Report: AMD Radeon RX 6700/XT will arrive at end of March

[[GSV]Trig]

I would have liked to go higher than my old RX580's 8Gb, but getting an AMD card was too hard so settled with the 3060Ti, still 8Gb but more than fast enough for the small amount of gaming I do, the problem is I'm now looking at spending £4/500 on a monitor...

[DanceswithUnix]

Well, I doubt it makes sense as a business as consoles are by far the least profitable thing AMD could do with those scares TSMC wafers. The wafer price/mm order must be something like this:

1. server CPUs,
2. desktop CPUs,
3. laptop CPUs,
4. distantly followed by GPUs
5. with console chips bringing up the even more distant rear.

It just seems that Microsoft and Sony have a very good deal where they get priority.

Given how profitable Nvidia have been in recent years selling GPUs, I suspect they should be higher up the list.

But it takes a lot of months from order for tested and packaged chips to come out of the factories, I'm glad I'm not the one having to decide what to order when.

[Luke7]

Can I just ask... why is everyone so upset about them not being available? The only reason is hype, surely? If you're desperate for a new GPU, well the prices of the last gen are sure to drop hard shortly. And they aren't bad cards. Otherwise, is your old GPU that bad? Is it suddenly unable to play games just as well as it did yesterday?

It feels as though people have bought in fully to the "you don't just want this, you NEEEED it and your old card is so... old" marketing. They have got very good at it. They have stoked up demand and then released limited supply, so they are in control of pricing and what card you buy. Remember, even big name companies have been caught scalping when furiously denouncing it.

The world hasn't changed, games haven't changed. The availability will improve and prices will settle.

"AMD don't have it, I'm buying Nvidia" - okay, but you'll pay a healthy premium, and you could just sit back and wait for a few months. Your old card will still work just fine. Watch the chaos, the scrabbling and when it all ends buy the card you want when YOU have the power over the companies. At the moment, demand exceeds supply and you have to get what you can get at the price you can get it.

In a few months, the companies will have to earn your custom, rather than having people literally begging them for their product. That makes for far better value and you can choose exactly what you want or need.

This comment makes far too many assumptions. I've *just* upgraded from an A10-7870k to a 5600X - I'm currently borrowing a GTX 960 which, whilst a shed load better than my old APU, is still pretty naff at playing games even from 2015 (low settings for Witcher 3) and is quite the bottleneck for my shiny CPU.

As for people wanting to upgrade... did you miss the RTX launch? They weren't widely considered worth it, so many people didn't upgrade which is one of the reasons for high demand this time around. As for those of us that hang on to hardware for many, many years, well yes, our GPUs simply don't allow for modern titles to be played nicely at all.

And whilst some of the disappointment with the delay may be down to "hype", we know for a fact that its going to be a better product that the 5700/5700 XT which, as you mention will come down in price, but in actuality, haven't yet. Why on earth would anyone spend X amount on what is effectively as last-gen GPU when there's a better performer (that we expected to be launched sooner) rolling out? Unless you've got so much money at your disposal that you simply don't care about depreciation on such a short timescale (something your comment suggests about yourself considering you seem somewhat oblivious to the masses of us with underpowered GPUs), it simply doesn't make sense to buy 5700/5700 XT now - and no one in their right mind would advise doing so until the prices drop significantly (which they haven't yet).

So yes, we're rightfully disappointed that the expected wait time for the 6700/6700 XT launch has increased - I don't understand why anyone would have difficulty in understanding why people would be "upset" (far too strong a word) by that.

[edmundhonda]

If you're desperate for a new GPU, well the prices of the last gen are sure to drop hard shortly. And they aren't bad cards.

From PCPartpicker: cheapest Turing card available is a 2060 non-super for £470; 2080 Super starts at £900, cheapest RX5700 XT is £550, even the RX580 starts at £230. Given that none of these cards are in major production, how cheap do you expect them to get?

[HW90]

I’d love to know the reasoning for this... I’ve heard rumours of AMD refocusing on console supply. Makes sense as a business I suppose. Not good for us though.

Well, I doubt it makes sense as a business as consoles are by far the least profitable thing AMD could do with those scares TSMC wafers. The wafer price/mm order must be something like this:

1. server CPUs,
2. desktop CPUs,
3. laptop CPUs,
4. distantly followed by GPUs
5. with console chips bringing up the even more distant rear.

It just seems that Microsoft and Sony have a very good deal where they get priority.

