![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
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.
RDNA 5 speculation
- Thread starter Marees
- Start date
My guess is that RDNA 4 was supposed to have software based crossfire for the top chips but didn't work out. So it had to be scrapped
So AMD is now working on incorporating it in RDNA 5. Remember in ATI days there used to be the x990 cards
Check the Active Interposer Dies (AIDs) in the linked report below
https://www.techpowerup.com/312364/amd-navi-4c-gpu-detailed-shader-engines-are-their-own-chiplets
Last edited:
Could be (I think there was some rumors of very late RDNA 5), but if the RDNA 4 is the RDNA 1-5700xt type of launch, short gap before the real complete stack:
Could it no go, about that fast :
5700xt: Jul 7th, 2019
6800xt: Oct 28th, 2020
early 2025 for RDNA 4, mid 2026 for RDNA 5 ?
Hope springs eternal
Chris_B
Supreme [H]ardness
- Joined
- May 29, 2001
- Messages
- 5,484
- Joined
- Jun 28, 2004
- Messages
- 3,107
RDNA 5 is on the roadmap for 2026. I wouldn't be surprised if AMD pulled in a paper launch at the end of 2025.
But that's contingent on...
How RDNA 4 is received will determine if we will even get a high-end RDNA 5 Sku and when the final RDNA 5 skus are announced.
The top RDNA 4 NAVI 48 SKU needs to compete effectively with the 4870 Super or 5070 Super equivalent
But that's contingent on...
How RDNA 4 is received will determine if we will even get a high-end RDNA 5 Sku and when the final RDNA 5 skus are announced.
The top RDNA 4 NAVI 48 SKU needs to compete effectively with the 4870 Super or 5070 Super equivalent
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 20,200
I think their top part is expected to be $599, possibly $699. They want to repeat Polaris and put one in every other gaming PC.
Top Polaris was 200 dollars
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 20,200
I meant in terms of proliferation. I think AMD has a $200 part in the works, but it might be $250 or $300. Also, money's not worth money any more, so they have to raise prices.
I fully expect Nvidia to paper launch most of their GPUs, too, then focus on workstation cards for 5x the profits.
I'm not so worried about production capacity. It's not like we're seeing shortages of anything else TSMC makes on advanced processes except for NV's datacenter AI accelerators, and those are limited by their fancy packaging facilities rather than the chip fabs. The $40k or whatever AI accelerators are a big complicated package with a bunch of chips in it. I think it's 2 compute dies, a couple others, HBM... much more complicated than a Radeon GPU or Ryzen desktop CPU. The consumer stuff isn't going to need Chip on Wafer on Substrate or whatever they call it like the AI accelerators.
I'm expecting cheaper options from AMD and NV this time around, partly because they have to deal with Intel Arc Battlemage chasing the low end and partly because both of them totally skipped it for the current generation. AMD made a 6500XT, 6400 and 6300 and NV made an RTX 3050, but this time around the bottom end was RX 7600 and RTX 4060 for desktops. I suspect AMD will likely have a less costly option than NV since NV doesn't seem likely to release anything that isn't fast enough to turn on ray tracing (with some help from DLSS) at 1080p, while AMD will be more realistic about it and simply admit that low end cards shouldn't be used for ray tracing.
(If we are talikng RDNA 4 & 50xx)
Nv can just rebadge the 4060 ti 8gb as 5060
AMD's Navi 44 should match the 4060 ti 8gb in all aspects. It will probably sell for $300. Cut down version could be $240
If 7700xt continues to sell, it would drop down to $350 or less
Coming back to topic
RDNA 5 is due end of next year (if all goes per plan)
AMD is planning monster crossfire dies to tackle nvidia. Maybe they might name the top card as RX 9990
I doubt RDNA 5 will have cheap cards as it would be on 3nm
By contrast upcoming battlemage, RDNA4, & 50XX are all on 4nm. So there should be 'budget' cards this time around
- Joined
- Jul 28, 2018
- Messages
- 197
It's all gone a bit quiet on this front... 
