, Nvidia are selling mountain loads of 5000 series.
Last edited:
If you buy something now and it breaks in 6 months time, the every worst that can happen is they refuse to replace on account it is too costly and simply give you your money back.
The only reason to hold off buying what you can both afford and need is if you have good reason to suspect prices will drop before the end of your warranty or patience.
Associate
- Joined
- 17 Mar 2025
- Posts
- 620
- Location
- solent sound
the issue with that is you get your money refunded MINUS a whack for the use you have had [or the use they assume you have had] and its a fairly big chunk from memory its worked out as if the day the warranty is up the item is worth zero but i think thats the best case scenario ie after a few months
you are taking a fairly sizable loss on the refund. hence why im ultra paranoid about my ram dying, not only did i buy just before the price increases [so in the case of needing to replace it im going to be having to pay the inflated amount] plus Id have a big chunk deducted from my refund becasue ive
had a few months of use out of it.
rescission apparantly is the word im looking for lol
But a product is still guaranteed to be, and remain free, of defects for the period set under consumer law I understand?
Confusing purchase guarantee for warranty, as applied to wear and tear items such as white goods, Electronics that are an integral part of a consumer item, rather than being the consumer item I mean.
Buy a car battery and it fails within its three year guarantee means either a refund or replacement, for instance.
Confusing purchase guarantee for warranty, as applied to wear and tear items such as white goods, Electronics that are an integral part of a consumer item, rather than being the consumer item I mean.
Buy a car battery and it fails within its three year guarantee means either a refund or replacement, for instance.
Associate
- Joined
- 17 Mar 2025
- Posts
- 620
- Location
- solent sound
apparently not so with computer components, theres been a fair bit of discussion on this subject recently on this forum.
some particularly bad examples where people really lost out because of it, and the obvious people worrying about similar happening with the ram crisis.
but yeah i would not go into this thinking I'm fully covered because it might bite me on the behind
some particularly bad examples where people really lost out because of it, and the obvious people worrying about similar happening with the ram crisis.
but yeah i would not go into this thinking I'm fully covered because it might bite me on the behind
I know my Ram was advertised and thus sold with a lifetime warranty - whether it meant the Ram's or mine isn't clear.
I know my RTX purchase has a three year warranty, extendable to 5 years. As long as it shows no signs of rage quitting, and it's entirely dead, I'm not sure how any degree of wear could be determined if I hadn't taken it out of the box in all that time because of a PSU shortage due to the AI buildout. Regardless, an extendable warranty is suggestive.
If one of my Noctua fans suddenly stops working within 12 months, I'm of the opinion they would want to to know why and thus not quibble.
In many cases though, the seller is the agent between consumer and supplier. They are first port of call followed by them furnishing their suppliers information if required to escalate to the supplier or manufacturer. Manufacturers provide the warranties after all, a seller the guarantee to attract sales otherwise under consumer protection act.
some years back mind, I bought from a shop a pair of 40GB HDD's for a raid 0 array (Athlon Socket 939 era). One failed after 8 moths so i took both drives back to the shop (1 being useless without the other). If they could have replaced the dud drive then it was on me to rebuild my array. Because they could not replace the drive and offered refund, I insisted they had to refund the other drive because I bought them as a matching pair specifically for a raid array guaranteed to work by the shop. I got full refund and bought a pair of WD 36GB raptors after that.
I
I know my RTX purchase has a three year warranty, extendable to 5 years. As long as it shows no signs of rage quitting, and it's entirely dead, I'm not sure how any degree of wear could be determined if I hadn't taken it out of the box in all that time because of a PSU shortage due to the AI buildout. Regardless, an extendable warranty is suggestive.
If one of my Noctua fans suddenly stops working within 12 months, I'm of the opinion they would want to to know why and thus not quibble.
