Demonstrating once again that Intel's primary problem is its inability to launch simple chiplets, and not all that much anymore with its manufacturing (yes it is a bit behind, but it might not even make the top 10 problems that Intel is facing anymore).
??? What does that have anything to do with this article on a monolithic AMD processor? And Intel's primary problem is its process, and 90% of its issues stem from that.
Yeah, but not long before that, L2 cache was chip-based and plugged into sockets on the motherboard. So, moving it in-package was actually a step towards integration rather than disaggregation.
Multi-chip modules have also been done in mainframes, for a long time.
Intel's monolithic designs have been mostly on time recently despite being huge, and their chiplets have been late. AMD's monolithic designs have universally been late since Zen 1.
This article is the latest in the trend of both of their problems trying to do too much at once. Many Intel delay articles feature comments juxtaposing their problems with AMD, and I added my followup with Intel to this one.
The biggest delay AMD had was when they introduced Zen+, delaying Zen 2 by one year. Outside of that there CPU division has been pretty much on time AFAIK. Now their GPU division... that's another story beginning with Vega and buggy drivers, which continue through to today (Like their HBM clocking problem), even into their 3rd iteration of RDNA.
AMD's problem is inefficiency in deploying and bringing laptops to market, it's dismal slowness and low availability..
Although AMD has the best CPU and iGPU for notebooks, in this regard intel is still light years ahead, intel announces a new line of CPUs and in 30 days it already has several models of laptops available for purchase. AMD takes months... until today there are no 6800u-based laptops to buy here. it sucks
It's more about the OEMs banking on the more well known name of Intel than anything AMD can do apart from advertisement.
Many of those who buy the computers for sale in supermarkets and the like are still very reluctant to buy higher-end AMD laptops out of fear they wouldn't be able to sell them. It's also the primary reason why their laptops have in general the smallest SSD option, as more would make it more expensive and they fear they wouldn't sell anymore if they're a few bucks extra just for more storage.
The problem with laptops as I understand it is that it's a far more collaborative exercise than with socketed desktop or even server parts.
Evidently in that market AMD can't just create a reference design and have notebook OEMs use that, nor can you just give them chips and drivers and have OEMs create finished laptops within months.
Both sides need to have and commit significantly sized engineering teams for a longish period to co-create products and it's not made easier by Intel showing their displeasure to OEMs that would like alternatives: at least not, when they're smaller than Apple.
Dismal availability isn't all AMD's fault, But having proven better parts than Intel and readiness to execute in principle (some products *did* get out) for several generations now is the only way to turn the tide... albeit slowly.
IMHO to a very large degree it's the ultrabook craze which has hurt modular designs, which could really help making notebooks both much more economical and ecological. That's partly Apple, but Intel was sure happy to weaponize it against AMD, too.
It's crazy that pretty much everything generationally different in a laptop fits into something like an Intel modular compute element, but nobody wants to sell (or buy?) laptops/NUCs based on that.
But then my laptops typically look almost new after ten years of use, while my kids manage to destroy them within two to four years.
My biggest problem with AMD APUs has been their driver support: their CPU portion basically still enjoys eternal life (except for Windows' 11 kill switch), but the iGPUs were sometimes already out of support while the APU was still sold as new (Richmond first, but last Kaveri based designs fared little better). AMD seems to have a five year life cycle for GPU architecture support, started from when first generational dGPUs are sold, not when the last APUs including them become end-of-life...
And then I just hate that all AMD drivers fail to work on Windows server editions, because evidently that costs a couple of extra dollars: As a developer I can't have my OS choice dictated by the GPU I happen to use somewhere in the life cycle of my systems!
Laptop vendors not updating customized iGPU drivers doesn't help either. I basically accepted that my Ryzen 5800U based laptop wasn't ever going to be a 3D machine when I bought it for it's fast and efficient 8 cores. But it still irks that I couldn't get updated drivers in the more than two years that since have gone by, without losing variable screen refresh and potentially the battery savings that go with that: evidently that requires custom integration support and OEMs don't budget driver updates, while lack of modularity/abstractions makes it necessary.