It's all gone a bit quiet on this front...
Waiting for more details on UDNA & how that impacts future RDNA architectures
Am I the only one who reads this as horrible news for gaming? I don't want AI in any form on my PC. I'm not going to be using a GPU for it. I foresee higher prices and GPU shortages again as a result of this. All of the cards get bought up for AI or Bitcoin mining again on launch.
Good guess would be that AI upscaling (final image and upcoming texture, then in situation generated asset, voice, dialogue) will be too powerful to not become quite used at least in indie to mid-budget gaming, and it connects gaming hardware to hundreds of billions in R&D and giant volume business. Soon that will sound like I do not want matrix multiplcation in any form on my pc (or the very common in the 90s, no GPU in any form on my desktop computers)
pro and cons.
pro and cons.
- Joined
- Jun 28, 2004
- Messages
- 3,107
AMD's plan is built on moving a mass volume of cards and getting developers onboard. That plan also includes a moderate profit. The larger portion of the profit will come from their AI GPU solutions priced significantly less than their Nvidia equivalent. All in all the market is going to have good gaming choices in the $500 and down category. AI is here to stay, this is going to become physicX 2.0 Soon AI components are just going to be baked into every card as an offload function and compiler libraries will start supporting the calls natively. If you don't have the specific offload support there will be fallback mechanisms similar to how modern games treat older cards.
drutman
Gawd
- Joined
- Jan 4, 2016
- Messages
- 687
AMD needs to stop trying to beat Nvidia performance and undercut their prices to take market share from Jenson. RT is still a novelty and doesn't increase gaming competitive advantage.
Plus Steam has a program that brings frame generation to any card or game not HW specific. Lossless scaling is app. AI market is NV to dominate, gamers are left in the rear now. Plus as lithography scale decreases cost will be higher and higher. The flooding of the quartz mine in the Carolinas will rear its ugly head soon.
Plus Steam has a program that brings frame generation to any card or game not HW specific. Lossless scaling is app. AI market is NV to dominate, gamers are left in the rear now. Plus as lithography scale decreases cost will be higher and higher. The flooding of the quartz mine in the Carolinas will rear its ugly head soon.
This strategy has already failed. The "we need AMD to offer competitive parts at lower prices so Nvidia will lower prices so I can afford them" is the mentality that won out.
Price alone is not going to move the needle for AMD. They need a "MUST HAVE" feature that NV doesn't have. For me, it is superior OSS Linux drivers, but that is a niche feature that is irrelevant to most.
But they have not really been trying to undercut nvidia though, their prices have not been not low enough to make it appealing when they have inferior features and worse efficiency and occasional drivers issues. AMD has been in effect a budget alternative (they do the same things, but a bit or a lot worse) for a long while now, but have not priced their products accordingly.
Most games have DLSS and Reflex whatnot, which are measurably and objectively superior to any similar feature AMD offers. Same goes for NVENC or overall efficiency, it's indisputable and measurable facts. It's literally a bad idea to buy an AMD GPU unless it's a massive price difference and you are sure it will pay off for your use case/lifestyle to save those bucks. I mean Christ, Space Marines 2 that came out a month ago has massive stability issues on AMD 7000 cards. That game is NEW and POPULAR, and AMD still messes it up.
Yes, there is always a few titles and use cases where AMD shines... that you can count on one hand. For the vast majority of consumers, this is a non-argument.
I don't think they can do anything other than play on pricing, they are really far behind and don't have the means to magically close that gap.
Most games have DLSS and Reflex whatnot, which are measurably and objectively superior to any similar feature AMD offers. Same goes for NVENC or overall efficiency, it's indisputable and measurable facts. It's literally a bad idea to buy an AMD GPU unless it's a massive price difference and you are sure it will pay off for your use case/lifestyle to save those bucks. I mean Christ, Space Marines 2 that came out a month ago has massive stability issues on AMD 7000 cards. That game is NEW and POPULAR, and AMD still messes it up.
Yes, there is always a few titles and use cases where AMD shines... that you can count on one hand. For the vast majority of consumers, this is a non-argument.