In many cases though, the seller is the agent between consumer and supplier. They are first port of call followed by them furnishing their suppliers information if required to escalate to the supplier or manufacturer. Manufacturers provide the warranties after all, a seller the guarantee to attract sales otherwise under consumer protection act.
some years back mind, I bought from a shop a pair of 40GB HDD's for a raid 0 array (Athlon Socket 939 era). One failed after 8 moths so i took both drives back to the shop (1 being useless without the other). If they could have replaced the dud drive then it was on me to rebuild my array. Because they could not replace the drive and offered refund, I insisted they had to refund the other drive because I bought them as a matching pair specifically for a raid array guaranteed to work by the shop. I got full refund and bought a pair of WD 36GB raptors after that.
I
what you really mean is the HUUUUUUUGE price increase vs performance.Really looking forward to Rubin leaks, hoping it's going to be another Maxwell moment with a huge performance per watt improvement
The more expensive it is, the easier it will be to get one on release daywhat you really mean is the HUUUUUUUGE price increase vs performance.
The more expensive it is, the easier it will be to get one on release day
Don't worry as Nvidia has already covered that angle of thinking.
The more expensive it is, The less there will be available to buy on release day
Last edited:
I don't see any reason nvidia wouldn't double down on their strategy of the last few years. If you told me in 2015 that in 2026 I'd be willing to spend 3k~ canadian on a graphics card if I could just find an upgrade over 5 years - I'd say that's just silly sounding, you're a big silly liar and I'm going to detonate a ballistic fist nuke rite square on ur manpleaser for ur silliness, courtesy of my piston pectorals.
And also that the graphics don't even really look different anyways
And also that the graphics don't even really look different anyways
In 2016 i paid around £400 for the top nvidia GPU at the time (MSI GTX 1080)
(Not including the highly over priced Titan cards)
Last edited:
My last new build, prior to my current one, was in 2012 when AMD were trading flagship blows with Nvidia and a Radeon 7970 Ghz Edition cost me around £370. I replaced this with a GTX 1070 5 or 6 years later for less, thinking the cost of the Titans to be barking mad....
The hit or miss graphics fidelity improvements of late (and incremental improvements from Ada Lovelace > Blackwell > likely Rubin) arise from the traditional graphics hardware blocks of ROPS and basic shader layouts stagnating. For standard (rasterized) 3D gaming, the physical hardware changes are minimal and most modern AAA titles are built as console ports, thus tailored to AMD hardware which lack the advanced hardware blocks found in Nvidia architectures. It simply is hit or miss whether developers embrace this tech fully, if at all, though Nvida funds developer relation teams to get patches into PC game ports.
The angle Nvidia is taking with Rubin GPU's concerns improving internal memory bandwidth by using a smaller node process with a more efficient layout to shorten paths which further reduces energy consumption. The transistor increase is primarily about shifting the traditional focus from raw math calculations to data-movement efficiency by enhancing RT and Tensor cores. While this ultimately aims to address the memory wall bottlenecking AI reasoning models, for gaming, speed + efficiency = reduced memory footprint and a lower current draw - essential given a shrinking of GPU die making heat dissipation even harder, which is another cost driver in the bill of materials.
In essence, the RTX 6000 series aims to achieve Ray Reconstruction and AI frame generation without the latency overheads of upscaling by enabling full path tracing in full resolution - it should at least theoretically make badly ported PC games more playable as a bonus.
A problem is the data centre variants will use stacked Rubin dies which are only reliable at such high current draw and temperature in sustained linear workloads created by training models for weeks at a time. The anticipated demand , if not already suggestively subjecting the desirable and useful RTX 6000 series gaming cards to a paper launch and into the hands of scalpers, is more likely than not going to yield a £1000 minimum commodity as the norm.
I collaborate my earlier thoughts by reckoning a tried and tested Blackwell is the current smart purchase if one has a back catalogue of games they want to play, while allowing some degree of future proofing. The as-yet untested RTX 6000 only makes sense to me to wait out if significantly improved ML and generation capabilities in addition to true path tracing without penalty in games is the absolute priority. A last generation card is viable in the same way a next generation card will impact current generation cards - it falls on availability and a buyers patience.