I am also wondering if AMD not having the drivers ready for prime time is the actual reason for the delay. Given TSMC's apparent availability of capacity of their advanced EUV nodes, that (drivers) seems to be more likely than not getting enough chips.
Basic 8GB RAM 256GB SSD is enough for most of the people. Considering the power/efficency/quality ratio of Apple MacBook Airs for asked price, it is not expensive at all. Add to that powerfull free apps like Garageband/iMovie/iWork etc. + cheap like MainStage and there is no competition in this area from Wintel side. In fact comparable Win laptops of similar quality are on the same price level. Of course, it all depends of user needs. For musician, photo/video editor there's only one reasonable option - Apple. There's no Wintel product that can compete with Apple in these areas in my opinion. More than tat - closed Apple ecosystem is an advantage in my eyes, because of security and efficiency reasons. Don't gest me wrong - Windows is not bad in these days (I'm using Wintel products at work), but I prefer MacOS and Apple hardware/software because of above reasons.
What do you mean Apple tax? Do you expect companies to develop hardware&software for free? Looking at the top hardware prices of other vendors, there is no, or small difference between Apple and their products. So, why don't you talk about Samsung, Huawei etc. taxes? If the price of any product borthers you, don't buy it. It's simple. I'm tired of people moaning all the time... You don't have to buy Bentley to reach the piont you want to drive.
Apple machines simply cost more for the same specs as most non-Apple. It's been ever thus. Why do you think they're so profitable? It has a lot to do with their enormous margins.
> Looking at the top hardware prices of other vendors, there is no, or small difference
I don't think you're truly comparing like-for-like. Or you're comparing MSRP, rather than actual street price. Even then, I'll grant that big PC OEMs have prestige products they try to price like Apple, but if you look at where their volume sales happen, it's much further down-market, where they do have a significant price advantage vs. Apple.
> If the price of any product borthers you, don't buy it.
That's basically what I said.
> I'm tired of people moaning all the time...
You want to talk up the advantages of Macs and yet somehow it's unfair for someone to point out the downsides? If you didn't want to end up here, then don't bring it up in a public forum where people are allowed to respond. I guess you're a prime candidate for the fascist world of Apple.
> Apple machines simply cost more for the same specs as most non-Apple. It's been ever thus.
One more time: "Looking at the top hardware prices of other vendors, there is no, or small difference." What you don't understand in this sentence? Similar quality - similar price. It's simple.
> Even then, I'll grant that big PC OEMs have prestige products they try to price like Apple, but if you look at where their volume sales happen, it's much further down-market, where they do have a significant price advantage vs. Apple.
Yep, Hyundai have the same price advantages over Mercedes... What are you talking about? You expect Apple to sell their premium quality built laptops for the price of cheap low quality plastic PC products. Get real!
> I guess you're a prime candidate for the fascist world of Apple.
Very weak conclusion based on nohing, and you just proved that you're Apple hater. Nothing but irrelevant arguments, and unrealistic expectations (which for me is moaning). Just tahe a breath, and read one more time what I wrote.
I use both - Apple and non-apple hardware. There are some laptops from other vendors, but they're in Apple price range. High quality doesnt't come cheap. That's what I'm trying to say.
"Platform readiness" So, judging by the fact that it's only delayed by a month, what was the problem? Exploitable hardware bug? Ryzen USB bug fixes? GPU drivers?
Maybe it's driver issues: The AI HW engine is 100% new for an x86 Win platform, so 1st benchmarks will determine if people give it a thumbs up or a "this is useless".
Maybe it's quality RAM availability: There are barely any DDR5 SODIMM with speeds that can actually take advantage of the 7940HS's performance. In theory the 7040HS series maxes out to DDR5-5600 or LPDDR5x-7500... and high speed RAM is especially important for the 780M (can already be seen on 680M benchmarks when going from DDR4 to DDR5).
Maybe it's a combination of things, and they feel that they want the 1st benchmarks to shine a bit better. You know... 1st impressions are important!
