AMD Radeon RX 9000 series GPUs

[Lange_666]

Prices at my local PC shop: cards that are in stock

RX 9070: 799 € - 849 €
RX 9070 XT: 899 € - 999 €

RX 7900 XTX: 1049 € - 1149 €

 

NVidia:
5090: 4999 €  (however, another store has them listed from 2609 € to 3199 € but nothing in stock)

5080: 1899 € - 2499 € (however, another store has them listed from 1239 € to 1889 € but nothing in stock)
5070Ti: 1499 €
5070: 999 €

[Qcumber]

18 minutes ago, Lange_666 said:

5070Ti: 1499 €

Ouch!!!

[Aapje]

12 hours ago, LucShep said:

likely having a 24GB version later - which (as I said), if happening, will be again a sh!tshow with availability and ludicrous street price.

You have completely failed to give a reasonable argument why it is likely that we'll see long term pricing of 2500-3000 euros for such a card (which is what you claimed) and poor availability in the long term.

This is how things went for an MSRP model of the 4080 Super:

null

So as you can see, we had an inflated early price, then apparently a lack of (sufficient) new shipments for a while causing the prices to spike, and then when a big new shipment came in, the prices went down a lot. After two months, the cheapest shop(s) hit MSRP, and then the pricing stayed very flat (the slight price changes are probably mostly due to exchange rate changes) and there was plenty of stock until the product was pulled off the market and the last remaining units spiked in price. So we had something like 9 months of perfectly fine pricing and availability.

And at no point did the price spike anywhere close to 2500/3000.

So on the one hand, we have you pulling these big numbers out of air, with absolutely no explanation of how you got to these numbers, and then on the other hand, there is me, who is actually providing evidence.

And the reason why I'm jumping down your throat on this, is because this kind of 'sky is falling'-narrative just encourages people to give into the FOMO and order overpriced cards, when they already have a decent card and they would save quite a bit by just waiting out that early volatility.

That said, we seem to have a lot more volatility for the 50-series right now due to Nvidia messing some things up, but unless the chips are fatally broken and they need to respin them, it should be solvable in a decent timeframe. And the current issues absolutely cannot be assumed to affect a potential Super-refresh.

12 hours ago, LucShep said:

Same thing may, or not, happen with 3GB VRAM modules, that's what I meant. You don't know, and I don't know.

 

We absolutely do know that these modules have already been developed and are being tested by GPU makers:

https://news.samsung.com/global/samsung-develops-industrys-first-24gb-gddr7-dram-for-next-generation-ai-computing

And you need to keep in mind that in a market with just a few potential buyers of GDDR-modules, the manufacturers don't spend a lots of money on new tech for the LOLs. They align their roadmaps, to make sure that they don't develop tech and built new production lines for nothing.

The main uncertainty is when these things come to market, not so much whether they will.

12 hours ago, LucShep said:

Even if basing on a roadmap ilustration from a memory manufacturer, it doesn't mean that anything substancial will happen VRAM amount wise for gaming market GPUs.

It's just common sense for them to want to use this, since they are now clearly facing a lot of customers who refuse to buy 8 GB cards, which is why they released clamshell versions of the 4060 Ti and RX 7600. A 12 GB low end card fits in the line up much better, and is cheaper than a 16 GB card.

A proper lineup has gradually increasing specs across the line up, but both AMD and Nvidia went from from 16 GB to 12 GB in their lineup, when going to the x070/x700 cards. And we already see that 12 GB cards are choking on Indiana Jones and the Great Circle, which has mandatory raytracing, so they simply can't keep putting 12 GB on the mid-tier x070 cards.

The only option to give people sufficient memory without increasing the bus size, which they clearly do not want, is to switch to the 3 GB modules. 

12 hours ago, LucShep said:

It's Nvidia, AMD and Intel who decides how (and how many of) those VRAM modules are used for X and Y model and market segment. Not the memory makers.
More than likely what we'll end up seeing is 18GB VRAM (6x 3GB modules) instead of 16GB VRAM (8x 2GB modules).

