AMD Trims Q3 Forecast, $1 Billion Missing, Client Processor Revenue down 40%, Halved Quarter-over-Quarter

[AsRock]

To add, If you look into the details of the AM5 platform / chipset / motherboard features, really most of the major feature sets that sets AM5 apart from AM4 are only available to the very high end motherboards so far.

As example, the lower end X670 motherboards don't have a PCI 5 x16 slot, they're using PCIe 4 x16 apparently to save on cost. Most of them will simply have a PCIe 4 x16 and one PCIe 5 x4 to CPU.

I'm also seeing that none of the B650 boards will have PCIe 5 x16 based on AMDs high level diagrams of B650 connectivity. It appears this is at AMDs behest since there's no technical reason I can think of why it couldn't exist as that connectivity comes from the CPU, not the chipset.

To make matters a bit worse, the bandwidth to the chipset on all of the AMD Zen 4 chipsets is half of what it is to a Z690 or Z790. You could literally saturate that chipset<->CPU bus with a single PCIe 4 x4 m.2 SSD. So all of those connections its capable of driving are nice, as long as you only use a couple of those devices at once. Meanwhile most of the PCIe 5 lanes are not used.

Basically unless you get a $450+ motherboard, a bunch of what you pay for with Zen 4 \ AM5 is going to waste.

So with that backdrop, it makes total sense that the two best selling AM5 motherboards at Microcenter are a $699 X670E Crosshair Hero and the $999 Crosshair Extreme. I pulled this up for multiple locations across the country and these are always the two most popular :

View attachment 264641

Well they did skimp on the AM4 boards to as the second PCIe-slot is only x4 in a lot of cases. Seems like some thing being pushed and is not really needed.

Theoretically it's possible that AMD motherboard vendors are selling lots of AM4 motherboards, but you still need to drop a processor into them. What are these people doing? Dropping in previously owned CPUs?

AMD's Client business missed by $1B in revenue. That means they didn't sell a bunch of Zen 3 (Ryzen 5000 series) parts: both desktop and mobile.

You have tons of options going for AM4 and ram prices are really nice too, Hell a old? Zen2 chip will meet most peoples needs and they are fairly cheap too, well you can get the 3900X for $350 with cooler shipped and if your just a gamer your talking a lot less than that.

 

[cvaldes]

You have tons of options going for AM4 and ram prices are really nice too, Hell a old? Zen2 chip will meet most peoples needs and they are fairly cheap too, well you can get the 3900X for $350 with cooler shipped and if your just a gamer your talking a lot less than that.

LOL, I contributed to AMD's revenue quarters ago. They aren't getting anything from me in the immediate future.

Right now there are four desktop PC builds in the house, all Ryzen (three are Zen 3 builds, my daily driver desktop PC is Zen 2, 3700X). AMD has earned plenty from me as an individual consumer.

 

[ARF]

Maybe there will be major price cut of AM4 and AM5 in the coming months. That would be awesome

More awesome would be if we get the graphics cards back to normal. Because they are anything but normal now.

A 335-mm^2 chip today:

vs a 366-mm^2 chip back in 2015:

 

[Valantar]

Do they? Have you looked at AMDs official B650 block diagram? If they allow it and it's up to the manufacturer, why would that diagram show it as PCIe 4 x16?

B650 and X670 are PEG PCIe 4.0 minimum; B650E and X670E are PEG PCIe 5.0 minimum. All platforms can support m.2 PCIe 5.0 - though this depends on the board implementation, obviously. As the similarly named chipsets are otherwise identical in terms of features, B650E is B650 with PCIe 5.0.

So PCIe 5 isn't really useful is the argument, but we need to pay for it anyway right.

Uh ... did I claim that these two factors were somehow logically linked? They're both facts, whether you like them or not. PCIe 5.0 costs more to implement, so yes, one would need to pay (extra) for it for boards implementing it. It is also not really useful for consumers. Its lack of a real-world use doesn't have the power to magically remove the costs of implementing it.

In any case, I don't really agree with you. There's a super common daily use for high chipset bandwidth. It's called doing a backup.