Perhaps AMD has finally figured out that 1st impressions can almost make or break a product. They do a great job of fixing those launch bugs and even increasing performance over the lifetime of their products. They just sputter out of the gate more often than not. A lot of people read those initial reviews and just assume that their drivers suck and their hardware is buggy or half-baked.
There will be no useful AI benchmarks at launch, because no one will know how to use it.
The only thing that matters is app performance, and it will take some time to hack it into Stable Diffusion, the LLaMA repos and such. That even took time for Apple, and they literally paid engineers to develop their own Stable Diffusion fork.
> There will be no useful AI benchmarks at launch, because no one will know how to use it.
I'm sure they'll have a couple of things working, like background removal and noise reduction for video conferencing. These will be 1st party plugins, or maybe developed in close partnership with 2nd party software vendors.
> it will take some time to hack it into Stable Diffusion
I doubt XDA could handle something like that. It's more about small, realtime models.
Probably firmware. Doubt its a hardware bug since the RDNA3 and ZEN4 cores are all carryover, but perhaps an issue with the AI cores? Can't be serious if they are delaying only a few weeks.
Am I alone in wishing that companies would just say what the problem is? I'd have a lot more respect for that level of honesty. Because, we've all been there when things all go south and something misses a launch window. It happens. Owning up to it doesn't seem to 'happen' as much as it should.
That's cool, I'm sure the 35-45W 7040HS will be great when it arrives.
But I really want to see some lower power chips for use in Steam deck-like handheld devices: I expect we'll see a 15-28W 7040U sometime this year.
A <10W Y-series chip with RDNA3 would be amazing if they could make it work, but I'm not sure if we'll ever get that. If it happened, though, it could be the heart of a true GPD Win 2 successor.
The "cheap" SKUs also cut out most of the iGPU, and having the 8 core SKUs in a handheld is both costly and power inefficient (since it leeches power from the iGPU).
Van Gogh was an anomaly that OEMs rejected until Valve picked it up, hence I don't think AMD will try that again.
They would have rejected it for very good reasons. It's Zen 2-based and TSMC N7, but AMD launched Rembrandt with Zen 3+ and RDNA2 on N6, around then. At the bottom of the market, AMD's existing N7 Zen 2 APUs with a Vega iGPU were fine, so no demand there. In the mid-market and above, Rembrandt was all-around better. Van Gogh was a weird chip.
BTW, I thought Van Gogh was custom-made under contract, similar to what AMD built for the big consoles. Do you know otherwise? Or were you just assuming AMD had it all done and Valve just ordered it as-is?
Van Gogh was on a leaked AMD laptop roadmap as a 2021 9W laptop APU. It is in line with Cezzane, and a predecessor to Dragon Crest.
I dont think I can post the link, but search for "Van Gogh Videocardz"
The "OEMs rejecting it" is speculation on my part. But I have always been skeptical of the Deck APU being "custom made," as Valve does not have the volume to justify that.
Right now, the only way to run big GPT-like models locally is to run a 32GB+ Mac, which costs as much as a car. Even on desktops, splitting the models between 3090s or 4090s isn't quite working yet. But a 64GB Asus G14 or whatever could be (theoretically) affordable, especially if bits of the project can run on an RTX 4060 or whatever.
It's going to run extremely slow on this Ryzen NPU. The reason is because Apple Silicon use the GPU and its unified memory VRAM to run large models. The Ryzen NPU will still have access to very slow DRAM since Ryzen APUs do not support unified memory.
The elephant in the room is memory bandwidth, where Apple uses up to 512-bit (M1 Max) or 1024-bit (M1 Ultra) LPDDR5 to provide enough bandwidth for decent GPU performance.
@mode_13h The regular M1 and M2 still perform well, with the NPU being far more power efficient than the GPU. And anyway, speed isnt a factor as much as "does the model even fit into the accelerator's RAM."
Lemur may actually be right. We shall see how AMD has their Phoenix accelerator set up, but the Rembrandt IGP (for instance) is not precisely unified memory.
