Utterly baffling to me that they decided to go with this naming. Intel has been nothing but chided for it, and they decide to emulate it? When they finally had a naming scheme that was approaching something useful? I can't begin to fathom how this decision was made.
It's peak corporate nonsense you see they emulate Intel's naming scheme, but are 3xx instead of 2xx so they're better you see! Oh and let's add in AI cuz that's the new thing our shareholders want more of! It's so utterly ridiculous.
They have been doing this nonsense since Ryzen came out. Ryzen 3, Ryzen 5, Ryzen 7, etc, etc. And their chipsets also, X350 vs Z290, etc, etc. Just always just copy Intel's naming then use a bigger number.
Does it really matter? Not like these will be available in any numbers until end of next year. Where intel does paper launches, amd does these in such tiny numbers while continuing to sell the older ones well into next year.
I wouldn't call a naming scheme you need to decode "useful". Most people looking to buy a laptop aren't following the industry closely or anything like that, so the older naming scheme was very difficult to understand. You really do just want a "bigger number = better" kind of system.
According to the product pages, the GPUs of both SKUs run at 2900MHz.
Interestingly, from the die image in the slides, it looks like there are two L3 caches. This matches the 24MB figure on the product pages, and suggests that the Zen 5 cores have 16MB while the Zen 5c cores have their own 8MB.
"Interestingly, from the die image in the slides, it looks like there are two L3 caches. This matches the 24MB figure on the product pages, and suggests that the Zen 5 cores have 16MB while the Zen 5c cores have their own 8MB."
Indeed. AMD has not previously shown the ability to put regular and compact cores in the same CCX. So we're expecting to find two CCXs on Strix. Which will make things interesting, as even on-die, there's some extra latency going between CCXes. (On the plus side, this means AMD has developed a 4 core CCX for TSMC 4nm)
It is worrying AMD didnt say a single word on battery life. Intel's Meteor Lake showed impressive battery life on light tasks like looping video playback. Which is also Qualcomm's forte.
By any stretch of the imagination, Zen 5 should be very efficient, when compared to Zen 4 at least. The clock speeds haven't budged except downwards, and the N4 node could have -22% power.
If they've done a good job with the 2-CCX approach, the Zen 5c cores should be handling the light tasks like web browsing and video playback, while the Zen 5 cores remain mostly off.
But it's bound to be extremely hard to come up with meaningful numbers, because of the broad range of computing, graphics and even inference power this chip will be able to support.
Your workload is bound to differ from many others and the only sure thing is that a power virus will suck them rather near their configurable TCP setting... which again will be very wide and hard to compare.
Their main design point seems to point towards portable consoles, which means while you may be able to see relatively long gaming times, the ability to just support three days of creative writing, with mostly typing and a bit of surfing, may be cut short vs. what a tablet, mobile phone or even a Snapdragon can do.
There is no equivalent of the ultra-low power Intel cores nor do we have something like the latest ARM 500 cores, which would be able to run a co-pilot-less Office for days on battery fumes.
I'm pretty glad that those three days to a week without a socket scenarios basically don't exist in my life and I can focus on having a bit more power under the hood, even on the run.
RAM limitations may be the bigger issue, I really find it hard to work with less than 32GB, I'd never complain about an extra €100 to go to 64GB, but for some odd reason the market seems to follow the fruity cult's bad example when it comes to taxing RAM.
Glad to see that the Graphics are getting a bump to 16 CUs on the top-end, hopefully it'll have enough memory bandwidth to keep them fed.
However naming both the 16 CU and 12 CU versions "Radeon 980M" was a absolute dick move by AMD. It's like they're competing for the worst naming scheme possible.
This mixing of different core types reminds me of Intel and there Core differential based on power. It would be nice to just keep it simple. Here it may be the different Core types allow for different PCIE and memory bus?
AMD is not using different core designs, not really. Same CPU core, cache is a little different and they are packed more dense so can't be clocked as high.
Kind of like what Qualcomm did with their Kyro-100 processor inside the QSD 820 chipset. It was a 2+2 design, but both were the same cores. Except one was built bigger, faster, more cache, and the other was done all for efficiency.
It worked. And it worked well. But Qualcomm still cancelled the project and fired their engineers and whole department shortly after. Which is super confusing. Because it worked really well.