It's less profitable, but it's a large amount of money which significantly subsidises overall R&D costs and probably gets them some extra manufacturing discounts at TSMC so that's a non-issue because by making less profit or even taking a loss on the consoles, they can make more profit elsewhere.

[Iota]

They can't squander it there are no chips or spare capacity. They can't make enough chips yet according to some reports have more market share than Intel in the CPU market...
It's really a no win thing. All of their eggs came together to be launched in a timeframe that was never going to work with a global pandemic

Pandemic, console launch, new CPU launch, new GPU launch. Pandemic aside, it's a fairly decent mix for saturating booked capacity that would be challenging to predict in advance.

Can I just ask... why is everyone so upset about them not being available? The only reason is hype, surely?

Blame marketing. They've done a great job, everyone must have all of the shiny things now.

So yes, we're rightfully disappointed that the expected wait time for the 6700/6700 XT launch has increased - I don't understand why anyone would have difficulty in understanding why people would be "upset" (far too strong a word) by that.

It really is a question of patience. I certainly couldn't get a 3090 FE at launch date and had to wait a few months before it became available. My 2080 FE sufficed in the meantime, it's not like it stopped working overnight.

[badass]

Can I just ask... why is everyone so upset about them not being available? The only reason is hype, surely? If you're desperate for a new GPU, well the prices of the last gen are sure to drop hard shortly

Last gen prices have go up by a lot as well. In short, anyone that wants any kind of graphics card now has to pay through the nose. The price of ethereum has ensured that. Even an ancient RX 580 is pretty profitable mining right now.

In a few months, the companies will have to earn your custom, rather than having people literally begging them for their product. That makes for far better value and you can choose exactly what you want or need.

I agree with the sentiment but it depends on your definition of "a few months". I doubt prices will settle until q3 at least.

[watercooled]

Definitely the console supply.

That is likely what's eating a lot of AMD's allocation, assuming the console orders aren't allocated separately by TSMC. Console processors are very much front-loaded in terms of supply with huge orders initially then tailing off somewhat after launch is out of the way.

I think they also have an understanding that while the 6xxx series are more competitive than previous series, they're not as competitive as they need to be, so they're better off focusing their efforts on putting out the console chips where that's not an issue and then putting more money into R&D for RX 7xxx to get the raytracing performance at a more competitive level and introduce chiplet GPUs to get costs down.

Not sure I agree with you there. They're more than competitive and are significantly more efficient (that was apparently critically important when Nvidia were leading there, so I don't see that changing now). Not everyone is interested in RT or upscaling/sharpening with a fancy name.

From talking to people I hear it's much simpler than anybody thinks...
There are NO chips to go around because they are being made as fast as they can and still woefully short of fulfilling demand at 7nm. The sheer amount of chips being requested at 7nm is incredible. Shortages will last for years. About the only good reason to go with Samsung for NV

Demand is tight at 7nm but no idea where you pulled the 'years' part from, that's nonsense. Three MAJOR customers are shifting away from TSMC 7nm who between them made up the vast majority of capacity; Apple, Qualcomm and Huawei. The capacity available to AMD will continue to increase, and the demand from consoles will eventually drop because of what I explained in my first reply. If the situation was as dire as you suggested, AMD wouldn't have bothered producing huge GPU dies at TSMC; they would have either sought production elsewhere or just stuck to midrange stuff again.

This is expected. They're selling all their lower CU parts to Sony and Microsoft

No they're not. The consoles use totally separate dies and have absolutely nothing to do with any of AMD's retail products.

Given how profitable Nvidia have been in recent years selling GPUs, I suspect they should be higher up the list.

I'm assuming he literally just meant profit per mm^2 of die sold; CPUs are far smaller and therefore higher sale price per mm^2. For a given wafer allocation, CPUs should be far more profitable than GPUs given you fit far more per wafer, attain higher yields, and still sell at higher margins. Mindshare etc is another matter, but we're talking about a wafer-limited situation.

[watercooled]

Pandemic, console launch, new CPU launch, new GPU launch. Pandemic aside, it's a fairly decent mix for saturating booked capacity that would be challenging to predict in advance.

It really is a perfect storm. It's normal for stuff to sell out on launch (lots of people seem to have either forgotten that or have only just entered the PC market!) whether it's GPUs or CPUs from AMD/Intel/Nvida. Provided what is being sold is remotely competitive for the price of course.

Blame marketing. They've done a great job, everyone must have all of the shiny things now.

