Monday, May 23rd 2022
AMD Unveils 5 nm Ryzen 7000 "Zen 4" Desktop Processors & AM5 DDR5 Platform
AMD today unveiled its next-generation Ryzen 7000 desktop processors, based on the Socket AM5 desktop platform. The new Ryzen 7000 series processors introduce the new "Zen 4" microarchitecture, with the company claiming a 15% single-threaded uplift over "Zen 3" (16-core/32-thread Zen 4 processor prototype compared to a Ryzen 9 5950X). Other key specs about the architecture put out by AMD include a doubling in per-core L2 cache to 1 MB, up from 512 KB on all older versions of "Zen." The Ryzen 7000 desktop CPUs will boost to frequencies above 5.5 GHz. Based on the way AMD has worded their claims, it seems that the "+15%" number includes IPC gains, plus gains from higher clocks, plus what the DDR4 to DDR5 transition achieves. With Zen 4, AMD is introducing a new instruction set for AI compute acceleration. The transition to the LGA1718 Socket AM5 allows AMD to use next-generation I/O, including DDR5 memory, and PCI-Express Gen 5, both for the graphics card, and the M.2 NVMe slot attached to the CPU socket.
Much like Ryzen 3000 "Matisse," and Ryzen 5000 "Vermeer," the Ryzen 7000 "Raphael" desktop processor is a multi-chip module with up to two "Zen 4" CCDs (CPU core dies), and one I/O controller die. The CCDs are built on the 5 nm silicon fabrication process, while the I/O die is built on the 6 nm process, a significant upgrade from previous-generation I/O dies that were built on 12 nm. The leap to 5 nm for the CCD enables AMD to cram up to 16 "Zen 4" cores per socket, all of which are "performance" cores. The "Zen 4" CPU core is larger, on account of more number-crunching machinery to achieve the IPC increase and new instruction-sets, as well as the larger per-core L2 cache. The cIOD packs a pleasant surprise—an iGPU based on the RDNA2 graphics architecture! Now most Ryzen 7000 processors will pack integrated graphics, just like Intel Core desktop processors.
The Socket AM5 platform is capable of up to 24 PCI-Express 5.0 lanes from the processor. 16 of these are meant for the PCI-Express graphics slots (PEG), while four of these go toward an M.2 NVMe slot attached to the CPU—if you recall, Intel "Alder Lake" processors have 16 Gen 5 lanes toward PEG, but the CPU-attached NVMe slot runs at Gen 4. The processor features dual-channel DDR5 (four sub-channel) memory, identical to "Alder Lake," but with no DDR4 memory support. Unlike Intel, the AM5 Socket retains cooler compatibility with AM4, so the cooler you have sitting on your Ryzen CPU right now, will work perfectly fine.
The platform also puts out up to 14 USB 20 Gbps ports, including type-C. With onboard graphics now making it to most processor models, motherboards will feature up to four DisplayPort 2 or HDMI 2.1 ports. The company will also standardize Wi-Fi 6E + Bluetooth WLAN solutions it co-developed with MediaTek, weaning motherboard designers away from Intel-made WLAN solutions.
At its launch, in Fall 2022, AMD's AM5 platform will come with three motherboard chipset options—the AMD X670 Extreme (X670E), the AMD X670, and the AMD B650. The X670 Extreme was probably made by re-purposing the new-generation 6 nm cIOD die to work as a motherboard chipset, which means its 24 PCIe Gen 5 lanes work toward building an "all Gen 5" motherboard platform. The X670 (non-extreme), is very likely a rebadged X570, which means you get up to 20 Gen 4 PCIe lanes from the chipset, while retaining PCIe Gen 5 PEG and CPU-attached NVMe connectivity. The B650 chipset is designed to offer Gen 4 PCIe PEG, Gen 5 CPU-attached NVMe, and likely Gen 3 connectivity from the chipset.
AMD is betting big on next-generation M.2 NVMe SSDs with PCI-Express Gen 5, and is gunning to be the first desktop platform with PCIe Gen 5-based M.2 slots. The company is said to be working with Phison to optimize the first round of Gen 5 SSDs for the platform.
All major motherboard vendors are ready with Socket AM5 motherboards. AMD showcased a handful, including the ASUS ROG Crosshair X670E Extreme, the ASRock X670E Taichi, MSI MEG X670E ACE, GIGABYTE X670E AORUS Xtreme, and the BIOSTAR X670E Valkyrie.
AMD is working to introduce several platform-level innovations like it did with Smart Access Memory with its Radeon RX 6000 series, which builds on top of the PCIe Resizable BAR technology by the PCI-SIG. The new AMD Smart Access Storage technology builds on Microsoft DirectStorage, by adding AMD platform-awareness, and optimization for AMD CPU and GPU architectures. DirectStorage enables direct transfers between a storage device and the GPU memory, without the data having to route through the CPU cores. In terms of power delivery Zen 4 uses the same SVI3 voltage control interface that we saw introduced on the Ryzen Mobile 6000 Series. For desktop this means the ability to address a higher number of VRM phases and to process voltage changes much faster than with SVI2 on AM4.
Taking a closer look at the AMD Footnotes, "RPL-001", we find out that the "15% IPC gain" figure is measured using Cinebench and compares a Ryzen 9 5950X processor (not 5800X3D), on a Socket AM4 platform with DDR4-3600 CL16 memory, to the new Zen 4 platform running at DDR5-6000 CL30 memory. If we go by the measurements from our Alder Lake DDR5 Performance Scaling article, then this memory difference alone will account for roughly 5% of the 15% gains.