... and you are doing a backup of a chipset-connected storage device to another chipset-connected storage device? That sounds like a rather unusual use case IMO. OS drives generally go in the primary, CPU-connected m.2 slot, and backup drives generally aren't m.2 anyway - I really can't imagine there are many people with a primary OS m.2 drive, a secondary m.2 drive for whatever, and a tertiary backup m.2 drive, and who also need daily backups of whatever is on that secondary m.2 drive. And, of course, the obvious solution even if this edge case is indeed what is happening: run your backups when you're not using the PC (unless they're running on the fly, in which case the amount of new data at any given time won't be sufficient to create any kind of lasting bottleneck).

More awesome would be if we get the graphics cards back to normal. Because they are anything but normal now.

A 335-mm^2 chip today:

View attachment 264654

vs a 366-mm^2 chip back in 2015:

View attachment 264655

This is hardly a surprise, nor just caused by greed. BOM costs for GPUs are massively higher, not only due to ~2-3x cost increases for many materials, but these GPUs have much more VRAM, much more expensive VRAM, beefier (and more expensive) VRMs, higher quality PCBs to accommodate higher speed VRAM and PCIe, and of course TSMC 7nm is still a rather expensive process node, while TSMC 28nm was cheap even when new, and especially four years into mass production. TSMC has also been raising prices due to running out of capacity, which pushes things even further.

Chip production isn't as simple as saying "the same die area should cost the same over time" - you need to take other factors into consideration as well, and you really can't ignore the BOM cost increases from more and more advanced componentry. That doesn't mean the last generation of GPUs isn't overpriced - it is, and it was clearly not designed for cost optimization. There is still quite a bit that can be done to bring down prices at the design and development stages (lowering PCB quality requirements through design tweaks; narrowing RAM buses, etc.), but some of that has also already been done.

 

[ARF]

This is hardly a surprise, nor just caused by greed. BOM costs for GPUs are massively higher, not only due to ~2-3x cost increases for many materials, but these GPUs have much more VRAM, much more expensive VRAM, beefier (and more expensive) VRMs, higher quality PCBs to accommodate higher speed VRAM and PCIe, and of course TSMC 7nm is still a rather expensive process node, while TSMC 28nm was cheap even when new, and especially four years into mass production. TSMC has also been raising prices due to running out of capacity, which pushes things even further.

Chip production isn't as simple as saying "the same die area should cost the same over time" - you need to take other factors into consideration as well, and you really can't ignore the BOM cost increases from more and more advanced componentry. That doesn't mean the last generation of GPUs isn't overpriced - it is, and it was clearly not designed for cost optimization. There is still quite a bit that can be done to bring down prices at the design and development stages (lowering PCB quality requirements through design tweaks; narrowing RAM buses, etc.), but some of that has also already been done.

What about scalping, like profit margin going from 30% in 2015 to 230% now in 2022? Let's be honest? ?

B650 and X670 are PEG PCIe 4.0 minimum; B650E and X670E are PEG PCIe 5.0 minimum. All platforms can support m.2 PCIe 5.0 - though this depends on the board implementation, obviously. As the similarly named chipsets are otherwise identical in terms of features, B650E is B650 with PCIe 5.0.

No one has ever asked them for the PCIe 5.0 marketing BS. So now, they have to take it back and remove it in order to cut the costs.

 

[the54thvoid]

you need to take other factors into consideration as well, and you really can't ignore the BOM cost increases from more and more advanced componentry.

@Valantar - there is a flaw in that logic which is the componentry which was cutting edge back then, also cost a lot. Each new advance in technology carries a cost - we saw it with HBM, but the cost wasn't astronomical. What we're seeing now is being unequally skewed by shareholder appeasement (following what was expected after mining income). If JSH or Lisa Su suddenly became charitable, they'd be kicked off the board.

 

[ARF]

... unless they justify it by saying that they wanna save the company from going under. We are in a deep recession, the manufacturing processes tick-tock is coming to an end in the next 5-10 years.
What are they thinking I don't know? Most Europeans have money now either for the heating bills or for the food... who is going to buy computer parts?

 

[TheoneandonlyMrK]

TBH, you can run Zen4 in eco mode (aka 65W power limit) and get only 3-10% performance hit in games. Zen4 is very efficient, it's the pricing of the whole AM5 platform that sucks big time.

Tbf there is definitely a early adopters tax on everything besides memory, that's not new.
Hopefully Raptor lake brings some price pressure to the market.
Full shelves will also help with the price reduction initiative.

 

[ARF]

Tbf there is definitely a early adopters tax on everything besides memory, that's not new.
Hopefully Raptor lake brings some price pressure to the market.
Full shelves will also help with the price reduction initiative.