And now, Qualcomm is attempting to do something similar with the Snapdragon EliteX chipsets. They're 6+6 design with one focused on performance, and the other for efficiency, but it doesn't look as drastic difference as it was with the old QSD 820 designs.
This sentence is difficult to parse, maybe the comparison is missing:
"Compared to previous generations, the 3rd Gen Ryzen AI, made by Xilinx and now AMD acquired, they are called AMD Xilinx, as the Ryzen AI NPU within the Ryzen 8040 mobile series topped out at just 16 TOPS; this fell short of the Microsoft Copilot requirements by over half."
To my understanding the "Block FP16" is how INT8 is implemented already in most models. It loses the big advantage of FP16 (i.e., mixing small and large numbers) with no improvement when compared to INT8.
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
27 Comments
Back to Article
Dolda2000 - Sunday, June 2, 2024 - link
Utterly baffling to me that they decided to go with this naming. Intel has been nothing but chided for it, and they decide to emulate it? When they finally had a naming scheme that was approaching something useful? I can't begin to fathom how this decision was made.thestryker - Monday, June 3, 2024 - link
It's peak corporate nonsense you see they emulate Intel's naming scheme, but are 3xx instead of 2xx so they're better you see! Oh and let's add in AI cuz that's the new thing our shareholders want more of! It's so utterly ridiculous.TheinsanegamerN - Monday, June 3, 2024 - link
Surprised their x3d chips didnt get renamed to the "L337 69-420 haxor weed bro dortios core ULTIMATE RGB edition" CPUs.Gasaraki88 - Monday, June 3, 2024 - link
They have been doing this nonsense since Ryzen came out. Ryzen 3, Ryzen 5, Ryzen 7, etc, etc. And their chipsets also, X350 vs Z290, etc, etc. Just always just copy Intel's naming then use a bigger number.sharath.naik - Thursday, June 6, 2024 - link
Does it really matter? Not like these will be available in any numbers until end of next year. Where intel does paper launches, amd does these in such tiny numbers while continuing to sell the older ones well into next year.Kushan - Monday, June 3, 2024 - link
I wouldn't call a naming scheme you need to decode "useful". Most people looking to buy a laptop aren't following the industry closely or anything like that, so the older naming scheme was very difficult to understand. You really do just want a "bigger number = better" kind of system.Dolda2000 - Monday, June 3, 2024 - link
I'm not saying it was great, but at the very least you could get useful info out of decoding it.Alistair - Tuesday, June 4, 2024 - link
First comment seriously? Who cares, Tempest in a Teapot. Can we talk about the tech instead of some faux controversy over naming.R0H1T - Sunday, June 2, 2024 - link
Gavin is that a typo with 22 threads for the HX 370 ?Ryan Smith - Sunday, June 2, 2024 - link
Yes, it is. Thnaks!heffeque - Monday, June 3, 2024 - link
Another typo: Ryzen AI 9 365 has a 880M, not 890M (the 890M is only for the 370).del42sa - Monday, June 3, 2024 - link
what about GPU clock ?ET - Monday, June 3, 2024 - link
According to the product pages, the GPUs of both SKUs run at 2900MHz.Interestingly, from the die image in the slides, it looks like there are two L3 caches. This matches the 24MB figure on the product pages, and suggests that the Zen 5 cores have 16MB while the Zen 5c cores have their own 8MB.
Ryan Smith - Monday, June 3, 2024 - link
"Interestingly, from the die image in the slides, it looks like there are two L3 caches. This matches the 24MB figure on the product pages, and suggests that the Zen 5 cores have 16MB while the Zen 5c cores have their own 8MB."Indeed. AMD has not previously shown the ability to put regular and compact cores in the same CCX. So we're expecting to find two CCXs on Strix. Which will make things interesting, as even on-die, there's some extra latency going between CCXes. (On the plus side, this means AMD has developed a 4 core CCX for TSMC 4nm)
Dante Verizon - Monday, June 3, 2024 - link
Now we know that the Qualcomm chip hasn't even been launched and is already dead.meacupla - Monday, June 3, 2024 - link
I'm waiting to see battery run times.Infy2 - Monday, June 3, 2024 - link
It is worrying AMD didnt say a single word on battery life. Intel's Meteor Lake showed impressive battery life on light tasks like looping video playback. Which is also Qualcomm's forte.nandnandnand - Monday, June 3, 2024 - link
By any stretch of the imagination, Zen 5 should be very efficient, when compared to Zen 4 at least. The clock speeds haven't budged except downwards, and the N4 node could have -22% power.If they've done a good job with the 2-CCX approach, the Zen 5c cores should be handling the light tasks like web browsing and video playback, while the Zen 5 cores remain mostly off.