That would mean a further decrease in bus size, and Nvidia is already pretty low. So I think that this is unlikely.

12 hours ago, LucShep said:

The general consensus is that there is no need for more than 16GB VRAM for PC gaming, not for the time expectation of this generation (as in, for at least 2 years).

It's not just about what is needed, but to create a line up that is good at upselling. I think that this makes a lot of sense:

x060: 12 GB (and probably also a slightly cheaper 8 GB version)

x070: 18 GB

x080: 24 GB

x090: 32 GB

Remember how Intel used to focus a lot on pushed the Mhz's as high as they could, because the common people didn't understand IPC, so they just would go: 'bigger number better'

This line up would be great at clearly distinguishing the tiers based on VRAM.

12 hours ago, LucShep said:

Have you noticed how much the GPUs have exponentially increased price in just five years?

No, because it simply is not true.

12 hours ago, LucShep said:

Prices barely decrease during the whole lifespan of a GPU generation (rare exceptions to some Intel and AMD GPUs that simply don't sell), the opposite actually happening in this last generation.

You mean the 4080 Super being cheaper than the 4080?

12 hours ago, LucShep said:

Yes, it'll reach a point (more than today) when people simply can't keep up, and many (I'd say most?) pass on it and will be left with obsolete GPUs, or resignate into buying lower segment than they would have years before.

The more the improvement in price/performance stagnates, the slower GPUs will become obsolete. So it makes sense for the market to then transition to higher pricing (since you can use the GPU for longer), with much lower sales volume.

Ultimately, this in large part a technological issue. If the days of big 'free' improvements per generation are over, then we will have to adapt to that new reality. A large part of the anger is because people feel entitled to big gains for the same money every X years.

[LucShep]

5 hours ago, Aapje said:

You have completely failed to give a reasonable argument why it is likely that we'll see long term pricing of 2500-3000 euros for such a card (which is what you claimed) and poor availability in the long term.

This is how things went for an MSRP model of the 4080 Super:

null

So as you can see

.....what I see is that you don't know what you're talking about.
The RTX4080S is actually the worst example you could find to defend your nonsense argument, because that model came out later exactly to disguise the fact that the original RTX4080 was a monumental failure from Nvidia - it was left abandoned and ignored on the shelves, while the RTX4090 imediately sold out, scalping and hording galore (its price still overly inflated today!), inclusively the RTX4070 was outselling it by some margin. 

That's why the RTX4080S came out later with such a considerably lower MSRP, just to entice sales numbers.
But then the novelty interest of 4000 series had already ran out. While it sold "decently", it was never that great, not really a huge success - more like the "oh well, can't reach the 4090 I guess I'll go to that" usual pick of the list.
Hence maintaining a reasonable stability in price, never inflating to the levels you see, for instances, for the 5080 now. Though the RTX4080S price did actually increase in last months, prior and during the RTX5000 series launch. 

And, FWIW.....

 

5 hours ago, Aapje said:

We absolutely do know that these modules have already been developed and are being tested by GPU makers:

https://news.samsung.com/global/samsung-develops-industrys-first-24gb-gddr7-dram-for-next-generation-ai-computing

And you need to keep in mind that in a market with just a few potential buyers of GDDR-modules, the manufacturers don't spend a lots of money on new tech for the LOLs. They align their roadmaps, to make sure that they don't develop tech and built new production lines for nothing.

The main uncertainty is when these things come to market, not so much whether they will.

It's just common sense for them to want to use this
 

It has NOTHING to do with common sense, it has all to do with bottom line for manufacturers, what gives them most profit for least cost.
You're oversimplifying and underestimating the necessary measures and costs to take GDDR7 to levels that make GDDR6 obsolete for gaming GPUs (not for AI computing / data center!!), which it simply doesn't atm. And those 3GB modules will be GDDR7.   Here are some points you're not taking into account: 

• Hardware Costs: GDDR7 memory modules are more expensive than GDDR6/X modules, primarily due to the higher bandwidth and power efficiency requirements, which leads to even higher priced gaming GPUs than they already were/are.
   