I don't think they can do anything other than play on pricing, they are really far behind and don't have the means to magically close that gap.
Last edited:
The problem for AMD is if they cut prices nvidia also will follow suit. And nv has a much larger war chest. Best for AMD to not go into direct head to head with nvidia
And there is 2 other problem
1) As mentionned, underpricing is generally not enough it needs a significant price cut as Nvidia can sell less performance for the same dollar usually.
2) Nvidia card by units could be significantly cheaper to make on their end
- AIB could be accepting lesser margin for higher volume
- The very large binning of wide family product
- superior tech/drivers, deliver more fps by die size&memory bandwidth
AD104 now goes from the 7680 core 4070ti down to a variant of the 4060 using just 4352 cores and RTX 3500 desktop and mobile taht used only 5120 all along, 300mm die that is capable to sell at an high price down to just 56% of the chips enabled, that quite the business model.
AD106 you have 4608 (4070 mobile) and 3072 (66% of a small die) with the 4060 on it, all up and down the stack they are fully used with pro RTX-Mobile-Desktop version of the same product, high volume also mean amortizing the R&D over more units (that where consoles market could be really important to amd).
I.e. to be able to undercut Nvidia pricing enough, there is 2 possible things
- A technical reason for their cards to cost less to make, that what the chiplets design tried and failed, that something that could do it
- Undercutting things that Nvidia really do not want to because it would hurt the pro line, like vram amount, issue being if you need to run on a ~10gb console and 6/8gb nvidia cards because of market volume, you will not see many games that do not play well if you do not have 16/20GB of vram.
Unlike the previous plan of not moving mass volume and not get developers onboard
there could be more to the plan than the high end design did not scale as well as expected, actual reason why it would help, but it is not obvious how much better and cheaper the 6700xt would have been without a 6800-6900 card, is there a cost to the mid-range to design something that can scale higher ? Maybe, specially if that mean it must be a chiplet design, but it is never said or asked how concentrating on the mid-range will significantly help it, over the usual and always the less you do the more attention to it you can give.
Last edited:
- Joined
- Jun 28, 2004
- Messages
- 3,107
AMD is planning going head to head with nVidia.... but only on the 4070 and down, where they have a good chance of competing.And there is 2 other problem
1) As mentionned, underpricing is generally not enough it needs a significant price cut as Nvidia can sell less performance for the same dollar usually.
2) Nvidia card by units could be significantly cheaper to make on their end
- AIB could be accepting lesser margin for higher volume
- The very large binning of wide family product
- superior tech/drivers, deliver more fps by die size&memory bandwidth
AD104 now goes from the 7680 core 4070ti down to a variant of the 4060 using just 4352 cores and RTX 3500 desktop and mobile taht used only 5120 all along, 300mm die that is capable to sell at an high price down to just 56% of the chips enabled, that quite the business model.
AD106 you have 4608 (4070 mobile) and 3072 (66% of a small die) with the 4060 on it, all up and down the stack they are fully used with pro RTX-Mobile-Desktop version of the same product, high volume also mean amortizing the R&D over more units (that where consoles market could be really important to amd).
I.e. to be able to undercut Nvidia pricing enough, there is 2 possible things
- A technical reason for their cards to cost less to make, that what the chiplets design tried and failed, that something that could do it
- Undercutting things that Nvidia really do not want to because it would hurt the pro line, like vram amount, issue being if you need to run on a ~10gb console and 6/8gb nvidia cards because of market volume, you will not see many games that do not play well if you do not have 16/20GB of vram.
Unlike the previous plan of not moving mass volume and not get developers onboard
Video cards have gotten stupidly expensive and it's about time the pricing is brought back down to earth. I truly hope AMD pulls this off, if only to bring a little sanity to the pricing.
And how much not having a "real" 8800xt-8900xt sku help them doing that, in what way ? Why would the 8600-8600xt-8700-8700xt need to be more expensive or worst performing if they make a 8800xt version ?
