AusWolfThe article was on about consumer cards. Would I ever pay more than £1,000 for a gaming card? Well, if you triple my salary, then sure. Otherwise, not a chance.

Agreed $1000 is my upper limit for a GPU.

And me selling my 6800XT for $600 and buying the 7900XTX for $1200 so cost me $600 for the upgrade was why I did it.

If I had to start from scratch with no gpu to sell I probably won't have done it.

MakaveliAlot of people in forums not just this one seems to either not understand or skip over the business side of both of these companies.

1. The enterprise market means more to them than to consumers.
2. Shareholders take priority over consumers
3. Profit > feels good or doing the right thing

Pay more attention to what they do and how they operate and less of the AMD vs Nvidia.

I can appreciate that I know that they are. So is the Company I work for but while our techs will only do their work, the competition (which is also public) will cut our cables when they enter a customer's home. Culture matters at Companies and identities are earned. In my 40 years of being around PCs I can say objectively that AMD are a more holistic company than Nvidia. Getting upgrades for free without changing the GPU also feels pretty good I have to say. Open Source is also something that AMD has always been into.

kapone32I can appreciate that I know that they are. So is the Company I work for but while our techs will only do their work, the competition (which is also public) will cut our cables when they enter a customer's home. Culture matters at Companies and identities are earned. In my 40 years of being around PCs I can say objectively that AMD are a more holistic company than Nvidia. Getting upgrades for free without changing the GPU also feels pretty good I have to say. Open Source is also something that AMD has always been into.

This implies NvIdia cards receive no performance improvements, feature updates or functionality in the lifetime of the hardware. You're just wrong, man. AMD's not a good guy. Take this image out of your head. Even AMD's relative "friendliness" towards open source is because they don't actually develop the praised Linux drivers...

Dr. DroThis implies NvIdia cards receive no performance improvements, feature updates or functionality in the lifetime of the hardware. You're just wrong, man. AMD's not a good guy. Take this image out of your head. Even AMD's relative "friendliness" towards open source is because they don't actually develop the praised Linux drivers...

You would be surprised by how much a GPU maker can hinder the open source community if they want to make their lives difficult. In any case, AMD does contribute to the AMDGPU code. This is also shown by the sharp increase in their commits to the Linux Kernel after they started working on AMDGPU.

Aside: I wonder why Phoronix chose to use red for Intel and blue (actually purple) for AMD.

AnotherReaderYou would be surprised by how much a GPU maker can hinder the open source community if they want to make their lives difficult. In any case, AMD does contribute to the AMDGPU code. This is also shown by the sharp increase in their commits to the Linux Kernel after they started working on AMDGPU.

Aside: I wonder why Phoronix chose to use red for Intel and blue (actually purple) for AMD.

No, I know that. Especially when they don't release the GPU's technical data to OSS developers, which is what Nvidia is "guilty" of doing. But it's not entirely to AMD's credit here in this specific situation.

Dr. DroNo, I know that. Especially when they don't release the GPU's technical data to OSS developers, which is what Nvidia is "guilty" of doing. But it's not entirely to AMD's credit here in this specific situation.

Yes, providing proper documentation is not sufficient, but they contribute to the project as well and haven't left all the work to the open source community. Of course, they aren't the first ones in the PC space to do so; Intel's Linux support is exemplary.

TechLurkerI read this as more that RDNA4's potential would allow for a strong multi-chip GPU, if they can improve the links between cores more. RDNA3 is a worthwhile attempt that showed it can work, they need to mature it more and if the cores for RDNA4 are as good as claimed, pairing two of them for even more performance would be a viable option. But the weak link has been the interconnects and programming to make the most of it, which AMD is probably still working on.

That said, rumors claim RDNA5 is going back to a monolithic chip, or a hybrid monolithic+chiplet, as it's showing more promise in the near-term, while RDNA4 is a second attempt to refine the multi-GPU chiplet route until they can improve it for RDNA6 or 7. At the same time, RDNA3 is showing very good performance in AI workloads, so it's possible that their approach there will quickly lead to dividends in the rising AI sector (moreso since CDNA3 was based on RDNA3 work), and if RDNA4 is just refined RDNA3, could lead to even better AI performance.