Temper that excitement. I predict you'll be disappointed.
AI performance is predicted by 2 things: die size and memory bandwidth. This little XDNA block has neither going in its favor. If it even *can* run large models, they won't be fast.
Too bad! I am in the market for a light 14" or 15" notebook for the last several months, and have held off buying one as a Phoenix-based system with 32 GB RAM would check off most of the boxes. I hope that the announced Phoenix notebooks will now really be available next month, my old light notebook is approaching EOL (keyboard, battery doesn't hold it's charge as well anymore...), and it's getting annoying to work with. I also hope that AMD and OEMs aren't holding off on moving Phoenix into the market because they're sitting on many 6000 APU-based notebooks they want to sell off first. I'm not interested in one of those, and have my reasons why not.
> I also hope that AMD and OEMs aren't holding off on moving Phoenix into the market > because they're sitting on many 6000 APU-based notebooks they want to sell off first.
Yeah, I also had that thought.
AMD: "Uh, OEMs... we're going to need to delay the launch by a couple weeks..."
OEMs: "Cool. How about delaying by a couple months, instead?"
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
57 Comments
Back to Article
lmcd - Friday, March 17, 2023 - link
Demonstrating once again that Intel's primary problem is its inability to launch simple chiplets, and not all that much anymore with its manufacturing (yes it is a bit behind, but it might not even make the top 10 problems that Intel is facing anymore).bubblyboo - Friday, March 17, 2023 - link
???What does that have anything to do with this article on a monolithic AMD processor? And Intel's primary problem is its process, and 90% of its issues stem from that.
Jorgp2 - Friday, March 17, 2023 - link
It makes even less sense when you consider Intel made chiplet CPUs before AMD.Samus - Sunday, March 19, 2023 - link
The Pentium Pro (from 1995) had an on-package cache chiplet - TWO DECADES before AMD.mode_13h - Sunday, March 19, 2023 - link
Yeah, but not long before that, L2 cache was chip-based and plugged into sockets on the motherboard. So, moving it in-package was actually a step towards integration rather than disaggregation.Multi-chip modules have also been done in mainframes, for a long time.
lmcd - Friday, March 17, 2023 - link
Intel's monolithic designs have been mostly on time recently despite being huge, and their chiplets have been late. AMD's monolithic designs have universally been late since Zen 1.This article is the latest in the trend of both of their problems trying to do too much at once. Many Intel delay articles feature comments juxtaposing their problems with AMD, and I added my followup with Intel to this one.
ballsystemlord - Friday, March 17, 2023 - link
The biggest delay AMD had was when they introduced Zen+, delaying Zen 2 by one year. Outside of that there CPU division has been pretty much on time AFAIK.Now their GPU division... that's another story beginning with Vega and buggy drivers, which continue through to today (Like their HBM clocking problem), even into their 3rd iteration of RDNA.
lmcd - Thursday, March 23, 2023 - link
None of their desktop CPUs have been monolithic since Zen+mode_13h - Friday, March 24, 2023 - link
Huh? The 5700G would like a word with you.lmcd - Wednesday, March 29, 2023 - link
That thing is literally an overclocked mobile chip. And arrived late!Dante Verizon - Friday, March 17, 2023 - link
AMD's problem is inefficiency in deploying and bringing laptops to market, it's dismal slowness and low availability..Although AMD has the best CPU and iGPU for notebooks, in this regard intel is still light years ahead, intel announces a new line of CPUs and in 30 days it already has several models of laptops available for purchase. AMD takes months... until today there are no 6800u-based laptops to buy here. it sucks
Bofferbrauer - Saturday, March 18, 2023 - link
It's more about the OEMs banking on the more well known name of Intel than anything AMD can do apart from advertisement.Many of those who buy the computers for sale in supermarkets and the like are still very reluctant to buy higher-end AMD laptops out of fear they wouldn't be able to sell them. It's also the primary reason why their laptops have in general the smallest SSD option, as more would make it more expensive and they fear they wouldn't sell anymore if they're a few bucks extra just for more storage.
abufrejoval - Saturday, March 18, 2023 - link
The problem with laptops as I understand it is that it's a far more collaborative exercise than with socketed desktop or even server parts.Evidently in that market AMD can't just create a reference design and have notebook OEMs use that, nor can you just give them chips and drivers and have OEMs create finished laptops within months.