abufrejoval - Monday, June 3, 2024 - link
Who wouldn't?But it's bound to be extremely hard to come up with meaningful numbers, because of the broad range of computing, graphics and even inference power this chip will be able to support.
Your workload is bound to differ from many others and the only sure thing is that a power virus will suck them rather near their configurable TCP setting... which again will be very wide and hard to compare.
Their main design point seems to point towards portable consoles, which means while you may be able to see relatively long gaming times, the ability to just support three days of creative writing, with mostly typing and a bit of surfing, may be cut short vs. what a tablet, mobile phone or even a Snapdragon can do.
There is no equivalent of the ultra-low power Intel cores nor do we have something like the latest ARM 500 cores, which would be able to run a co-pilot-less Office for days on battery fumes.
I'm pretty glad that those three days to a week without a socket scenarios basically don't exist in my life and I can focus on having a bit more power under the hood, even on the run.
RAM limitations may be the bigger issue, I really find it hard to work with less than 32GB, I'd never complain about an extra €100 to go to 64GB, but for some odd reason the market seems to follow the fruity cult's bad example when it comes to taxing RAM.
nfriedly - Monday, June 3, 2024 - link
Glad to see that the Graphics are getting a bump to 16 CUs on the top-end, hopefully it'll have enough memory bandwidth to keep them fed.However naming both the 16 CU and 12 CU versions "Radeon 980M" was a absolute dick move by AMD. It's like they're competing for the worst naming scheme possible.
nfriedly - Monday, June 3, 2024 - link
Oh, wait, I take that back. It's just a typo in your table. The slide has the 12 CU version as 880M.[email protected] - Monday, June 3, 2024 - link
This mixing of different core types reminds me of Intel and there Core differential based on power. It would be nice to just keep it simple. Here it may be the different Core types allow for different PCIE and memory bus?Alistair - Tuesday, June 4, 2024 - link
AMD is not using different core designs, not really. Same CPU core, cache is a little different and they are packed more dense so can't be clocked as high.They are both P cores with minor differences.
Kangal - Wednesday, June 5, 2024 - link
Kind of like what Qualcomm did with their Kyro-100 processor inside the QSD 820 chipset. It was a 2+2 design, but both were the same cores. Except one was built bigger, faster, more cache, and the other was done all for efficiency.It worked. And it worked well. But Qualcomm still cancelled the project and fired their engineers and whole department shortly after. Which is super confusing. Because it worked really well.
And now, Qualcomm is attempting to do something similar with the Snapdragon EliteX chipsets. They're 6+6 design with one focused on performance, and the other for efficiency, but it doesn't look as drastic difference as it was with the old QSD 820 designs.
Gc - Monday, June 3, 2024 - link
This sentence is difficult to parse, maybe the comparison is missing:"Compared to previous generations, the 3rd Gen Ryzen AI, made by Xilinx and now AMD acquired, they are called AMD Xilinx, as the Ryzen AI NPU within the Ryzen 8040 mobile series topped out at just 16 TOPS; this fell short of the Microsoft Copilot requirements by over half."
Nate_on_HW - Tuesday, June 4, 2024 - link
I found it interesting that they also talked about the INT8 OPS throughput of the GPU and CPUWould find it interesting to get those numbers on AMDs &Qualcomms chip and maybe plot each module of the SoCs as "TOPS/Watt" (for comparison)
I wonder if the new windows11 "on-device ML-models" would use the whole chip for computing or only the NPU.
Rudde - Wednesday, June 5, 2024 - link
To my understanding the "Block FP16" is how INT8 is implemented already in most models. It loses the big advantage of FP16 (i.e., mixing small and large numbers) with no improvement when compared to INT8.