If AMD is also leveraging 2 rival groups within for GPU development like they are for Ryzen (to ensure they always have options), then I expect we'll probably see alternating periods between performance uplift (monolithic or semi-monolithic) and efficiency gains (chiplet-based) until the chiplet method becomes good enough to compete in both while saving cost.

Those are the opposite to the rumours I've read. High end RDNA 4 was having trouble with chiplets because it was going to be 20 chiplets vs RDNA3's 7 chiplets IIRC. They were having huge problems getting it to work at the power and performance levels they wanted, and decided to drop it and release lower end monolithic RDNA4. This would allow RDNA5 more money, people and time to achieve it's goals while staying with a chiplet design. Getting high-end RDNA4 to work would have greatly eaten into RDNA5's time-line. Given that Blackwell for desktop won't be out probably until Q2 2025 and RDNA5 is Q4 (maybe) 2025, AMD won't be too far behind Blackwell. But I still don't see how AMD can offer a 7900 class mid-tier RDNA4 product for half the price and not make the 7900's look like a stupid investment in H2 2024.

chipsandcheese.com/2024/01/28/examining-amds-rdna-4-changes-in-llvm/

Chips And Cheese has done an exceptional job extracting details about RDNA 4 from LLVM. It seems to me that changes are mainly aimed at efficiency and AI.

kapone32This was a News story in 2016. You can believe what you want. Watch the Level One video if you doubt me.

And those are financial details from *checks page* today.

Either way your claim that AMD is not greedy because it is owned by Saudi Royal Family has to be up there with the most bizzare claims I've heard, even if it were true it doesn't make a lick of sense.

R-T-BAnd those are financial details from *checks page* today.

Either way your claim that AMD is not greedy because it is owned by Saudi Royal Family has to be up there with the most bizzare claims I've heard, even if it were true it doesn't make a lick of sense.

I am not saying they are not greedy just not the same as Nvidia.

Not saying any company is greedy. I hope you guys are past that notion arguing which one is better and which is worse or good or bad. I mean, these are companies so their first priority is to make money. I mean, if you had a business you would do what's best for your company and try to increase profits by adjusting to the market. If you have an opportunity you would use it right? So are these companies. Some actions from those can be debatable of theirs righteousness. Defending any is foolish. They are doing what they are doing to secure their businesses. Anyone has a right to their opinion but I don't think greed per se is driving these companies. They have a business and they want to keep it and prosper. God know when another pandemic or war or any other event screwing economy come.

I would want to see AMD moving chiplets to GPU segment. It worked with CPUs i really can't see why it would not have worked with GPUs. I think there are more obstacles with GPUs (latency is the worst one I can think of from the top of my head and scheduling i think). Wonder, if AMD would need to use some sort of I/O chiplet like with CPUs. Probably yes to avoid the chiplets to be seen as 2 different graphics chip.

ratirtNot saying any company is greedy. I hope you guys are past that notion arguing which one is better and which is worse or good or bad.

Some of us are. All companies bottom line is the shareholders dollar, unless of course they are a private enterprise or something in which case the owners word is law. That can directly represent the owners morals of course.

2024/2025 will be a dark year for high end gaming GPU pricing.

JismAMD opted for the use of chiplets due to the ever rising cost of going or building monolithic GPU's as Nvidia is doing. On top of that with monolithic you risk having more "faulty" chips out of a wafer, thriving up costs because you can sell less.

The RDNA3 series might not have bin that impressive, it will take a few generations to fully uncork it's potential. It was just the start.

The good margins are within the midrange of GPU's - not the 2000$ costing GPU's. Failed? I think you half understand what they are doing.

Whole CDNA (=Mi300X) is expected to skyrocket in regards of sales.

The origin of the problem lies in TSMC and its ever increasing wafer prices. If N2 node costs as much as 25-30,000$, imagine how much a moderate-sized 400-sq.mm die would cost?
People are not going to pay 3000$ for an RTX 6070 with 350 sq.mm die.

AMD must move away from TSMC. I see that Samsung charges better.

BTW, TSMC no longer makes higher profits because of more sales, but simply because of higher pricings. This is not sustainable and its bubble will go boom... and disappear.


www.trendforce.com/news/2024/01/24/news-tsmcs-2023-wafer-average-selling-price-rises-by-22-driven-by-n3-process-success/

3valatzyThe origin of the problem lies in TSMC and its ever increasing wafer prices. If N2 node costs as much as 25-30,000$, imagine how much a moderate-sized 400-sq.mm die would cost?
People are not going to pay 3000$ for an RTX 6070 with 350 sq.mm die.