Both sides need to have and commit significantly sized engineering teams for a longish period to co-create products and it's not made easier by Intel showing their displeasure to OEMs that would like alternatives: at least not, when they're smaller than Apple.
Dismal availability isn't all AMD's fault, But having proven better parts than Intel and readiness to execute in principle (some products *did* get out) for several generations now is the only way to turn the tide... albeit slowly.
IMHO to a very large degree it's the ultrabook craze which has hurt modular designs, which could really help making notebooks both much more economical and ecological. That's partly Apple, but Intel was sure happy to weaponize it against AMD, too.
It's crazy that pretty much everything generationally different in a laptop fits into something like an Intel modular compute element, but nobody wants to sell (or buy?) laptops/NUCs based on that.
But then my laptops typically look almost new after ten years of use, while my kids manage to destroy them within two to four years.
My biggest problem with AMD APUs has been their driver support: their CPU portion basically still enjoys eternal life (except for Windows' 11 kill switch), but the iGPUs were sometimes already out of support while the APU was still sold as new (Richmond first, but last Kaveri based designs fared little better). AMD seems to have a five year life cycle for GPU architecture support, started from when first generational dGPUs are sold, not when the last APUs including them become end-of-life...
And then I just hate that all AMD drivers fail to work on Windows server editions, because evidently that costs a couple of extra dollars: As a developer I can't have my OS choice dictated by the GPU I happen to use somewhere in the life cycle of my systems!
Laptop vendors not updating customized iGPU drivers doesn't help either. I basically accepted that my Ryzen 5800U based laptop wasn't ever going to be a 3D machine when I bought it for it's fast and efficient 8 cores. But it still irks that I couldn't get updated drivers in the more than two years that since have gone by, without losing variable screen refresh and potentially the battery savings that go with that: evidently that requires custom integration support and OEMs don't budget driver updates, while lack of modularity/abstractions makes it necessary.
eastcoast_pete - Sunday, March 19, 2023 - link
I am also wondering if AMD not having the drivers ready for prime time is the actual reason for the delay. Given TSMC's apparent availability of capacity of their advanced EUV nodes, that (drivers) seems to be more likely than not getting enough chips.lemurbutton - Sunday, March 19, 2023 - link
>Although AMD has the best CPU and iGPU for notebooks,No they don't. Apple Silicon is years ahead.
eastcoast_pete - Sunday, March 19, 2023 - link
May well be, but they're walled off inside their own ecosystem, and MacBooks with at least 16 GB RAM and a 1 TB SSD aren't exactly cheap.max - Monday, March 20, 2023 - link
Basic 8GB RAM 256GB SSD is enough for most of the people. Considering the power/efficency/quality ratio of Apple MacBook Airs for asked price, it is not expensive at all. Add to that powerfull free apps like Garageband/iMovie/iWork etc. + cheap like MainStage and there is no competition in this area from Wintel side. In fact comparable Win laptops of similar quality are on the same price level. Of course, it all depends of user needs. For musician, photo/video editor there's only one reasonable option - Apple. There's no Wintel product that can compete with Apple in these areas in my opinion. More than tat - closed Apple ecosystem is an advantage in my eyes, because of security and efficiency reasons. Don't gest me wrong - Windows is not bad in these days (I'm using Wintel products at work), but I prefer MacOS and Apple hardware/software because of above reasons.mode_13h - Tuesday, March 21, 2023 - link
Hey, if you don't mind paying the Apple tax and don't mind vendor lock-in......then I bid you to enjoy your stay at the Hotel California.
max - Wednesday, March 29, 2023 - link
What do you mean Apple tax? Do you expect companies to develop hardware&software for free? Looking at the top hardware prices of other vendors, there is no, or small difference between Apple and their products. So, why don't you talk about Samsung, Huawei etc. taxes? If the price of any product borthers you, don't buy it. It's simple. I'm tired of people moaning all the time... You don't have to buy Bentley to reach the piont you want to drive.mode_13h - Thursday, March 30, 2023 - link
> What do you mean Apple tax?Apple machines simply cost more for the same specs as most non-Apple. It's been ever thus. Why do you think they're so profitable? It has a lot to do with their enormous margins.
