News AMD Posts First Loss in Years as Consumer Chip Sales Plummet

[mitch074]

Longer socket cycles don't matter much when most people upgrade every 5+ years. And in case you missed the memo, AMD wouldn't commit to 4+ years long socket life with AM5. It likely doesn't want to lock itself into a potential repeat of having a hard time getting Zen 2 to work on AM4 and the backwards/forwards compatibility nightmare that came with Zen 3.

They hadn't committed to AM4 past 2020 originally either - and here we are, 2023, AM4 is still somewhat relevant. It is surprising how long they managed to support that socket (while there were some hurdles on the road we still got there)... Personally I upgraded my own main rig from a 2700X to a 5900X (X470 based, from Zen+ to Zen 3), I think I'm set until 2027. This should be more than enough.
It wasn't exactly nightmarish - there were 3 main problems in all and most got resolved.

1. PCI-E 4.0 support on X470/B450 chipsets : didn't go past beta. It could have worked had mobo makers actually not cheapened out on the hardware.
2. B550 not being backward-compatible with Zen+ and older : to be fair, I don't think that problem really lasted - people who wanted a test bench that could fully support everything from gen 1 to gen 5 could scrounge up the cash for a X570.
3. B350 and X370 not getting Zen3 support : AMD got some bad will from refusing to support this at first, but they still did it. Interestingly, no one really pissed at Intel for refusing support 9th gen support on B250.

It may be because I only worked with somewhat good boards, but apart from one case of an early B350 board not using the video on a 3200G, I've never had cross-gen compatibility problems. Historically, AMD did well in supporting their sockets - I can only remember one case of failed support (with AM3 to AM3+) and even then, they had actual engineering problems.

 

[SunMaster]

By going from 118W to 235-247W... 45% improvement in performance for around 100% more power, rough napkin math makes this less than 50% as efficient as the previous gen, but then if you don't care about power you can also go to intel and have it run unlimited.

Cute. I can make my 5950x operate on 270 watts with PBO, with close to 0 performance gain for the last 100 watts. I'm sure you can use that information to something that suits your agenda.

Perhaps you should lay off the napkin math for AMD and perhaps do Intel napkin math instead.

It is very hard to take anything you write seriously, because it's always slanted.

 

[ottonis]

I think there are three main problems with AMD's business:

1. For some reason, AMD never really got sustained traction in the notebook segment. For quite a time, even if a first tier laptop manufacturer provided maybe one (of many existing) line of notebooks with AMD chips, they usually sucked because of:
- non-optimal screen
- non-optimal graphics card
- non optimal I/O (e.g. no TB or equivalent)
- non optimal driver support (e.g. the infamous USB-driver bug which caused massive latencies/interrupt cycles on USB, thus rendering external audio interfaces useless for real time, low-latency audio applications.

any of those or a combination of them.

And this is such a pity because the latest AMD notebook APUs (the Ryzen 6x00 line) were really great in terms of power/performance ratio, efficiency, battery life etc.
Mobile computing is a huge chunk of market share, so AMD would be well adviced to put massively more efforts in collaborating with first tier laptop vendors and even provide them some extrremely low prices and great driver support.

2. The Graphics card section missed the low hanging fruits!
AMD is virtually non-existent in the entire video-editing / rendering / VFX /content creation community - which has substantially grown over the last half of a decade.
Why? because they didn't bother to outperform nvidia in terms of hardware video encoding and decoding engines. The performance of nVidias NVENC engines has only marginally improved over the past few years. It would have been an easy task to slap 3-4 more video encoders/decoders and further optimize them for H264, H265, AV1 etc. NVENC and even Intel's Quick-Sync Video have a huge problem: at similar quality, the output a 100-200% larger video file sizes compared to software encoders. AMD could have addressed this issue and optimized their hard- and software just to be better, smaller, faster - and they would be ruling the content creation market.