AMD must move away from TSMC. I see that Samsung charges better.

BTW, TSMC no longer makes higher profits because of more sales, but simply because of higher pricings. This is not sustainable and its bubble will go boom... and disappear.


www.trendforce.com/news/2024/01/24/news-tsmcs-2023-wafer-average-selling-price-rises-by-22-driven-by-n3-process-success/

If you want a 900 watt TDP 7900 XTX, yeah use Samsung to build it :laugh:

TSMC isn't the problem. Allocation wouldn't change even with more capacity because to AMD, Radeon is neither their biggest nor its most important business.

Dr. DroIf you want a 900 watt TDP 7900 XTX, yeah use Samsung to build it :laugh:

TSMC isn't the problem. Allocation wouldn't change even with more capacity because to AMD, Radeon is neither their biggest nor its most important business.

It's using chiplets on the old 7nm process (N6) and Navi 33 (RX 7600 XT) is on the old 7nm process (N6).
Samsung for sure has better offers than that!

The chiplets introduce performance issues, yeah, economically they look like a solution but it's more like a fake solution because they leave performance in the glue between those chiplets!

3valatzyIt's using chiplets on the old 7nm process (N6) and Navi 33 (RX 7600 XT) is on the old 7nm process (N6).
Samsung for sure has better offers than that!

The chiplets introduce performance issues, yeah, economically they look like a solution but it's more like a fake solution because they leave performance in the glue between those chiplets!

Chiplets are flexible but they're not without their drawbacks, which is why that rumor that said AMD was developing a GPU that had each and every internal block on a separate chiplet is complete BS (you don't use 2D chiplets but rather 3D stacked packaging to do something like this while minimizing the latency incurred and maximizing the bandwidth available in between modules), not to mention actually packaging an assembled processor out of 30+ sub-10mm² chiplets would be an absolute uncoolable nightmare of an extremely fragile Lovecraftian horror - it's just not happening.

The technology would permit mixing say, Samsung 8N (the node used on Ampere) for auxiliaries with a TSMC N4P for the GPU core, but I still strongly suspect this is neither necessary nor affordable at the present time, and would likely not increase the volume of cards that AMD shifts. Remember, the problem isn't stock, it's the product itself, and the only market where AMD has managed to make a significant incursion was the European market because they're extremely sensitive to pricing, it's a penny pinching consumer culture and they'll avoid spending a single euro above what's absolutely required, and they are very willing to make concessions in experience and functionality if it can be considered "superfluous" and it means they get a deal.

Mixing the foundries is also an option but I meant AMD dropping TSMC altogether and going with Samsung 3GAE (150 MTr/sq. mm) (shipping since 2022) and Samsung 4LPP+ (137 MTr/sq. mm) (in production as of 2023).

Dr. DroRemember, the problem isn't stock, it's the product itself

That's what I am also saying - worse product deserves lower pricing. AMD's current pricing for the 7900 XTX is much higher than what a normal market would pay.
People are sensitive to pricing these days more than previously first because it's stagflation, and because we are used to much cheaper prices.

RX 7900 XTX if released in 2015 would have cost 450$.

Dr. DroChiplets are flexible but they're not without their drawbacks, which is why that rumor that said AMD was developing a GPU that had each and every internal block on a separate chiplet is complete BS (you don't use 2D chiplets but rather 3D stacked packaging to do something like this while minimizing the latency incurred and maximizing the bandwidth available in between modules), not to mention actually packaging an assembled processor out of 30+ sub-10mm² chiplets would be an absolute uncoolable nightmare of an extremely fragile Lovecraftian horror - it's just not happening.

The technology would permit mixing say, Samsung 8N (the node used on Ampere) for auxiliaries with a TSMC N4P for the GPU core, but I still strongly suspect this is neither necessary nor affordable at the present time, and would likely not increase the volume of cards that AMD shifts. Remember, the problem isn't stock, it's the product itself, and the only market where AMD has managed to make a significant incursion was the European market because they're extremely sensitive to pricing, it's a penny pinching consumer culture and they'll avoid spending a single euro above what's absolutely required, and they are very willing to make concessions in experience and functionality if it can be considered "superfluous" and it means they get a deal.