• Power Consumption: While GDDR7 is designed to be more power-efficient, the initial investment in GDDR7-based hardware is higher due to the cost of the more advanced memory technology, and already the case for Nvidia 5000 series.
   
• System Design: The design of the system architecture also impacts the cost of using GDDR7 memory. For example, the need for more advanced cooling on GPUs or higher-speed interconnects adds to the already higher overall cost, in a market already cursed with inflated prices.
   
• Higher Working Temperatures: While bandwith is a bit larger, GDDR7 speeds are (for now) necessarily capped due to higher temperatures, to equivalent speeds in practice to what is already in use for GDDR6/X, and as the bandwidth was already eficient for gaming before, it makes GDDR6/X still competitive enough while being more cost effective for this end use (gaming).

RTX5000 already uses it, but how much of its silly high price is due to GDDR7 adoption?
It's not a "lets just slap some mem modules in the PCB and go home", and not something to rave about for increased VRAM gaming GPUs either.
It increases production cost and final price for gaming GPUs, even more, for very little benefit (none for now) - when GPUs already were/are at used car prices.
 

5 hours ago, Aapje said:

18 hours ago, LucShep said:

Have you noticed how much the GPUs have exponentially increased price in just five years? This isn't "almost certainly temporary" (your words, not mine).

No, because it simply is not true.

LOL "okay bro"....  
You must live on another planet then?
If you don't see increased price in GPUs from 2019/2020 to 2024/2025, gen to gen, I think every person in this forum board will be surprised.

Anyway, you like to look for charts and numbers, go do it, from one generation to another of Nvidia 2000, to 3000, to 4000, to 5000 series.
Go research, for example, how much the street-price (not MSRP bullsh!t!) of an RTX2060, RTX2070, RTX2080 (Super or not) was before RTX3000 series replacements came out and how much those then went for. And then, even after the pandemic and mining craze ended (the excuse then for crazy prices), how the RTX4000 series and now RTX5000 series kept such prices up. 
And same for AMD with the RX5700/XT versus RX6700/XT, RX7700/XT and upcoming RX9060/XT.
 

5 hours ago, Aapje said:

Ultimately, this in large part a technological issue. If the days of big 'free' improvements per generation are over, then we will have to adapt to that new reality. A large part of the anger is because people feel entitled to big gains for the same money every X years.

I agree there.
The problem is when prices are so much higher (for whatever reasons we may discuss) but performance uplifts are not significant enough.
It's discouraging, for both the users/buyers to invest in new hardware, and the PC gaming developers to do better.
PC gaming has always had its highs and lows, but it's not in a happy/healthy place right now. Since the economy and geopolitical scene is at a weird spot, it'll depend on the most clever, innovative and risk-taking type of people to push things further, for the better and cheaper.
 

Edited March 7 by LucShep

[Qcumber]

This is getting very off topic. 

[Aapje]

@LucShep

Quote

That's why the RTX4080S came out later with such a considerably lower MSRP, just to entice sales numbers.

So instead of admitting that I am right and you are wrong, you are simply stating reasons why the MSRP went down (which goes against your claims) and that there was plenty of supply for a long time (same).

I see a chronic inability to simply admit that you made false claims, and instead you make all kinds of statements that actually prove me right, but then pretend that they prove you right, when they don't do so at all.

And your mostly offtopic rant about GDDR7 still does not actually support your own argument. Despite all the negatives you list, it was adopted, so why wouldn't the 3 GB modules be adopted then, especially as the downsides of GDDR7 will get less over time?

And you are simply telling falsehoods about the pricing. You claimed an exponential price increase, and this is simply false. Untrue. You are saying things that are not correct. No matter how you try to spin it, MSRP or street pricing, there simply is no exponential curve that you can draw of the pricing, especially if you factor in inflation.

There is little point in having a discussion if you keep telling these falsehoods and refuse to admit to the truth when challenged, but start moving the goal posts.