3. Pricing policy
When AMD came out with the Zen architecture, they beat Intel by price vs performance. Now that the playing field is fairly even and Intel just has the historical and current market dominance, AMD giving up on competitive pricing was very harmful to themselves: Who's gonna purchase Ryzen 7000er CPUs alongside expensive mainboards and RAM?
AMD should up their prices when they have reached more than 50% of the market share - not before that. Otherwise they will lose ground even further.

 

[TerryLaze]

This is perfectly demonstrated by the 7950X3D, which is only a few percent slower than the 7950X but for about half the power. Like so:

Yes, looking at a completely different CPU, shame they didn't start with that.

Cute. I can make my 5950x operate on 270 watts with PBO, with close to 0 performance gain for the last 100 watts. I'm sure you can use that information to something that suits your agenda.

Perhaps you should lay off the napkin math for AMD and perhaps do Intel napkin math instead.

It is very hard to take anything you write seriously, because it's always slanted.

How is it slanted when all sites show the same?!
They do use roughly twice the power that's not slanted that's how it is.

 

[ohio_buckeye]

I don’t know if 50% market share is the main number for upping prices, but I can say I’ve always built amd systems and the pricing on the am5 platform has me looking more at Intel. I’ve been using amd systems since 1997 when I was a teenager. So then upping pricing too much can definitely erode good well with consumers.

 

[InvalidError]

IIRC, the issue with Zen 2 was just the matter of enabling PCIe 4.0 on prior-generation boards. That's the only thing they backed away from, as I remember it.

Enabling Zen 3, on older boards, was an issue of EEPROM capacity limits on those early boards. Presumably, they've learned their lesson and given themselves ample breathing room.

There was a story about how Zen 2's substrate had to go from 10 to 14 layers in order to make the thing work on AM4 at all. You only need this much extra routing space when there is a major mismatch between socket pinout and where things actually are on the package, otherwise you only need to send signals and power almost straight up through the stack. On AM4, SOC voltage is under the left corner CCD while core voltage is clumped up under the right-side CCD with PCIe and USB IO peppered along the CCD/top edge. Fine when you have a single chiplet or monolithic die, not so great when the IOD is on the bottom edge, core power is needed across both CCDs and IO has to scoot across the whole thing in to reach the CCD edge.

Avoiding all of this hassle is why AMD went with new sockets for TR and EPYC. Were it not for it having committed to at least four years of socket support, AM4 would likely have bitten the dust back then too.

 

[PEnns]

Obscene pricing has consequences.

You'd think AMD has seen Nvidia's debacle by now.

 

[RedBear87]

Obscene pricing has consequences.

You'd think AMD has seen Nvidia's debacle by now.

What debacle? Nvidia's stocks are at one year record, while everyone and Elon Musk are buying their GPUs to make and deploy AI models... at this rate their data center's revenues will more than double their gaming GPU's soon enough.

 

[TerryLaze]

Obscene pricing has consequences.

You'd think AMD has seen Nvidia's debacle by now.

It's not like AMD is doing it out of greed like nvidia probably is.
AMD was making very little money from ZEN, and in general, until covid and they only had a few good-ish years. They were just about to break even now, they have to charge prices that will keep them alive.

AMD Net Income 2010-2023 | AMD

AMD annual/quarterly net income history and growth rate from 2010 to 2023. Net income can be defined as company's net profit or loss after all revenues, income items, and expenses have been accounted for. <ul style='margin-top:10px;'> <li>AMD net income for the quarter ending...

www.macrotrends.net

AMD Annual Net Income
(Millions of US $)
2022 $1,320
2021 $3,162
2020 $2,490
2019 $341
2018 $337
2017 $-33

AMD Annual Retained Earnings (Accumulated Deficit)
(Millions of US $)
2022 $-131
2021 $-1,451
2020 $-4,605
2019 $-7,095
2018 $-7,436
2017 $-7,775

 

[hannibal]

Maybe they had better get serious and bring prices down to where they should be

Noup... they better reduce gaming based products production, like Nvidia is doing.