> Looking at the top hardware prices of other vendors, there is no, or small difference
I don't think you're truly comparing like-for-like. Or you're comparing MSRP, rather than actual street price. Even then, I'll grant that big PC OEMs have prestige products they try to price like Apple, but if you look at where their volume sales happen, it's much further down-market, where they do have a significant price advantage vs. Apple.
> If the price of any product borthers you, don't buy it.
That's basically what I said.
> I'm tired of people moaning all the time...
You want to talk up the advantages of Macs and yet somehow it's unfair for someone to point out the downsides? If you didn't want to end up here, then don't bring it up in a public forum where people are allowed to respond. I guess you're a prime candidate for the fascist world of Apple.
max - Tuesday, April 18, 2023 - link
> Apple machines simply cost more for the same specs as most non-Apple. It's been ever thus.One more time: "Looking at the top hardware prices of other vendors, there is no, or small difference." What you don't understand in this sentence? Similar quality - similar price. It's simple.
> Even then, I'll grant that big PC OEMs have prestige products they try to price like Apple, but if you look at where their volume sales happen, it's much further down-market, where they do have a significant price advantage vs. Apple.
Yep, Hyundai have the same price advantages over Mercedes... What are you talking about? You expect Apple to sell their premium quality built laptops for the price of cheap low quality plastic PC products. Get real!
> I guess you're a prime candidate for the fascist world of Apple.
Very weak conclusion based on nohing, and you just proved that you're Apple hater. Nothing but irrelevant arguments, and unrealistic expectations (which for me is moaning). Just tahe a breath, and read one more time what I wrote.
I use both - Apple and non-apple hardware. There are some laptops from other vendors, but they're in Apple price range. High quality doesnt't come cheap. That's what I'm trying to say.
TheinsanegamerN - Monday, March 20, 2023 - link
Well unless I can buy a dell latitude, or thinkpad, running windows, with said M series in them, they are useless.max - Monday, March 20, 2023 - link
... useless for you. OK, nothing wrong with that. People do have different needs.meacupla - Friday, March 17, 2023 - link
"Platform readiness"So, judging by the fact that it's only delayed by a month, what was the problem?
Exploitable hardware bug? Ryzen USB bug fixes? GPU drivers?
Dante Verizon - Friday, March 17, 2023 - link
I suppose to align the efficiency curve and stability, since the rumors point to a slight clock reduction in the iGPU.meacupla - Friday, March 17, 2023 - link
Ah, so hot GPU.Okay, I guess these will be handicapped without good cooling then.
DaveLT - Saturday, March 18, 2023 - link
That's not it... Just the usual QA required with new platforms.heffeque - Saturday, March 18, 2023 - link
Maybe it's driver issues:The AI HW engine is 100% new for an x86 Win platform, so 1st benchmarks will determine if people give it a thumbs up or a "this is useless".
Maybe it's quality RAM availability:
There are barely any DDR5 SODIMM with speeds that can actually take advantage of the 7940HS's performance. In theory the 7040HS series maxes out to DDR5-5600 or LPDDR5x-7500... and high speed RAM is especially important for the 780M (can already be seen on 680M benchmarks when going from DDR4 to DDR5).
Maybe it's a combination of things, and they feel that they want the 1st benchmarks to shine a bit better. You know... 1st impressions are important!