People say it as if it was obvious or as if AMD did not go head-to-head and try to sell volume in the 5700xt-6700xt-7800xt and lower range all along, maybe it is, but it is not for me how significant it changes things that they do not make an extremely similar but 80 CU version of their 8700xt.
As for Dev buying in, console made that relatively working well enough I think, is FSR 2 issue, not being present in enough title ? Getting close to 200 of them and about all 2024-2025 game will have FSR 2 support. They said some PR spin, people repeat it, without asking the following why/how and by how much it help reducing the price and boosting the performance of the mid-range to not launch an actual 8800xt-8900xt, by telling us how it helped the 5700xt and was a detriment to the 6700xt-6750xt for example.
- Joined
- Jun 28, 2004
- Messages
- 3,107
The previoous method which you are endorsing was to focus on building a competitive high end die, making several cut down versions of that and making a lessor die and using that to cover the low end of the market.
The big issue is that a high-end die is complex, expensive to manufacture and generally a low volume seller. In order to pay for the more expensive die, any cut down version of that die has to maintain a certain price point or there would be little or no profit. The lower-end die wasn't much of a consideration, as those aren't exactly jumping off the shelves.
The reality check is that nVidia completely cleaned AMDs clock on the high end graphics cards resulting in AMD having to lower the pricing on a product that was expensive to create. Lower volume of sales made it even less profitable. Lowering the pricing of each card leads to the same issue ( no profit after the bills are paid).
The big difference is that the upcoming NAVI 48 isn't going to be anywhere near as expensive to produce, as the die isn't as large or as complex. That lack of complexity leads to higher yields and a lower cost to produce for each functional die per wafer.
That different approach means AMD can afford to sell Navi 48 at a lower price point as it costs less to build cards from it. AMD currently can't compete from an R&D POV against the 4090 at any price point anyone is willing to pay.
The opportunity is that the 4070 and 4060 are still significantly overpriced and THAT is something AMD can design, implement and compete against.
Nvidia gave up trying in the midrange, the massive performance gaps, and ridiculously moderate price gaps between the 4090, 4080 and 4070 prove that true.
Nvidia could respond, and likely will... but again that'll be an outcome of actual market competition.
The other wildcard is "ray tracing" is great on nVidia's high end cards, but the midrage and lower end cards it just isn't as impressive. AMD might have superior ray tracing capability in the midrange and low-end cards, the rumors seem to indicate that as a possibility.
It's a total supposition on my part, but I do see nVidia creating new cards using existing die's, just to level the playing field with what AMD is going to bring.
As far as I am concerned the graphics market has gone stagnant. I 'm still using a 1080 GTX and I have little interest in upgrading at the current market rates.
But you'd get my attention if one of those newer cards ends up something akin to a 4070 super on performance and offers that kind of experience at a $200-$250 discount.
arthur_tuxedo
Gawd
- Joined
- Jan 8, 2003
- Messages
- 842
AMD’s problem is that they can never seem to deliver a one-two punch. The 6800 XT clobbered the 3080 10Gb in value for money, efficiency, and longevity, but the 7900 series was a big disappointment and 8900 is a no-show. They shouldn’t have put all of their eggs in the MCM basket. Assuming they work the bugs out, you can extrapolate some very impressive performance numbers from top RDNA 5 based on the midrange RDNA 4’s performance expectations of being somewhere near a 4070 Ti Super. But RDNA 5 is not going to come out soon enough to help consumer choice and pricing for at least another year.
That can be overstated, $650 vs $700, in 2024 the 3080 still tie or win many benchmark (some by more than 10%):
https://www.techpowerup.com/review/black-myth-wukong-fps-performance-benchmark/5.html
https://www.techpowerup.com/review/silent-hill-2-fps-performance-benchmark/5.html
https://www.techpowerup.com/review/senuas-saga-hellblade-2-benchmark/5.html
https://www.techpowerup.com/review/avatar-fop-performance-benchmark/5.html
https://www.techpowerup.com/review/star-wars-outlaws-fps-performance-benchmark/5.html
average in a recent 2024 test suite:
It was a winner, but not by some clobbering amount and DLSS could end up making it close to some.