 

[InvalidError]

Noup... they better reduce gaming based products production, like Nvidia is doing.

If you hollow out the entry-level such that nobody can be bothered to try PC gaming beyond what can run on IGPs, you are likely destroying future prospects at the mid-to-high end.

 

[bit_user]

By going from 118W to 235-247W... 45% improvement in performance for around 100% more power,

We've been through this many times. Zen 4 doesn't need so much power to deliver most of its performance. If you put it on the same power budget as Zen 3, it still significantly outperforms it.

The efficiency of Zen 4 will become more apparent, as reviews roll in for Zen 4 laptops. Here's some data from an early review. Check out the Cinebench R23 scores, across different power levels, for the R7 7840 HS:

https://wccftech.com/amd-phoenix-ry...ntel-same-power-gaming-efficiency-leadership/
​

 

[bit_user]

1. For some reason, AMD never really got sustained traction in the notebook segment. For quite a time, even if a first tier laptop manufacturer provided maybe one (of many existing) line of notebooks with AMD chips, they usually sucked because of:
- non-optimal screen
- non-optimal graphics card
- non optimal I/O (e.g. no TB or equivalent)
- non optimal driver support (e.g. the infamous USB-driver bug which caused massive latencies/interrupt cycles on USB, thus rendering external audio interfaces useless for real time, low-latency audio applications.

any of those or a combination of them.

And this is such a pity because the latest AMD notebook APUs (the Ryzen 6x00 line) were really great in terms of power/performance ratio, efficiency, battery life etc.
Mobile computing is a huge chunk of market share, so AMD would be well adviced to put massively more efforts in collaborating with first tier laptop vendors and even provide them some extrremely low prices and great driver support.

I remember reading somewhere that laptop ODMs require a lot more hand-holding and support than other types of platforms. I think AMD has been ramping up their internal resources needed to support this segment, but it takes a while to do.

2. The Graphics card section missed the low hanging fruits!
AMD is virtually non-existent in the entire video-editing / rendering / VFX /content creation community - which has substantially grown over the last half of a decade.
Why? because they didn't bother to outperform nvidia in terms of hardware video encoding and decoding engines. The performance of nVidias NVENC engines has only marginally improved over the past few years. It would have been an easy task to slap 3-4 more video encoders/decoders and further optimize them for H264, H265, AV1 etc. NVENC and even Intel's Quick-Sync Video have a huge problem: at similar quality, the output a 100-200% larger video file sizes compared to software encoders. AMD could have addressed this issue and optimized their hard- and software just to be better, smaller, faster - and they would be ruling the content creation market.

I think they have put a lot into video compression performance on newer codecs, as we can clearly see from @JarredWaltonGPU 's excellent comparison, earlier this year:

​

https://www.tomshardware.com/news/amd-intel-nvidia-video-encoding-performance-quality-tested​

​

If you look at H.265 and especially AV1, AMD has really closed the quality gap vs. Nvidia. Unfortunately, they don't seem to have revisited H.264 to improve quality of that codec. Performance-wise, they now appear to be in the lead.

​

​

​

The problem AMD might now face is convincing app developers to support their hardware? I suspect many apps are using Nvidia's native APIs, rather than going through DirectShow.

 

[bit_user]

How is it slanted when all sites show the same?!

Your analysis is too simplistic You treat things as linear that are not. That, and picking which products and metrics is where the slant comes in..

As I said, the Zen 4 laptop CPUs should do an excellent job of showcasing Zen 4's inherent efficiency.

 

[TerryLaze]

We've been through this many times. Zen 4 doesn't need so much power to deliver most of its performance. If you put it on the same power budget as Zen 3, it still significantly outperforms it.

We've been through this many times. It still does use that much though.
Just as nobody cares that you can limit power on intel, nobody cares that you can limit power on ryzen.

Your analysis is too simplistic You treat things as linear that are not. That, and picking which products and metrics is where the slant comes in..

