Tuesday, July 16th 2024
AMD Granite Ridge and Strix Point Zen 5 Die-sizes and Transistor Counts Confirmed
AMD is about give the new "Zen 5" microarchitecture a near-simultaneous launch across both its client segments—desktop and mobile. The desktop front is held by the Ryzen 9000 "Granite Ridge" Socket AM5 processors; while Ryzen AI 300 "Strix Point" powers the company's crucial effort to capture Microsoft Copilot+ AI PC market share. We recently did a technical deep-dive on the two. HardwareLuxx.de scored two important bits of specs for both processors in its Q&A interaction with AMD—die sizes and transistor counts.
To begin with, "Strix Point" is a monolithic silicon, which is confirmed to be built on the TSMC N4P foundry node (4 nm). This is a slight upgrade over the N4 node that the company built its previous generation "Phoenix" and "Hawk Point" processors on. The "Strix Point" silicon measures 232.5 mm² in area, which is significantly larger than the 178 mm² of "Hawk Point" and "Phoenix." The added die area comes from there being 12 CPU cores instead of 8, and 16 iGPU compute units instead of 12; and a larger NPU. There are many other factors, such as the larger 24 MB CPU L3 cache; and the sizes of the "Zen 5" and "Zen 5c" cores themselves.
The "Granite Ridge" desktop processor is a chiplet-based processor, much like the Ryzen 7000 "Raphael." AMD is confirmed to be reusing the 6 nm client I/O die (cIOD) from "Raphael." This chip measures 122 mm², and packs 3.4 billion transistors. For reference, the cIOD of Ryzen 5000 "Vermeer" and Ryzen 3000 "Matisse" is built on the Global Foundries 12 nm node, measures a similar 125 mm², but with a much lower transistor count of 2.09 billion. The key contributor to the transistor count increase is the tiny iGPU that the Socket AM5 cIOD comes with. It may have just 1 workgroup processor (2 CU), but comes with the same display engines and media engines as the iGPU on APUs.
And now, onto the CPU complex dies (CCDs), the key area of silicon innovation for AMD desktop processors. The 8-core "Zen 5" CCD is codenamed "Eldora," and is built on the 4 nm foundry node. The HardwareLuxx.de report says that this the same N4P node as "Strix Point," but we've heard several other credible sources claiming that it is the more advanced N4X node, which favors high frequencies. The "Zen 5" CCD has a transistor count of 8.315 billion, which is a significant increase over the 6.5 billion of "Durango," the 8-core CCD based on "Zen 4," powering "Raphael."
What's most interesting is that this staggering 28% increase in transistor counts from the "Zen 4" Durango CCD to the "Zen 5" Eldora CCD comes at a die-area decrease of 0.5%. That's right, the "Zen 5" CCD measures 70.6 mm², whereas the "Zen 4" CCD is 71 mm². The "Zen 4" CCD is built on TSMC N5 (5 nm), which goes to show the leap in transistor densities achieved from the switch to N4P (or N4X).
A maxed out Ryzen 9 9950X processor hence has a total transistor count of 20.03 billion, while the single-CCD Ryzen 7 9700X has a transistor count of 11.715 billion.
AMD Ryzen 9000 series goes on sale from July 31, 2024.
Source:
HardwareLuxx.de
To begin with, "Strix Point" is a monolithic silicon, which is confirmed to be built on the TSMC N4P foundry node (4 nm). This is a slight upgrade over the N4 node that the company built its previous generation "Phoenix" and "Hawk Point" processors on. The "Strix Point" silicon measures 232.5 mm² in area, which is significantly larger than the 178 mm² of "Hawk Point" and "Phoenix." The added die area comes from there being 12 CPU cores instead of 8, and 16 iGPU compute units instead of 12; and a larger NPU. There are many other factors, such as the larger 24 MB CPU L3 cache; and the sizes of the "Zen 5" and "Zen 5c" cores themselves.