A strategy like the rumoured one for high-end RDNA4 would have only been a good idea if they followed in MI300's footsteps by making these choices:

• moderately small GCD dies; each MI300 XCD is 115 mm^2 and has 40 Compute Units (contrast with 96 CU for the 7900 XTX GCD)
• base die with last level cache, multimedia engine, and off-chip interfaces, i.e. DRAM and PCIe

Note that unlike the MI300, the base die would serve as an interposer too. This wouldn't be as expensive as MI300, but I suspect that unlike the current strategy, it would be too expensive to scale down. So this would have only been a good choice for a 5090 competitor.

Let's see what they actually do with RDNA4. I haven't read any leaks about CU counts or VRAM bus width so I suspect they will be following the RDNA 3 strategy of different sized GCDs and 1 type of MCD. If my hypothesis about the strategy for high-end RDNA4 being a poor fit for cheaper SKUs is right, then we will likely see at least a successor to the 7800 XT.

AnotherReaderLet's see what they actually do with RDNA4. I haven't read any leaks about CU counts or VRAM bus width so I suspect they will be following the RDNA 3 strategy of different sized GCDs and 1 type of MCD. If my hypothesis about the strategy for high-end RDNA4 being a poor fit for cheaper SKUs is right, then we will likely see at least a successor to the 7800 XT.

Navi 48 will be the successor of Navi 32 with one of the following configurations, maybe both for the XT and XL versions.
4096 shaders across 192-bit memory bus.


www.tweaktown.com/news/94533/amds-next-gen-rdna-4-navi-44-and-48-from-radeon-rx-8000-series-gpus-appear-in-linux/index.html

3valatzyNavi 48 will be the successor of Navi 32 with one of the following configurations, maybe both for the XT and XL versions.
4096 shaders across 192-bit memory bus.


www.tweaktown.com/news/94533/amds-next-gen-rdna-4-navi-44-and-48-from-radeon-rx-8000-series-gpus-appear-in-linux/index.html

Thanks. Somehow I missed this rumour. This seems plausible though I'm not sure about the GDDR7. Navi 48 has about the same WGP count as Navi 32 (30 WGP). The rumours of it matching the 7900 XTX seem rather unlikely. If they have increased clocks to what the initial presentations for RDNA3 claimed, then it could match the 7900 XT though it might require some increases in performance per WGP as well as that gap is pretty large.

AnotherReaderThanks. Somehow I missed this rumour. This seems plausible though I'm not sure about the GDDR7. Navi 48 has about the same WGP count as Navi 32 (30 WGP). The rumours of it matching the 7900 XTX seem rather unlikely. If they have increased clocks to what the initial presentations for RDNA3 claimed, then it could match the 7900 XT though it might require some increases in performance per WGP as well as that gap is pretty large.

With those specifications, it indeed looks more like the Navi 33 (RX 7600 XT) successor, and a shrink with some tweaks of Navi 32 (RX 7700 XT / RX 7800 XT).
Performance guesstimate - around RX 6950 XT or RX 7900 XT.

Price should be 399$ to be well accepted by both the reviewers and the gamers.

While they are possible (and adding features not clearly apparent by the specs) I really don't agree with those rumors. They don't make much sense on their face, but architectures can drastically change in ways not apparent until we see testing/reviews./white papers/etc.

While I am in-fact pulling this out of my speculative ass, I think 4096/8192 appears much more plausible.

Shrinking N33 and pumping clocks (up to ~3264/24000 at stock) would appear a reasonable and cost-efficient thing to do, could compete with products below AD104, perhaps at a size similar to AD107 (which N33 already beats but at a larger size, granted on an older [then cheaper, but 4nm is now also cheap(er)] process). Shrinking from 6nm->4nm could/should increase density by ~40% (44% by my linear math, but I don't know exactly how much more space the lesser-scaling parts like cache would take up, nor do I know how much space arch changes/increasing clocks would effect it). You're really getting into the weeds when you think about the fact 7nm SRAM cell size is .027µ (no idea on 6nm) and 4nm is ~.0199mm2µ (same size as n3b, n3e is WORSE), while decap (for higher clocks) on rv770->rv790 was ~26mm2 (way back on 55nm).