As I said, the Zen 4 laptop CPUs should do an excellent job of showcasing Zen 4's inherent efficiency.

So this was about the normal desktop CPUs, I was talking about the normal desktop CPUs, you are talking about laptop CPUs, but I am the one picking products and metrics?!

 

[emitfudd]

I bought stock in AMD and 7 other companies earlier this year. AMD is the only one with a loss.

 

[bit_user]

We've been through this many times. It still does use that much though.
Just as nobody cares that you can limit power on intel, nobody cares that you can limit power on ryzen.

That's only when you're looking at X-series and K-series @ stock. The story changes, when you look at non-X and non-K, or if you start limiting either below the stock power limits.

Here's what happens, when you dial back the power limits on a Zen 4 desktop CPU. In this case, look at the 65W Ryzen 9 7900 (non-X) @ stock:

​

Source: https://www.techpowerup.com/review/amd-ryzen-9-7900/21.html​

​

 

[bit_user]

I bought stock in AMD and 7 other companies earlier this year. AMD is the only one with a loss.

If you're playing in the stock market, you probably know that some stocks are overpriced. Just because a stock goes down doesn't mean the company is doing poorly - it could just be that some investors overvalued it. Picking individual stocks is really about trying to find stocks that represent a good value, rather than jumping on the bandwagon and buying what's currently hot.

I expect AMD will continue to execute well, but I'm not buying their stock. I don't know if they can execute well enough to justify an even higher price. Maybe, if they can finally tap into this AI boom.... but they've been chasing that carrot for a long time, with little to show for it.

 

[TerryLaze]

That's only when you're looking at X-series and K-series @ stock. The story changes, when you look at non-X and non-K, or if you start limiting either below the stock power limits.

Here's what happens, when you dial back the power limits on a Zen 4 desktop CPU. In this case, look at the 65W Ryzen 9 7900 (non-X) @ stock:

​

Source: https://www.techpowerup.com/review/amd-ryzen-9-7900/21.html​

​

Which one of these do you think is the dialed back one?!
65W stock is 88W PPT, or real max power draw, stock is not dialed-back, stock is stock.
The pic shows stock vs PBO/OC with the CPU losing half the efficiency when you go higher than the already high 88W stock.
What's new about overclocking reducing efficiency?!
That's why people think that k CPUs have bad efficiency, because every review site only shows power unlimited extreme overclock numbers, it's not news.

 

[bit_user]

Which one of these do you think is the dialed back one?!
65W stock is 88W PPT, or real max power draw, stock is not dialed-back, stock is stock.

You sound confused. If we're talking about Zen 4 desktop CPUs, then my point is that the X-models boost far above their peak efficiency point. So, if we're talking efficiency, then we ought to look at the 65 W (88 W PPT) models @ stock.

The pic shows stock vs PBO/OC with the CPU losing half the efficiency when you go higher than the already high 88W stock.

Yes, that's exactly the point. You were treating it as linear, but it's not. If Zen 4's performance scaled linearly with power, then it wouldn't lose efficiency as power increased.

The moral of the story is that you cannot summarize an entire product line by looking only at its flagship model. Especially when it comes to efficiency.

 

[KyaraM]

What debacle? Nvidia's stocks are at one year record, while everyone and Elon Musk are buying their GPUs to make and deploy AI models... at this rate their data center's revenues will more than double their gaming GPU's soon enough.

Rule number 1:
Intel and Nvidia are always only one step away from disaster and everything they do is a debacle. This never applies to AMD. They are too good to fail. At least when listening to their fanboys...

Personally, I enjoy reading through the copium especially when thinking back to the same people proclaiming Intel essentially dead already just recently, after their report. Quite funny, to be honest.

 

[TerryLaze]

You sound confused. If we're talking about Zen 4 desktop CPUs, then my point is that the X-models boost far above their peak efficiency point. So, if we're talking efficiency, then we ought to look at the 65 W (88 W PPT) models @ stock.