Very much this, the hype around these launches was insane! Even the value comparisons are more than a bit bizarre - Ampere does offer something worth upgrading from previous generations but that's nothing to wet yourself over, it should be expected! It seems Turing just set everyone's expectations at rock-bottom because it offered practically zero value improvement over Pascal. About the only card worth 'upgrading' to was the 2080Ti because it was the fastest available, assuming money is no object and you must just absolutely have the latest shiny thing. Every performance band was significantly jacked up in price, and those prices have remained.

It really is a question of patience. I certainly couldn't get a 3090 FE at launch date and had to wait a few months before it became available. My 2080 FE sufficed in the meantime, it's not like it stopped working overnight.

Again, you'd really think everything else did stop working overnight with the panic to buy. I even know a few people who had no intention of upgrading and had never bought a particularly expensive card before, panic buying a 3080 because the scarcity seemingly drives up desirability for a lot of people!

[DanceswithUnix]

That is likely what's eating a lot of AMD's allocation, assuming the console orders aren't allocated separately by TSMC. Console processors are very much front-loaded in terms of supply with huge orders initially then tailing off somewhat after launch is out of the way.

I'm assuming he literally just meant profit per mm^2 of die sold; CPUs are far smaller and therefore higher sale price per mm^2. For a given wafer allocation, CPUs should be far more profitable than GPUs given you fit far more per wafer, attain higher yields, and still sell at higher margins. Mindshare etc is another matter, but we're talking about a wafer-limited situation.

I might have imagined once hearing that Microsoft had their own allocation, AMD just designed the thing and collect royalties. But even in that scenario a fully booked out 7nm process is still stopping AMD from getting more wafer starts, regardless of who it is currently maxing out the fabs.

The GPU profit is an interesting one. Even allowing for the IO die, the quite tiny CPU die on a 5800X should mean a pretty good profit margin. I presume as not every PC has a discrete GPU the overall market for GPUs is also lower. That all makes perfect sense, and yet Nvidia seem to be swimming around in more money than Scrooge McDuck. So it can't be that hand to mouth compared with CPUs

[edmundhonda]

Three MAJOR customers are shifting away from TSMC 7nm who between them made up the vast majority of capacity; Apple

The entire iPhone 12 range and newest iPad Air are all on 5nm, so Apple's contribution to a shift was 6 months ago.

[Iota]

I presume as not every PC has a discrete GPU the overall market for GPUs is also lower. That all makes perfect sense, and yet Nvidia seem to be swimming around in more money than Scrooge McDuck. So it can't be that hand to mouth compared with CPUs

I think Nvidia have pushed into diversifying, it really isn't just about the desktop GPU space now, that doesn't account for all of their business.

[kompukare]

I think they also have an understanding that while the 6xxx series are more competitive than previous series, they're not as competitive as they need to be, so they're better off focusing their efforts on putting out the console chips where that's not an issue and then putting more money into R&D for RX 7xxx to get the raytracing performance at a more competitive level and introduce chiplet GPUs to get costs down.

Not sure I agree with you there. They're more than competitive and are significantly more efficient (that was apparently critically important when Nvidia were leading there, so I don't see that changing now). Not everyone is interested in RT or upscaling/sharpening with a fancy name.

Yes, is all a bit surprising isn't it?
Almost like Nvidia's marketing is so powerful that they set the whole narrative of which metrics users and reviewers should concentrate on.
When they had the lead at perf/watt, that was what everyone talked about, now it has to be something else.

Due to going with Samsung and feeling the need to clock Ampere as high as possible, Nvidia have squandered the perf/watt advantage they had. And probably the perf/area one too although comparing Samsung 8nm to TSMC's 7nm isn't that totally straightforward. The area perf seems to suffered due to spending so much die space on the tensor cores. Which Nvidia really wanted for their highly profitable compute markets...

... So by repositioning those tensor cores looking for a solution they came up with...

The fake-res upscaling over-sharpening trick. The only way to run raytracing at 'high' resolution at an acceptable speed.

However, is 4K with over-sharpened upscaling actually 4K? To think LG got so much slack (rightly IMO) for WRGB, and now Nvidia has everyone praising upscaled fake 4K as the greatest thing ever?

Yes, I'll admit that DLSS is a clever use of tensor cores and a very clever upscaler, but is still an upscaler.

[watercooled]

I might have imagined once hearing that Microsoft had their own allocation, AMD just designed the thing and collect royalties. But even in that scenario a fully booked out 7nm process is still stopping AMD from getting more wafer starts, regardless of who it is currently maxing out the fabs.