Einy0 - Saturday, March 18, 2023 - link
Perhaps AMD has finally figured out that 1st impressions can almost make or break a product. They do a great job of fixing those launch bugs and even increasing performance over the lifetime of their products. They just sputter out of the gate more often than not. A lot of people read those initial reviews and just assume that their drivers suck and their hardware is buggy or half-baked.TheinsanegamerN - Monday, March 20, 2023 - link
it would be shocking if AMD launched a product in a ready state for once.brucethemoose - Saturday, March 18, 2023 - link
There will be no useful AI benchmarks at launch, because no one will know how to use it.The only thing that matters is app performance, and it will take some time to hack it into Stable Diffusion, the LLaMA repos and such. That even took time for Apple, and they literally paid engineers to develop their own Stable Diffusion fork.
mode_13h - Sunday, March 19, 2023 - link
> There will be no useful AI benchmarks at launch, because no one will know how to use it.I'm sure they'll have a couple of things working, like background removal and noise reduction for video conferencing. These will be 1st party plugins, or maybe developed in close partnership with 2nd party software vendors.
> it will take some time to hack it into Stable Diffusion
I doubt XDA could handle something like that. It's more about small, realtime models.
Samus - Sunday, March 19, 2023 - link
Probably firmware. Doubt its a hardware bug since the RDNA3 and ZEN4 cores are all carryover, but perhaps an issue with the AI cores? Can't be serious if they are delaying only a few weeks.dwillmore - Friday, March 17, 2023 - link
Am I alone in wishing that companies would just say what the problem is? I'd have a lot more respect for that level of honesty. Because, we've all been there when things all go south and something misses a launch window. It happens. Owning up to it doesn't seem to 'happen' as much as it should.ballsystemlord - Friday, March 17, 2023 - link
I second the motion.hallstein - Saturday, March 18, 2023 - link
It also stops people speculating wildly.mode_13h - Sunday, March 19, 2023 - link
If the problem involves some of their partner companies (either hardware or software), then it might not be AMD's decision to make.nfriedly - Friday, March 17, 2023 - link
That's cool, I'm sure the 35-45W 7040HS will be great when it arrives.But I really want to see some lower power chips for use in Steam deck-like handheld devices: I expect we'll see a 15-28W 7040U sometime this year.
A <10W Y-series chip with RDNA3 would be amazing if they could make it work, but I'm not sure if we'll ever get that. If it happened, though, it could be the heart of a true GPD Win 2 successor.
DaveLT - Saturday, March 18, 2023 - link
Yup there are 7040U samples floating around alreadybrucethemoose - Saturday, March 18, 2023 - link
The "cheap" SKUs also cut out most of the iGPU, and having the 8 core SKUs in a handheld is both costly and power inefficient (since it leeches power from the iGPU).Van Gogh was an anomaly that OEMs rejected until Valve picked it up, hence I don't think AMD will try that again.
mode_13h - Sunday, March 19, 2023 - link
> Van Gogh was an anomaly that OEMs rejectedThey would have rejected it for very good reasons. It's Zen 2-based and TSMC N7, but AMD launched Rembrandt with Zen 3+ and RDNA2 on N6, around then. At the bottom of the market, AMD's existing N7 Zen 2 APUs with a Vega iGPU were fine, so no demand there. In the mid-market and above, Rembrandt was all-around better. Van Gogh was a weird chip.
BTW, I thought Van Gogh was custom-made under contract, similar to what AMD built for the big consoles. Do you know otherwise? Or were you just assuming AMD had it all done and Valve just ordered it as-is?
brucethemoose - Tuesday, March 21, 2023 - link
Van Gogh was on a leaked AMD laptop roadmap as a 2021 9W laptop APU. It is in line with Cezzane, and a predecessor to Dragon Crest.I dont think I can post the link, but search for "Van Gogh Videocardz"
The "OEMs rejecting it" is speculation on my part. But I have always been skeptical of the Deck APU being "custom made," as Valve does not have the volume to justify that.
mode_13h - Thursday, March 23, 2023 - link
You can post links here. As long as they're in context, I think it's find.mode_13h - Thursday, March 23, 2023 - link
^fine.> I have always been skeptical of the Deck APU being "custom made,"
> as Valve does not have the volume to justify that.