Yes, that's exactly the point. You were treating it as linear, but it's not. If Zen 4's performance scaled linearly with power, then it wouldn't lose efficiency as power increased.

The moral of the story is that you cannot summarize an entire product line by looking only at its flagship model. Especially when it comes to efficiency.

It wasn't about efficiency and you thinking that it was just shows that you are just parroting pre-formulated propaganda without even thinking about what the other person said.
Miles said "Zen 4 is a relatively small gain over Zen 3 for a much higher cost" and while he probably meant cost of entry or platform, power is a cost as well,
on which you responded with a graph showing the 7950x being 45% faster, if only running server type workloads, without any regard to power draw so you didn't reply on the basis of efficiency, but you are now trying to turn it around to efficiency in general just so you can keep defending the undefendable.
Zen 4 still uses twice the amount of power for that 45% better performance that you showed.

Also you are the one that treated it as linear by trying to suggest that everything you run on zen 4 will be 45% faster.

 

[bit_user]

It wasn't about efficiency

I made a point about performance, which you then turned into an argument about efficiency, because you can't allow anyone to say anything good about AMD without letting it go unchallenged.

you thinking that it was just shows that you are just parroting pre-formulated propaganda

I'll not be lectured about propaganda by you, of all people.

Miles said "Zen 4 is a relatively small gain over Zen 3

Which is objectively and provably incorrect, as I showed.

while he probably meant cost of entry or platform, power is a cost as well,

When people mean power, they say power. Nothing in that post referenced heat or power - it's just your wishful thinking that you're hearing.

on which you responded with a graph showing the 7950x being 45% faster, if only running server type workloads,

It was Toms' Multithreaded composite score. The review included "desktop" and "workstation" benchmarks. No sever tests.

without any regard to power draw so you didn't reply on the basis of efficiency, but you are now trying to turn it around to efficiency in general just so you can keep defending the undefendable.

You are piling bad inferences atop bad conclusions and getting yourself twisted and crossed up, in the process.

You made this about efficiency, and I simply responded to your points. You tried to claim Zen 4 has poor efficiency, and I merely showed that to be an incorrect generalization.

Zen 4 still uses twice the amount of power for that 45% better performance that you showed.

There! You're doing it, again! That data is from the 7950X, not Zen 4 broadly. I just showed it doesn't apply to all Zen 4 desktop CPUs, so you can't even claim ignorance.

Also you are the one that treated it as linear by trying to suggest that everything you run on zen 4 will be 45% faster.

You don't seem to understand a basic tenet of logic. If someone makes a blanket statement, such as: "Zen 4 is a relatively small gain over Zen 3", all I have to do is show even one case where that's not true, in order to invalidate the assertion. That's exactly what I did. I showed the 7950X was 45.5% faster, in Tom's multithreaded composite score.

 

[TerryLaze]

There! You're doing it, again! That data is from the 7950X, not Zen 4 broadly. I just showed it doesn't apply to all Zen 4 desktop CPUs, so you can't even claim ignorance.

You showed the 7900 that has no 5900 counterpart so what exactly are you showing there other than overclocking makes efficiency bad?
Actually you are showing the 5900x which in your words has worse efficiency since it's an X part, having the exact same efficiency as the overclocked 7900 non-x part....
And 45% above the 5900x result would be 31000 points which needs the 7900 non-x to be overclocked to still not quite reach.
Hey, all I have to do is show even one case where that's not true, in order to invalidate the assertion.

"Zen 4 is a relatively small gain over Zen 3 for a much higher cost"
You keep cutting that last part out to move the goalpost.
You showed that it has better performance I showed that the cost for it is much higher.
Efficiency is only a by product of that and I didn't make it be about that, twice the power is a much higher cost and that is is objectively and provably CORRECT.

"Which is objectively and provably incorrect, as I showed."
If you move the goal post enough then everything is objectively and provably incorrect.

 

[Zerk2012]

I think this thread has ran it's course and should be closed.