The GPU profit is an interesting one. Even allowing for the IO die, the quite tiny CPU die on a 5800X should mean a pretty good profit margin. I presume as not every PC has a discrete GPU the overall market for GPUs is also lower. That all makes perfect sense, and yet Nvidia seem to be swimming around in more money than Scrooge McDuck. So it can't be that hand to mouth compared with CPUs

Yeah you get the same result either way - I was more wondering if the console allocations might have actually *reduced* AMD's own availability, which I presume would happen if the consoles came from their own allocation, or they merely limited AMD's growth by consuming some of the available capacity.

A member of this forum did some back-of-envelope calculations comparing die costs for CPUs and GPUs. Even when comparing against the much smaller 5700XT the difference was huge. Of course there are large error bars in those figures (and it doesn't factor in R&D or anything but raw wafer costs) but it was an interesting comparison. To a point I guess the IO die should help - it's not coming out of their 7nm allocation and I would imagine is quite cheap to produce now.

As Iota said I think a lot of Nvidia's profits come from their enterprise market now, plus the sheer volume of sales they've built up and, more cynically, the prices creeping up for each segment, especially those which have not had direct competition from AMD for some time.

Regarding availability again, it's just occurred to me that Ampere has been out for a full two months longer than Navi 21. I'd actually kinda grouped them all together.

The entire iPhone 12 range and newest iPad Air are all on 5nm, so Apple's contribution to a shift was 6 months ago.

Which is why I mentioned more than Apple, although Apple will still be consuming 7nm wafers - they didn't cease producing existing products overnight. Qualcomm and Huawei were also huge 7nm customers, Huawei leaving for obvious reasons, and Qualcomm transitioning to other nodes; their flagship 888 is to be produced on Samsung's 5nm node so leaving TSMC altogether. Besides, changes don't happen immediately, ramps take months so a customer transitioning does not make all of that capacity available for volume production overnight. Unless I'm mistaken, AMD are set to go from being a fair bit down the list to possibly TSMC's biggest 7nm client, and TSMC are continuing to add capacity to the node.

Availability seems to have somewhat stabilised for Ryzen 5000 CPUs now and all but the 5900X have been in stock for a while now.

I'm not denying wafer capacity is likely tight, but it is kinda a perfect storm - new GPUs, consoles and CPUs all releasing at the same time on the same node, during the middle of a pandemic, right before Christmas. In which case it makes complete sense that Nvidia went to Samsung 8nm otherwise we'd have *another* major product competing for capacity; a difference being that while AMD already had significant 7nm allocation because of their existing products, I expect Nvidia's would have been quite small since their previous generation was on 16/12nm, and they're producing the A100 with the 7nm allocation they do have. And Nvidia typically sell a lot more GPUs than AMD.

And another thing, lots of these new products are demanding GDDR6 and quite a lot of it - the consoles need 16GB each - more than many of the GPUs, and they sell in huge volumes. No doubt the consoles are eating both wafers and DRAM, but like I say they're quite front-heavy in terms of production - *massive* initial orders which then slow over time.

[watercooled]

Yes, is all a bit surprising isn't it?
Almost like Nvidia's marketing is so powerful that they set the whole narrative of which metrics users and reviewers should concentrate on.
When they had the lead at perf/watt, that was what everyone talked about, now it has to be something else.

Due to going with Samsung and feeling the need to clock Ampere as high as possible, Nvidia have squandered the perf/watt advantage they had. And probably the perf/area one too although comparing Samsung 8nm to TSMC's 7nm isn't that totally straightforward. The area perf seems to suffered due to spending so much die space on the tensor cores. Which Nvidia really wanted for their highly profitable compute markets...

... So by repositioning those tensor cores looking for a solution they came up with...

The fake-res upscaling over-sharpening trick. The only way to run raytracing at 'high' resolution at an acceptable speed.

However, is 4K with over-sharpened upscaling actually 4K? To think LG got so much slack (rightly IMO) for WRGB, and now Nvidia has everyone praising upscaled fake 4K as the greatest thing ever?

Yes, I'll admit that DLSS is a clever use of tensor cores and a very clever upscaler, but is still an upscaler.

I believe Samsung's 8nm is an iteration of their 10nm node much like TSMC 16 > 12nm. So there's that. I expect wafer costs would be lower vs TSMC 7nm though so that would offset the lower density to some degree.