Your PC bottleneck is not in the CPU, it's in the GPU. And definitely not in how wide the PCIe spec is.
In fact, we can't saturate PCIe 3.0 to the max today, most people still use old-fashioned HDDs and SATA SSDs.

PCIe 4.0 is still a dream. PCIe 5.0 is a ridiculous expensive nightmare and will be so for the next 5 years.
Unless MS finds a way to make PCIe 5.0 actually work.

 

[Valantar]

What about scalping, like profit margin going from 30% in 2015 to 230% now in 2022? Let's be honest?

Who has 230% profit margins? Most chipmakers report somewhere in the 40-60% range. And ... scalping? How does that relate to this discussion? MSRPs aren't affected by scalping. You were comparing MSRPs.

No one has ever asked them for the PCIe 5.0 marketing BS. So now, they have to take it back and remove it in order to cut the costs.

... uh, have you missed how literally the entire PC enthusiast scene consists of people shouting "MOAR PL0X" at the top of their lungs? Yes, there is a growing group taking a step back and saying "perhaps we have enough in some ways?", but the dominant approach is still asking for ever-increasing performance, even when it has no real benefits. Nothing sells new products like having a new, higher number to plaster onto the box. This is just as much of an enthusiast culture problem as it is a business culture problem.

@Valantar - there is a flaw in that logic which is the componentry which was cutting edge back then, also cost a lot. Each new advance in technology carries a cost - we saw it with HBM, but the cost wasn't astronomical. What we're seeing now is being unequally skewed by shareholder appeasement (following what was expected after mining income). If JSH or Lisa Su suddenly became charitable, they'd be kicked off the board.

Not a flaw in the logic, this is already taken into account. Yes, cutting-edge things are always expensive at their time, however, as bandwidths and power delivery needs have increased, the absolute cost of all of this has increased non-linearly. Why? Because as technology has developed, we have picked ever more of the low-hanging fruit, forcing ever more exotic solutions in order to maintain performance growth. It's (far) more obvious in things like external I/O, where USB 3.0 will never reach prices as low as USB 2.0, let alone anything faster than USB 3.0 ever getting there, but the same logic applies to BOM costs for internal componentry as well. High-end GPUs in 2015 had 4-6GB of GDDR5 (launched in 2007), while high-end GPUs today have 16-24GB of GDDR6/X, at 2-3x the bandwidth, memory that's produced on more advanced nodes that haven't seen downward per-bit cost scaling to match the capacity increase. Then you have ever-increasing VRM needs, both in quantity and quality, and the PCB design needs to make all this high speed memory, the high bandwidth display outputs, and the faster PCIe connections all work.

There is also an increasing culture for shareholder appeasement, yes, but that is stacked on top of steadily increasing BOM costs, which are in turn stacked on top of inflation. Corporations very obviously need to check their greed, but they also need to start designing more cost-optimized products and stop chasing performance in a way that throws all cost considerations to the wind, as we've been seeing for the past couple of generations now.

Unless MS finds a way to make PCIe 5.0 actually work.

How on earth would MS "find a way to make PCIe 5.0 work"? Literally none of the issues with PCIe 5.0 are related to OS support. It's unnecessarily fast, no consumer hardware can reasonably make use of it, and it's expensive due to physics. None of that is fixable by MS in any way.

 

[ARF]

... uh, have you missed how literally the entire PC enthusiast scene consists of people shouting "MOAR PL0X" at the top of their lungs? Yes, there is a growing group taking a step back and saying "perhaps we have enough in some ways?", but the dominant approach is still asking for ever-increasing performance, even when it has no real benefits. Nothing sells new products like having a new, higher number to plaster onto the box. This is just as much of an enthusiast culture problem as it is a business culture problem.

Maybe they don't listen to me shouting more cores please AMD is sitting on the old number 16.

How on earth would MS "find a way to make PCIe 5.0 work"? Literally none of the issues with PCIe 5.0 are related to OS support. It's unnecessarily fast, no consumer hardware can reasonably make use of it, and it's expensive due to physics. None of that is fixable by MS in any way.

I think it's something called DirectStorage but maybe I'm mistaken. Yes, definitely it is MS' obligations to make it work.

 

[TheoneandonlyMrK]

Your PC bottleneck is not in the CPU, it's in the GPU. And definitely not in how wide the PCIe spec is.
In fact, we can't saturate PCIe 3.0 to the max today, most people still use old-fashioned HDDs and SATA SSDs.