The processes, clocks, and available ram chips make this very possible, and it's placement makes sense (cheapest option that will get you PS5 performance), and I'd be very surprised if they couldn't make it work with 16GB for <$300.

I personally stopped thinking GDDR7 was plausible a while back as I don't *think* they will be available yet by the time this needs to launch. I *could* be wrong. Samsung has been very quiet about a release date (the only one we have is Micron at the EOY, I assume for Blackwell). It could happen, but I still think 24gbps GDDR6 is more likely due to the fact literally no other products have used it (it should be much cheaper) and it fits rather well with a small update to their current parts. Also, if you do the math, 24gbps is actually a better fit. The ram bandwidth would be a wash (or perhaps overclock slightly better), but unless AMD doubled up their L3 cache it would limit bandwidth more quickly (16MB of L3 constitues about 8.5% of bw perf if such a part is 2720/20000). It just sounds more expensive for very little gain...maybe ~5% and that's IF it could hit ~3700mhz, while losing 4GB of buffer. Not a trade I would make.

Likewise, doubling that up is very much akin to what AMD has done in the past. IMHO the actual shader count is likely to be close to 7800xt as that part is clocked at 2425mhz for a reason (part of which may be to make a similar part with higher clocks look like a generational leap), likely to look like a faster PC alternative to the PS5 pro for similarish (and eventually within it's lifetime) less money, which is something the PC gaming market really needs to do if it wants to survive.

8192*3200/7680*2425 = +40% (not counting whatever architectural enhancements may occur).

These just make the most sense to me. We know that 4nm(P?) can yield well at 3200mhz (base clock of Zen4c, typical operational speed of M2 is 3200-3300mhz), we also have seen it scale to 3500mhz (M2) and 3600-3700mhz (M2 Max/Ultra, Zen4c max boost). To me, it appears like the process is most comfortable around 3200-3500mhz mark, which fits perfectly with 24Gbps GDDR6 (including overclocking to ~25.6gbps) and similar L3 cache size to old parts.

This is bang-on to what TSMC claimed about n4p (6% greater density, 11% greater performance). Most (good yield) 5nm GPUs can hit 2900mhz at 1.08v (or lower). I don't see how ~3200mhz @ 1.2v would be a problem (with conceivable headroom to ~3500mhz), nor how overclocking to the above would be out of the question. If you do the math, (even with overclocking) it would appear to be able fit within 375w (or parts with 1x8-pin on the low-end and 2x8-pin on the high-end). Some of you may say "But that's the power draw of a 4080 Super for conceivably slightly less performance". That's very true, but it could also conceivably be the size of AD104, hence a similar MSRP and beat it's actual competition (4070 super), and conceivably tie (and/or beat in raster) 4070 Ti Super. Essentially this part would make a lot more sense than discounting 7900xt into the ground, likely past the point they would ever want to if they still wanted to make any margains. Likewise, I believe a 16GB (stronger than) PS5-like card would go down very well if under $300 (probably less than a discounted PS5), which just might not be possible for navi 33 due to it's size. It might very-well be possible due to 4nm lowering in cost (and more wafer allocation being avaiable) recently (which it certainly has due to Apple/nVIDIA moving to 3nm) as proven by Sony/AMD moving to 4nm, which they wouldn't do unless it was cost-effective.

I also agree that displacing 7900xtx is a tough thing to believe, which is why I think cancelling N4(c?), the chiplet 9 SED (17280sp?) design, was such a difficult decision on AMD's part, as one can imagine how that could be competative. They could always refresh the current chips with 24gbps memory, and we know they are capable of greater than 3000mhz (which is where a refresh/faster memory begins to make sense for N31), but they indeed would consume a ton of power and probably still not catch up to the 4090 (which nVIDIA themselves could then [and may still] refresh). I suppose if they want to make an overtly compelling 4080 Super competition they could, but to what end when Blackwell isn't too far away and the lowest-end "enthusiast" chip in that lineup could very well match a 7900xtx in raster (my best guess at this point is 192-bit/18GB/9216sp @ 3375/32000). AMD might have a slight advantage over-all conceivably in raster at the same price if you factor in overclocking, but that's where the arguement of power consumption(DLSS/RT) really could become a deciding factor. There's always the possibility of a 4nm respin, but I'm not holding my breath. Would love to be wrong as more options and competition is good for everyone.