I'm extremely cynical about DLSS and think it's quite wrong to call it '4k DLSS' when the image is upscaled 1080p or something. A more correct label would be '1080p with DLSS' or something. Playing a DVD through my 'upscaling' Blu-Ray player doesn't make it a 1080p movie. I also enjoy pointing out that the Xbox 360 used a very impressive upscaler and had games running at 60fps back in 2005!

I'm also cynical about the way the 'free' performance is viewed. It is not free at all - a significant portion of the die is dedicated to those tensor cores, area which could theoretically have been used for higher shader throughput. The reason it appears free is because that performance is simply unavailable to normal rendering. And yes, I see DLSS as more of an enabler to get worthwhile RT performance than anything else.

And agreed on RT performance - I was very surprised and disappointed with Turing's RT performance given how well Nvidia usually market and sell things. I expected it to drastically improve with Ampere and was disappointed again - yes, RT performance has improved but only in line with overall performance from what I can see: the performance hit from enabling it has barely changed from Turing.

I could continue the rant about how RT is being compared i.e. a poor implementation of rasterisation vs RT, showing effects which are quite possible without RT. But my posts are already long enough!

[philehidiot]

This comment makes far too many assumptions.

Well, yes. Oddly enough I can't type a comment on a forum that covers every eventuality for every individual. I make the point that people should wait until releases are done, demand has settled and they have the power in the negotiation.

I've *just* upgraded from an A10-7870k to a 5600X - I'm currently borrowing a GTX 960 which, whilst a shed load better than my old APU, is still pretty naff at playing games even from 2015 (low settings for Witcher 3) and is quite the bottleneck for my shiny CPU.

So you're not in the market for a new, high end GPU, which is what is available now. What if last generation's GPU was as good as the latest mid-range one you're looking at and priced better? Would you buy one of those? I would. Last gen flagships can be better buys.

As for people wanting to upgrade... did you miss the RTX launch? They weren't widely considered worth it, so many people didn't upgrade which is one of the reasons for high demand this time around. As for those of us that hang on to hardware for many, many years, well yes, our GPUs simply don't allow for modern titles to be played nicely at all.

Yep, I saw the RTX launch. I skipped the generation as they didn't make a product worthy of my money. Nor did they price it at a price point I cared to consider. I, like you, don't throw my money at the shiny.

I hang on to my hardware for years, also. That's why I'm not in a rush. Those who hang on to hardware for years are best placed to sit and wait for prices to become more reasonable. I turn down the settings as an inevitable concequence of this decision. As a result, I can sit and wait until a card I want comes into my price range.

And whilst some of the disappointment with the delay may be down to "hype", we know for a fact that its going to be a better product that the 5700/5700 XT which, as you mention will come down in price, but in actuality, haven't yet. Why on earth would anyone spend X amount on what is effectively as last-gen GPU when there's a better performer (that we expected to be launched sooner) rolling out? Unless you've got so much money at your disposal that you simply don't care about depreciation on such a short timescale (something your comment suggests about yourself considering you seem somewhat oblivious to the masses of us with underpowered GPUs), it simply doesn't make sense to buy 5700/5700 XT now - and no one in their right mind would advise doing so until the prices drop significantly (which they haven't yet).

Pretty sure I mentioned about waiting for the prices to drop... pretty sure that was the central theme.

Why on Earth would I spend X on a last gen GPU? It depends what X is and what the relative performance Vs price is. It's all relative.

When I'm making a decision on a GPU, I do a lot of work. I look at this gen, last gen and consider the next gen. I aggregate benchmarks I care about and I do a price Vs performance. If last gen is giving performance within the window I am considering and the relative price Vs performance is better, I will buy last generation. I will also wait until the prices have settled and performance is known before making a decision. Because I'll be living with that decision for a long time.

Additionally, you're making some serious assumptions yourself (whilst you're turning them into borderline insults). I bought my GPU with the plan that I'd be dropping the resolution towards the end of its life. I set that plan out, just as I do with my CPUs. When my old CPU started struggling seriously, I applied the end of life overclock I'd planned from the beginning and started looking. It took me months to find a product at a price I was happy with. Why? Because Zen 2 had launched and I was having none of that nonsese. My GPU will have the resolution dropped to 1/4 of starting before I consider a change.

So yes, we're rightfully disappointed that the expected wait time for the 6700/6700 XT launch has increased - I don't understand why anyone would have difficulty in understanding why people would be "upset" (far too strong a word) by that.

Because it's evolving. Let it evolve. I'm not "oblivious". What I'm pointing out is how these "masses" should be looking at the market to get the most bang for buck. And if that means hanging on for a few months, lumping it. It's what I have to do.