Edited March 7 by Aapje

[LucShep]

50 minutes ago, Aapje said:

@LucShep

So instead of admitting that I am right and you are wrong, you are simply stating reasons why the MSRP went down (which goes against your claims) and that there was plenty of supply for a long time (same).

I see a chronic inability to simply admit that you made false claims, and instead you make all kinds of statements that actually prove me right, but then pretend that they prove you right, when they don't do so at all.

And your mostly offtopic rant about GDDR7 still does not actually support your own argument. Despite all the negatives you list, it was adopted, so why wouldn't the 3 GB modules be adopted then, especially as the downsides of GDDR7 will get less over time?

And you are simply telling falsehoods about the pricing. You claimed an exponential price increase, and this is simply false. Untrue. You are saying things that are not correct. No matter how you try to spin it, MSRP or street pricing, there simply is no exponential curve that you can draw of the pricing, especially if you factor in inflation.

There is little point in having a discussion if you keep telling these falsehoods and refuse to admit to the truth when challenged, but start moving the goal posts.

 lol what? 

 ...okey dokey, bye 

[Lange_666]

Jeez dudes, get a life and fight it out in a PM chat.
Nobody, but really NOBODY is interested in this.

Moderators, isn't it time to put a lid on this topic?

[Blackhawk163]

Aapje at it again?  You seem to be the main character of all these debates, and in internet logo, that's not a good thing. I'm sorry mate but if you're not arguing with kksnowbear then you're doing it again with someone else. 

 

Best to move on.

 

 

And here I was hoping we had users with actual benchmarks. Sigh. 

[LucShep]

11 hours ago, Blackhawk163 said:

And here I was hoping we had users with actual benchmarks. Sigh. 
 

Yes, I think we all would like to know how the RX9070/XT go with DCS in 2D 1440P/4K and also VR. And how it compares to Nvidia's counterparts.

My guess is that very few people playing DCS managed to get a RX9070/XT and, of those, they probably won't be looking into these forums nor posting benchmarks.

Now with prices shooting up (see videos below about it), I guess we'll have to wait a little longer for that...
 

 

Edited March 8 by LucShep

[LucShep]

In case you missed it, this was posted on the VR section - RX9070XT running DCS in VR (Quest3):
 

[Blackhawk163]

2 hours ago, LucShep said:

In case you missed it, this was posted on the VR section - RX9070XT running DCS in VR (Quest3):
 

Saw that. Guess my 4080s will be sticking around much longer, maybe until the 6000 series as the prices on the 5090's are freaking insane especially given that the AIB's are in on scalping their own customers now it seems. What a pathetic launch. Imagine needing a new mid-upper range graphics cards these days. 

[LucShep]

5 minutes ago, Blackhawk163 said:

Saw that. Guess my 4080s will be sticking around much longer, maybe until the 6000 series as the prices on the 5090's are freaking insane especially given that the AIB's are in on scalping their own customers now it seems. What a pathetic launch. Imagine needing a new mid-upper range graphics cards these days. 

hmmmmmm let me prophesize RTX6000 series for you...? 

RTX6090:    MSRP 1999$ - street price 4500$
RTX6080:    MSRP 1299$ - street price 2500$
RTX6070Ti: MSRP 1049$ - street price 1600$
RTX6070:    MSRP 849$ - street price 1200$

"Impossible without AI..." The more you buy, the more you save".... 

[Pilotasso]

I dont think the prices will continue to raise linearly as they had in past years. Heres a few reasons why:

1- GPU crytpo mining is a thing of the past, plus we are going to a bear market sometime this year and that will last at least 18 months.

2- Realy poor condition of global economy, the average joe losing income and jobs, has no choice but to cut expenses, gaming being one of them. The youger generations give gaming more importance but putting supper on the table is going to be the priority. This could get worse due to major conlicts arising in developed countries (and this is going to throw a whole lot of new variables unfortunatly).