PCIe 4.0 is still a dream. PCIe 5.0 is a ridiculous expensive nightmare and will be so for the next 5 years.
Unless MS finds a way to make PCIe 5.0 actually work.

Indeed, that's why I snapped up a 5900X swap in for my 3800X a fair bit more CPU performance not such a high cost and I definitely do need to upgrade my GPU, the Vega plays anything but I am missing features, just waiting on rDNA 3 tbh and I'm set for a bit.

Except I have saturated pciex 3, 4x pciex 3 nvme in one x16 got there in reality though even I didn't. Keep it in use it Was pointless in any of my uses hopefully direct storage helps.

 

[R-T-B]

Oh did I mention OPEC

sir this is a tech site.

 

[Wirko]

Not a flaw in the logic, this is already taken into account. Yes, cutting-edge things are always expensive at their time, however, as bandwidths and power delivery needs have increased, the absolute cost of all of this has increased non-linearly. Why? Because as technology has developed, we have picked ever more of the low-hanging fruit, forcing ever more exotic solutions in order to maintain performance growth. It's (far) more obvious in things like external I/O, where USB 3.0 will never reach prices as low as USB 2.0, let alone anything faster than USB 3.0 ever getting there, but the same logic applies to BOM costs for internal componentry as well. High-end GPUs in 2015 had 4-6GB of GDDR5 (launched in 2007), while high-end GPUs today have 16-24GB of GDDR6/X, at 2-3x the bandwidth, memory that's produced on more advanced nodes that haven't seen downward per-bit cost scaling to match the capacity increase. Then you have ever-increasing VRM needs, both in quantity and quality, and the PCB design needs to make all this high speed memory, the high bandwidth display outputs, and the faster PCIe connections all work.

I checked if there's more than one Moore's law. Of course there is.

Moore's second law - Wikipedia

en.wikipedia.org

 

[AsRock]

LOL, I contributed to AMD's revenue quarters ago. They aren't getting anything from me in the immediate future.

Right now there are four desktop PC builds in the house, all Ryzen (three are Zen 3 builds, my daily driver desktop PC is Zen 2, 3700X). AMD has earned plenty from me as an individual consumer.

Yeah i think not to as seen as you have a 5900X lmao.

 

[Valantar]

Maybe they don't listen to me shouting more cores please AMD is sitting on the old number 16.

... And what on earth do you need more than 16 cores for that isn't a workstation task? Then again, thanks for proving my point I guess?

I think it's something called DirectStorage but maybe I'm mistaken. Yes, definitely it is MS' obligations to make it work.

This was a joke, right? DS has nothing to do with PCIe 5.0 - it will work on PCIe 3.0 even. And its main benefit is getting data more quickly to your GPU, bypassing the cpu as a decompression step, not making better use of PCIe bandwidth in and of itself. It'll lead to increases in SSD reads as the CPU's decompression won't bottleneck things any more, but that speedup will be major even on 3.0, and the difference between that and higher speed SSDs is likely to be small.

I checked if there's more than one Moore's law. Of course there is.

Moore's second law - Wikipedia

en.wikipedia.org

Makes sense. And I'd expect it's trajectory to be the opposite of the first Moore's Law, i.e. I'd expect fab cost increases to trend above 2x while transistor counts trend (well) below 2x.

 

[RandallFlagg]

Uh ... did I claim that these two factors were somehow logically linked? They're both facts, whether you like them or not. PCIe 5.0 costs more to implement, so yes, one would need to pay (extra) for it for boards implementing it. It is also not really useful for consumers. Its lack of a real-world use doesn't have the power to magically remove the costs of implementing it.

The board cost is obvious (though why it is so much more expensive on AM5 vs Z690 remains to be seen), and in any case is not what I meant.

What I meant is you are paying for it on the chip, but can't use it unless you *also* pay for it on the motherboard. It's great that Zen 4 has 24 PCIe 5 lanes, but when the motherboard only uses 4 of those in PCIe 5 mode it's a completely neutered point.

... and you are doing a backup of a chipset-connected storage device to another chipset-connected storage device? That sounds like a rather unusual use case IMO.

This is not unusual at all. I would bet the most typical configuration *right now* will be 2-3 m.2 SSDs and 1-2 SATA SSDs. SATA SSDs these days would be the most likely target - cheap with larger capacity. Putting them into a RAID 1 config wouldn't be too unusual either.