What runs counter to this were the rumors about various specs changes, such as a 256-bit memory interface, at one point.
mode_13h - Sunday, March 19, 2023 - link
Plus, Van Gogh required (then) expensive LPDDR5, which is another thing that made it inappropriate for entry-level notebooks.brucethemoose - Saturday, March 18, 2023 - link
I am quite excited about the Ryzen AI thing.Right now, the only way to run big GPT-like models locally is to run a 32GB+ Mac, which costs as much as a car. Even on desktops, splitting the models between 3090s or 4090s isn't quite working yet. But a 64GB Asus G14 or whatever could be (theoretically) affordable, especially if bits of the project can run on an RTX 4060 or whatever.
lemurbutton - Sunday, March 19, 2023 - link
It's going to run extremely slow on this Ryzen NPU. The reason is because Apple Silicon use the GPU and its unified memory VRAM to run large models. The Ryzen NPU will still have access to very slow DRAM since Ryzen APUs do not support unified memory.mode_13h - Sunday, March 19, 2023 - link
> Ryzen APUs do not support unified memory.What do you even mean by that? I'm sure they do.
The elephant in the room is memory bandwidth, where Apple uses up to 512-bit (M1 Max) or 1024-bit (M1 Ultra) LPDDR5 to provide enough bandwidth for decent GPU performance.
brucethemoose - Tuesday, March 21, 2023 - link
@mode_13h The regular M1 and M2 still perform well, with the NPU being far more power efficient than the GPU. And anyway, speed isnt a factor as much as "does the model even fit into the accelerator's RAM."Lemur may actually be right. We shall see how AMD has their Phoenix accelerator set up, but the Rembrandt IGP (for instance) is not precisely unified memory.
mode_13h - Thursday, March 23, 2023 - link
Doesn't matter how good the NPU is, performance on larger models will be bottlenecked by memory bandwidth. That's a fact.To inference, you typically have to read in the *entire* model from memory per iteration. Do the math.
mode_13h - Sunday, March 19, 2023 - link
Temper that excitement. I predict you'll be disappointed.AI performance is predicted by 2 things: die size and memory bandwidth. This little XDNA block has neither going in its favor. If it even *can* run large models, they won't be fast.
brucethemoose - Tuesday, March 21, 2023 - link
LLaMA and Stable Diffusion inference (for instance) are getting really fast on "weak" hardware. The limiting factor is really the RAM pool.Finetuning is another story, and you just want a huge Nvidia GPU for that. Or maybe Intel/AMD some day.
eastcoast_pete - Sunday, March 19, 2023 - link
Too bad! I am in the market for a light 14" or 15" notebook for the last several months, and have held off buying one as a Phoenix-based system with 32 GB RAM would check off most of the boxes. I hope that the announced Phoenix notebooks will now really be available next month, my old light notebook is approaching EOL (keyboard, battery doesn't hold it's charge as well anymore...), and it's getting annoying to work with.I also hope that AMD and OEMs aren't holding off on moving Phoenix into the market because they're sitting on many 6000 APU-based notebooks they want to sell off first. I'm not interested in one of those, and have my reasons why not.
mode_13h - Sunday, March 19, 2023 - link
> I also hope that AMD and OEMs aren't holding off on moving Phoenix into the market> because they're sitting on many 6000 APU-based notebooks they want to sell off first.
Yeah, I also had that thought.
AMD: "Uh, OEMs... we're going to need to delay the launch by a couple weeks..."
OEMs: "Cool. How about delaying by a couple months, instead?"
max - Monday, March 20, 2023 - link
Are there any reasons why you don't buy M1/M2 MacBook Air?TheinsanegamerN - Monday, March 20, 2023 - link
Presumably because he runs Windows, and said software is not on Mac.Most people dont like switching entire ecosystems just for a CPU.
max - Monday, March 20, 2023 - link
Yeah. I can presume that, but I'm just curious. I haven't noticed where he said that soft is not on MacOS. Let him answer if he wants.