And with 2024 drivers:
https://www.techpowerup.com/review/nvidia-geforce-rtx-4080-super-founders-edition/41.html
40 watt difference, 50 for max, for the model they had 3080 easier for 20ms spiking than the 6800xt, a win but not really clobbering.
- Joined
- Sep 7, 2011
- Messages
- 10,161
Well, that's the big issue. AMD is never the best for more than half a generation, and it's usually by 5-10%.
Whereas Nvidia is the best by %30-50 and rarely gives up that crown.
arthur_tuxedo
Gawd
- Joined
- Jan 8, 2003
- Messages
- 842
It was never $650 vs $700, though. For a long time it was $1,000 to $1,300 for 6800 XT vs up to $2,000 for 3080. When prices came back down to Earth, the 6800 XT hit its MSRP much sooner. The 3080 wasn’t widely available at MSRP until the generation was on its way out. Even if the pandemic craze pricing had never happened, shipments showed that the $700 was only intended to be a phantom price to trick reviewers, available only on low-volume FE and reference cards that would be constantly out of stock, and the real price was more like $800.
But even if both MSRP’s had been truthful, 40 to 50 watts is a huge difference in efficiency, and 10 GB VRAM was only barely enough for 4K in 2020, and is woefully insufficient now. You might see similar average FPS in bechmarks, but the 10 GB card is going to have a lot more stutters as time goes on.
You’re right that people who play a lot of games with good DLSS implementations and bad or missing FSR might still prefer the 3080, or if they do things like AI video upscaling that use the Tensor cores (although Topaz Video AI is buggier with Tensor and 10 GB is a huge limitation for upscaling), but I suspect that most gamers are glad they chose a 6800 XT over a 3080. I know I am.
Nicehash create stange price, they ended up probably similar value for money if you ran it enough hours a day here (i imagine that what was setting those relative price)
Balanced by games that do not have an raytracing off option, consoles and the popularity of 10gb or less GPU kind of force the hand of gamedev to make playable game with 10 or less gb. Are people playing Blackmyth (a record breaker seller) feel like the 3080 has less longevity than the 6800xt, it is a mixed bag.
520mm of tsmc 7 on 256 bits of regular GDDR6, was probably cheaper than a 628mm of samsung 8 on 320 bits with fancy gddr6x.
Are people playing BMW with Hard RT on on a 3080 ?
At what settings ??
Obviously not, at lower setting than that, but it is still an example at playable setting where the 3080 did not age worst than a 6800xt.
It is not like longevity did not end up being a bit of a mixed bag, has expected, is clobbering win really representative, a good one, but not one to put NVIDIA to the ground.
Last edited:
arthur_tuxedo
Gawd
- Joined
- Jan 8, 2003
- Messages
- 842
You make a good case that “clobbered” was too strong of a word. But whatever word I should have chosen, that’s what I’d like to see AMD bring to the table with Navi 5 in 2026.
- Joined
- Jul 5, 2012
- Messages
- 4,622
Don't we all! But let's face it the odds of that happening are extremely slim.
With the actual street price, for a gamer never running nicehash that was probably a correct word, I had more the $50 price gap in mind, it is just something I often see everywhere about the 2 cards, and yet it seem that half the big game released 4 years later run better on the 3080...
Next gen 9070 xt successor could have 36 wgp or 72 cu
https://x.com/Kepler_L2/status/1946289792657793093
https://x.com/Kepler_L2/status/1946289792657793093
Could AMD simply attach a MID (Media Interface Die) with Display/Media Engine and PCIe to the Magnus GCD and turn that into a desktop GPU?
If the new xbox has the 9070 xt successor in it, what will the system cost at launch 999? I feel like they need another generation like the Xbox One and PS 4 where they pull back on the specs a bit, things are way too expensive right now and I'm not so sure it's wise to sell at console at a huge loss.
Latest should be the latest to make per unit at the same level of wanted performance ? RDNA 5 or newer depending on launch date make a lot of sense.