Point is, the demarkation line (in my mind) is the PS5 and the PS5pro. To put it in more-simple terms, a 7600xt (which is old process tech and still over the golden $300) or overclocked 7800xt (OT: which some people won't do or consider when looking at review charts...even though you'll likely get close to a 20% gain vs a dirt-cheap MSRP 2425mhz stock model and does become a compelling option versus a 4070 Super, especially given the extra ram. /OT), so really they need a cheaper 7600xt and 7900xt (the latter of which is the gateway to high-end PC gaming). They need stock competition for the 4060 Ti 16GB (preferably under the magic $300 mark, which 7600xt is not and nVIDIA would likely never challenge UNLESS they release a point 4060 Ti Super with 4608sp and relegate the old part lower to compete; both hypothetical super and conceivable '8600xt' would probably OC to a similar level in raster) and 4070 Ti Super (preferably for the price of a 4070 Super or less; 7800xt can't hit that perf level and 7900xt costs too much to make). Those would be that thing. They might use a bit more power or run warmer (albeit most AIB coolers are plenty strong), but would make financial sense (wrt die size/cost) versus a chip one slot down in nVIDIA's lineup, which is typically where AMD likes to play. They've always done that: they price themselves against their weakest aspect versus nVIDIA. In this case it has been RT (by ~20%), which again I find amusing because not only do AMD's chips have stronger raster (especially after adding the delta from overclocking), nVIDIA outdates their last gen wrt RT with every new gen (while raster/RAM springs eternal for actual playability at a given resolution). Seriously, look at the playability of a 2000/3000 series in RT with similar raster performance (that's right, the old cards are largely unplayable)...but that's the market. Win for value if you don't constantly update for RT or to get back the missing ram nVIDIA took from you and that you eventually needed.

But hey, let's see how it pans out. AMD has disappointed before, but they also usually have at least one part that's a diamond wrt $/perf (especially if you overclock).

MakaveliAgreed $1000 is my upper limit for a GPU.

And me selling my 6800XT for $600 and buying the 7900XTX for $1200 so cost me $600 for the upgrade was why I did it.

If I had to start from scratch with no gpu to sell I probably won't have done it.

Actually, my limit is closer to 500, which can be stretched if the value is there, but I get what you mean. :)

There is 0% chance for RX7900XTX performance at $400.
Although Leakers said only Low-end Navi 43 and 44 for RDNA4, the timing is perfect for a monolithic 4nm Navi 42 part (4nm is probably the last node offering adequate size reduction for cache (1.35X) , it offers the potential for 1.15X frequency jump vs 7nm and by Q4 2024 it will be a mature process offering great yields.Even if RDNA4 is a small iteration focusing only on reducing size in ISO process and extracting a small bump in performance per SE for example (like only 5%) a 4nm 256bit 4SE design with128RBE and 80CU and coupled with GDDR7 (+64MB infinity cache) it will be able to match 7900XTX.It will be 24GB also due to the new GDDR7 increased capacity.This will probably be at $599 offering around +20% raster performance and +6GB vs a RTX 5070 at the same price.The main reasons they don't go for higher-tier models is because the are left behind in features and the vast majority buying $600 and above they will always go Nvidia if AMD offers only 10% more raster performance/$, forcing AMD to these tiers to offer 15-20% more performance/$ + increased memory capacity in order to have a chance making the whole endeavor not very desirable financially for AMD.Also in 2024 we will see increased capacities in memory sizes, for example a 256bit design with 24GB GDDR7 and probably by Q4 2026 32GB, so these 4 years (especially if PS5 Pro has the same 16GB capacity and depending when Xbox next will launch) the memory advantage AMD will offer per tier will not be so important for market perception and sales anymore.Also a OC 7900XTX is at the edge to not be considered CPU limited (4090 it is) so the percentage that they can increase GPU performance with a Zen5X3D is very specific and uninspiring (if it is CPU limited at the same level as a 7900XTX was at Q4 2022). Actually I don't see AMD returning to 384bit designs before Q4 2028 at the earliest (if ever...)