3- Deepseek, and the AI revolution becoming cheaper to attain AI. NVIDIA has a disporportional ammount of FAB allocation to the AI server clients. what hapens when they no longer need to buy as many chips? NVIDIA has no choice but to put those chips on the consumer market. They have to be aggressive as the comsumers are leaving markets.

4- AMD is right to cater to the budget gamer, by leveraging the console tech for PC gaming. This will normalise the hardware requirements you need. They are not there yet, as the current gen GPU's IMHO dont have enough VRAM (neither NVIDIA's "affordable" SKU's for that matter).

5- More FABs outside Taiwan in comming years.

Edited March 9 by Pilotasso

[EightyDuce]

1 hour ago, Pilotasso said:

I dont think the prices will continue to raise linearly as they had in past years. Heres a few reasons why:

1- GPU crytpo mining is a thing of the past, plus we are going to a bear market sometime this year and that will last at least 18 months.

2- Realy poor condition of global economy, the average joe losing income and jobs, has no choice but to cut expenses, gaming being one of them. The youger generations give gaming more importance but putting supper on the table is going to be the priority. This could get worse due to major conlicts arising in developed countries (and this is going to throw a whole lot of new variables unfortunatly).

3- Deepseek, and the AI revolution becoming cheaper to attain AI. NVIDIA has a disporportional ammount of FAB allocation to the AI server clients. what hapens when they no longer need to buy as many chips? NVIDIA has no choice but to put those chips on the consumer market. They have to be aggressive as the comsumers are leaving markets.

4- AMD is right to cater to the budget gamer, by leveraging the console tech for PC gaming. This will normalise the hardware requirements you need. They are not there yet, as the current gen GPU's IMHO dont have enough VRAM (neither NVIDIA's "affordable" SKU's for that matter).

5- More FABs outside Taiwan in comming years.

While crypto isn't as much of a demand as it previously was, AI craze isn't going anywhere anytime soon and demand for hardware is only going to increase. By far the largest bottleneck is TMSC and its ability to produce sufficient wafers to satisfy all orders. There are a ton of things that utilize the same or variation of the same manufacturing node and the majority is more than likely being funneled into enterprise/server/AI hardware as that's where your biggest ROI is. Gaming hardware at this point accounts for a small fraction of sales while taking a disproportionally large chunk of silicon to manufacture. For example, lets take the 4090: it has a 609mm^2 die the same as the Quadro RTX6000. 4090 MSRP is $1,699...Quadro RTX 6000 MSRP is $6,799. Even at inflated street price, they would have to sell 3 4090's to equal a single RTX 6000 sale. While the 4090s would easily sell, you could take that silicon and make 3 RTX 6000's and easily sell them for $20,000. Yes, there is some margin for the silicon that doesn't meet the spec for an RTX 6000 so it get's "cut down" to a 4090, but overall the math should be close enough as I don't think every 4090 is a failed RTX 6000, but rather an intentionally cut down "good" RTX 6000 die. This is unlikely to change with Blackwell. Same goes for AMD and their MI series.

But I'm not an economist or a hardware manufacturer and have zero inside knowledge of the inner-workings of this system. Just doing some napkin math while on the outside looking in.

[LucShep]

1 hour ago, EightyDuce said:

While crypto isn't as much of a demand as it previously was, AI craze isn't going anywhere anytime soon and demand for hardware is only going to increase. By far the largest bottleneck is TMSC and its ability to produce sufficient wafers to satisfy all orders. There are a ton of things that utilize the same or variation of the same manufacturing node and the majority is more than likely being funneled into enterprise/server/AI hardware as that's where your biggest ROI is. Gaming hardware at this point accounts for a small fraction of sales while taking a disproportionally large chunk of silicon to manufacture. For example, lets take the 4090: it has a 609mm^2 die the same as the Quadro RTX6000. 4090 MSRP is $1,699...Quadro RTX 6000 MSRP is $6,799. Even at inflated street price, they would have to sell 3 4090's to equal a single RTX 6000 sale. While the 4090s would easily sell, you could take that silicon and make 3 RTX 6000's and easily sell them for $20,000. Yes, there is some margin for the silicon that doesn't meet the spec for an RTX 6000 so it get's "cut down" to a 4090, but overall the math should be close enough as I don't think every 4090 is a failed RTX 6000, but rather an intentionally cut down "good" RTX 6000 die. This is unlikely to change with Blackwell. Same goes for AMD and their MI series.