And now, onto the CPU complex dies (CCDs), the key area of silicon innovation for AMD desktop processors. The 8-core "Zen 5" CCD is codenamed "Eldora," and is built on the 4 nm foundry node. The HardwareLuxx.de report says that this the same N4P node as "Strix Point," but we've heard several other credible sources claiming that it is the more advanced N4X node, which favors high frequencies. The "Zen 5" CCD has a transistor count of 8.315 billion, which is a significant increase over the 6.5 billion of "Durango," the 8-core CCD based on "Zen 4," powering "Raphael."
What's most interesting is that this staggering 28% increase in transistor counts from the "Zen 4" Durango CCD to the "Zen 5" Eldora CCD comes at a die-area decrease of 0.5%. That's right, the "Zen 5" CCD measures 70.6 mm², whereas the "Zen 4" CCD is 71 mm². The "Zen 4" CCD is built on TSMC N5 (5 nm), which goes to show the leap in transistor densities achieved from the switch to N4P (or N4X).
A maxed out Ryzen 9 9950X processor hence has a total transistor count of 20.03 billion, while the single-CCD Ryzen 7 9700X has a transistor count of 11.715 billion.
AMD Ryzen 9000 series goes on sale from July 31, 2024.
33 Comments on AMD Granite Ridge and Strix Point Zen 5 Die-sizes and Transistor Counts Confirmed
Would be kinda funny of the i7 on tsmc overclocks better and ends up more desirable lol but we still probably have 3-4 months to go.
Some say only the high end will use 20A some say only the low end so it's really all over the place most leaks are from march or older.
Yeeeesh
If they have actually achieved both higher density and cooler temps, it's pretty impressive but we'll have to wait and see.
More transistors and similar power envelopes in a smaller area means denser.
Denser means harder to cool. Just physics. I don't need a source for that, it's observational analysis from this die size info.
Hence why 200 W from an Intel chip of any of the past five generations runs at a lower temperature than 200 W from an AMD Zen 1-4 chip, under the same cooler. Look at any of the TPU cooler reviews Intel vs AMD socket heat load capacity. Intel chip is less dense with a larger, monolithic die so heat transfer to the IHS/cold plate is improved from there being more surface area for the same energy amount to transfer through.
Also why things like offset mounts or delidding improves temperatures on AMD, to get those small CPU dies in an optimal position to pull heat away.I don't see how it's physically possible to run cooler with same power running through a smaller die. Unless they've done something similar to what Intel did in one of the Skylake refreshes where they thinned the layer on top of the cpu die to improve heat transfer.
BTW Strix Halo will use 3nm I/O die and part of why it's going to be a year late when it ships.Yet TSMC themselves show N4P is 22% more efficient and 11% more performant. The Zen 5 cpus will runs several degrees cooler. The fact clock speeds are the same as Zen 4 shows there has been no penalty for N4P. It's refined N5 and was always going to be better.
Also, the die is the same size, just denser. While it's generally true that denser would mean harder to cool, there are ways to mitigate it to an extent (eg. where the caches/dark silicon are in relation to the hotspots, since they are mostly idle transistors etc etc)
kind of strange that the article is about AMD but seems like everyone are more interested in whatever intel is releasing instead of this new AMD release.
Personally, I need to upgrade my 5600X but undecided if i should get one now or wait for the X3D variant.
The chiplets are exactly where they were - only that the CCDs are a bit smaller.If you game, wait for the X3D. If you don't, you're fine with whatever.
Will wait a bit anyway, in case there are any launch issues.
Edit: I mean, if we're talking about 30-40% between the 5800X3D and the 9800X3D, then probably, yes. Otherwise, I have my doubts.
The X3d part would be a nice extra, but not necessarily a hard requirement.
Although I expect their MT/application perfomance to be spot on but I guess we will see. Gaming is easier to fudge with dubious settings and technically be accurate but not representative of actual perfomance.
There is a lot of things they can do to improve the situation regardless of where the dies are and all that visibly, and still come out 15C better.
they specified 7C so wonder what they've done, better solder job, dies are sanded like intel does, combination? many things they can do.