However, the trend is moving towards having all m.2, and it's moving that way fast. Almost everything now has 3 m.2 slots, and many midrange and up have 5. There's no reason to buy SATA anything now, as $140 gets you a cheap 2TB m.2.

Moreover, not only is all of this traffic going to go across you chipset, most if it will go across your chipset twice.

And this is most likely why Intel has quadrupled the IO bandwidth to the chipset since the Z390/490 days.

 

[mechtech]

More awesome would be if we get the graphics cards back to normal. Because they are anything but normal now.

A 335-mm^2 chip today:

View attachment 264654

vs a 366-mm^2 chip back in 2015:

View attachment 264655

They charge for transistors............not die area

 

[R0H1T]

sir this is a tech site.

Well it is related to tech in a way, directly even, so more $$$ spent on gas == less money for discretionary spends like for (PC) hardware.

 

[ARF]

... And what on earth do you need more than 16 cores for that isn't a workstation task?

What on Earth do you need PCIe 5.0 more than PCIe 3.0 or PCIe 4.0 which are not being saturated now?

 

[Valantar]

What on Earth do you need PCIe 5.0 more than PCIe 3.0 or PCIe 4.0 which are not being saturated now?

Exactly. It's all a waste. Completely pointless nonsense driving up prices for no reason.

 

[ARF]

Exactly. It's all a waste. Completely pointless nonsense driving up prices for no reason.

I really don't understand what they are thinking. The market is not ready for PCIe 5.0.
I will never put so large cooling on an SSD. It is stupid and underengineered:

ASRock Unveils Blazing M.2 PCIe Gen 5 SSD Cooler With Active Fan Heatsink, Compatible With Z790, X670E & B650 Motherboards (wccftech.com)

 

[pressing on]

I really don't understand what they are thinking. The market is not ready for PCIe 5.0.
I will never put so large cooling on an SSD. It is stupid and underengineered:

View attachment 264710
ASRock Unveils Blazing M.2 PCIe Gen 5 SSD Cooler With Active Fan Heatsink, Compatible With Z790, X670E & B650 Motherboards (wccftech.com)

The answer is future-proofing. People are willing to buy motherboards loaded with features they cannot currently use because they are planning to keep the boards for some time, maybe five years. For an AMD 600 series board that could well embrace Zen 4, Zen 5 and even Zen 6. BIOS updates might be necessary but not a new board. Who's to say where either PCI-E 5 graphics cards or M.2 NVME drives will be in say three years time. The alternative is to buy lesser spec motherboards more frequently but that's probably more typical of Intel buyers.

 

[ARF]

The answer is future-proofing. People are willing to buy motherboards loaded with features they cannot currently use because they are planning to keep the boards for some time, maybe five years. For an AMD 600 series board that could well embrace Zen 4, Zen 5 and even Zen 6. BIOS updates might be necessary but not a new board. Who's to say where either PCI-E 5 graphics cards or M.2 NVME drives will be in say three years time. The alternative is to buy lesser spec motherboards more frequently but that's probably more typical of Intel buyers.

Are you sure that anyone will care about PCIe 5.0 even after 5 years?
AMD's stupidity is infinite. Just release X670 with PCIe 4.0 and then in 5 years return to the PCIe 5.0. Maybe by then there will be more efficient manufacturing processes, so that everything will run cooler and quieter.

No, it's not the time now to oversupply with something that you don't even know if will work ever.

PCIe 5.0 might be DOA as is.

 

[Gaiacheck]

Are you sure that anyone will care about PCIe 5.0 even after 5 years?
AMD's stupidity is infinite. Just release X670 with PCIe 4.0 and then in 5 years return to the PCIe 5.0. Maybe by then there will be more efficient manufacturing processes, so that everything will run cooler and quieter.

No, it's not the time now to oversupply with something that you don't even know if will work ever.

PCIe 5.0 might be DOA as is.

People are overreacting, PCIe 3.0 lasted like 10 years and you will still have no problems running it with current gen hardware. I run 1080p144 with rx6600 on 3.0 mb, will I get better performance on 4.0? Probably. Will I notice it? Absolutely not.

I agree that 5.0 might just be pointless. Basically standard for hardware that doesn't exist yet. Future proofing for 10 years into future? bruh