But I'm not an economist or a hardware manufacturer and have zero inside knowledge of the inner-workings of this system. Just doing some napkin math while on the outside looking in.

Yep! 
And TMSC also does chips not just for Nvidia, but also for CPUs and GPUs of AMD, and also Intel.
Being the world's most valuable semiconductor company, producing for (pretty much) all major companies in the market, has some major drawbacks, especially for the end-user.

Basically it's now like a bottleneck, they can't produce fast enough and cheap enough.
And what can deliver highest margin once in the market is given priority over the rest (and gaming GPUs aren't it) by order of each manufacturer - the customers of TSMC.
What we need right now is more manufacturing capacity, but alas....

Buildzoid recently posted a video about this:

 

Edited March 9 by LucShep

[Qcumber]

Base prices are going up. At Overclockers the "out of stock" price of MSRP Rx 9070xt cards has risen from £569.99 (6/3/25) to £629.99 (7/3/25) to £649.99 (10/3/25). I would not be surprised if the "new" MSRP target is actually £669.99 or higher. 

[Qcumber]

The 9070xt does not look promising for VR at the moment.

 

 

[Aapje]

3 hours ago, Qcumber said:

The 9070xt does not look promising for VR at the moment.

That's a pretty generic statement when it's really only headset that seems to have issues right now. Pimax says that they are working on fixing it.

It's probably fixed by the time you can get the 9070 XT for MSRP (seems like a pretty safe bet  )

 

PS. I don't really understand the logic from the video where he is claiming that Pimax will take a long time to fix it, because Valve didn't fix something. Does he not understand that these are two different companies?

Edited March 14 by Aapje

[Qcumber]

Updated driver and results for 9070xt. Not tested in DCS yet but looks like a big improvement for VR. 

 

[Aapje]

In the past AMD took a long time to fix VR, so if they fix it so soon, that is a good sign. In a way it seems that AMD and Nvidia switched places when it comes to driver development.

[BitMaster]

When I zoom far out, it seems to me that Nvidia is letting loose on the Gamers and AMD is willing and more and more capable of catching them.

AMD is gaining momentum on all fronts, including relevance in the most profitable price range of GPU's, 500-1000€.

 

[Qcumber]

10 minutes ago, BitMaster said:

When I zoom far out, it seems to me that Nvidia is letting loose on the Gamers and AMD is willing and more and more capable of catching them.

AMD is gaining momentum on all fronts, including relevance in the most profitable price range of GPU's, 500-1000€.

 

I like what AMD are trying to do, but the recent increase in MSRP of the Rx 9070xt is a bit frustrating. Especially as it is very difficult to buy anything even at that new price. I am hoping that over the next few weeks the stock levels will make this card more accessible. They are certainly being more proactive than NVIDIA. 

[LucShep]

21 hours ago, BitMaster said:

When I zoom far out, it seems to me that Nvidia is letting loose on the Gamers and AMD is willing and more and more capable of catching them.

AMD is gaining momentum on all fronts, including relevance in the most profitable price range of GPU's, 500-1000€.

 

Agreed, I think we may be witnessing a slow and gradual shift in the gaming GPU arena, perhaps for next years to come. 

Similar to what has happened with CPUs right before the 2020 time period, when AMD Ryzen then became a true alternative to Intel "K" processors at very competitive prices.

The globally inflated prices and low availability, and also constant drivers development, are factors that can favor AMD GPUs right now - even if prices are bad, at the moment.
 

Edited March 23 by LucShep

[Aapje]

Yeah, it looks like a perfect storm of immense hubris, failure to execute and AMD closing the gap (FSR 4 mainly). If one of these things wouldn't have been true, customers would probably not think of switching en mass.