The footnotes also reference a "RPL-003" claim that's not used anywhere in our pre-briefing slide deck, but shown in the video presentation. In the presentation we're seeing a live demo comparison between a "Ryzen 7000 Series" processor and Intel's Core i9-12900K "Alder Lake." It's worth mentioning here that AMD isn't disclosing the exact processor model, only that it's a 16-core part, if we follow the Zen 3 naming, that would probably be the Ryzen 9 7950X flagship. The comparison runs the Blender rendering software, which loads all CPU cores. Here we see the Ryzen 7000 chip finish the task in 204 seconds, compared to the i9-12900K and its 297 seconds time, which is a huge 31% difference—very impressive. It's worth mentioning that the memory configurations are slightly mismatched. Intel is running with DDR5-6000 CL30, whereas the Ryzen is tested with DDR5-6400 CL32—lower latency for Intel, higher MHz for Ryzen. While ideally we'd like to see identical memory used, the differences due to the memory configuration should be very small.
AMD is targeting a Fall 2022 launch for the Ryzen 7000 "Zen 4" desktop processor family, which would put this sometime between September thru October. The company is likely to detail the "Zen 4" microarchitecture and the Ryzen 7000 SKU list in the coming weeks.
Update 21:00 UTC: AMD has clarified that the 170 W PPT power numbers seen are the absolute max limits, not "typical" like the 105 W, on Zen 3, which were often exceeded during heavy usage.
Update May 26th: AMD further clarified that the 170 W number is "TDP", not "PPT", which means that when the usual x1.35 factor is applied, actual power usage can go up to 230 W.
You can watch the whole presentation again at YouTube:
Much like Ryzen 3000 "Matisse," and Ryzen 5000 "Vermeer," the Ryzen 7000 "Raphael" desktop processor is a multi-chip module with up to two "Zen 4" CCDs (CPU core dies), and one I/O controller die. The CCDs are built on the 5 nm silicon fabrication process, while the I/O die is built on the 6 nm process, a significant upgrade from previous-generation I/O dies that were built on 12 nm. The leap to 5 nm for the CCD enables AMD to cram up to 16 "Zen 4" cores per socket, all of which are "performance" cores. The "Zen 4" CPU core is larger, on account of more number-crunching machinery to achieve the IPC increase and new instruction-sets, as well as the larger per-core L2 cache. The cIOD packs a pleasant surprise—an iGPU based on the RDNA2 graphics architecture! Now most Ryzen 7000 processors will pack integrated graphics, just like Intel Core desktop processors.
At its launch, in Fall 2022, AMD's AM5 platform will come with three motherboard chipset options—the AMD X670 Extreme (X670E), the AMD X670, and the AMD B650. The X670 Extreme was probably made by re-purposing the new-generation 6 nm cIOD die to work as a motherboard chipset, which means its 24 PCIe Gen 5 lanes work toward building an "all Gen 5" motherboard platform. The X670 (non-extreme), is very likely a rebadged X570, which means you get up to 20 Gen 4 PCIe lanes from the chipset, while retaining PCIe Gen 5 PEG and CPU-attached NVMe connectivity. The B650 chipset is designed to offer Gen 4 PCIe PEG, Gen 5 CPU-attached NVMe, and likely Gen 3 connectivity from the chipset.
AMD is working to introduce several platform-level innovations like it did with Smart Access Memory with its Radeon RX 6000 series, which builds on top of the PCIe Resizable BAR technology by the PCI-SIG. The new AMD Smart Access Storage technology builds on Microsoft DirectStorage, by adding AMD platform-awareness, and optimization for AMD CPU and GPU architectures. DirectStorage enables direct transfers between a storage device and the GPU memory, without the data having to route through the CPU cores. In terms of power delivery Zen 4 uses the same SVI3 voltage control interface that we saw introduced on the Ryzen Mobile 6000 Series. For desktop this means the ability to address a higher number of VRM phases and to process voltage changes much faster than with SVI2 on AM4.
Update 21:00 UTC: AMD has clarified that the 170 W PPT power numbers seen are the absolute max limits, not "typical" like the 105 W, on Zen 3, which were often exceeded during heavy usage.
Update May 26th: AMD further clarified that the 170 W number is "TDP", not "PPT", which means that when the usual x1.35 factor is applied, actual power usage can go up to 230 W.
You can watch the whole presentation again at YouTube:
211 Comments on AMD Unveils 5 nm Ryzen 7000 "Zen 4" Desktop Processors & AM5 DDR5 Platform
In terms of benchmarking apps that only test a single core there can be a difference of 20% percent, but with the Intel processor using faster DDR5 memory!
Just not a huge fan with AMD being so much later and just at best matching Alderlake ST performance on what seems like a much more advanced node. I gave intel crap and even ditched their platform entirely for taking so long to release something worthwhile over 9th gen. MT performance is just as important but ignoring ST performance is also foolish. Again just mildly disappointed and expected more especially after Zen2 and Zen3. Looking forward to seeing actual reviews of these and hopefully AMD is being conservative. There are a lot of fanboys that just want to see AMD or Intel die depending on what team they imaginarily think they are a part of but we as consumers need both chip vendors to be competitive because that will mean much better products for us and not just quad cores for 7 generations....
Good times already an installation video out....
Again, using the same example as above
Comparing 1s and 200s jobs
Using YOUR equations
1s is 99.5% faster than 200s and 200s is 19900% slower than 1s
So your statement " Presenting those two as 90% and 95% faster is a far more accurate representation of their absolute time expenditure." is totally flawed because the 19900% is STILL THERE.
The only thing there is "faster" and "slower" doesn't apply to "Time" alone.
"Time" is just "No. of seconds" and there is no "Speed" there.
"Speed" only happens when "Something divided by Time"
You cannot describe something is "faster" or "slower" without a "Speed" element
Therefore in the equation you must calculate the "Speed" first
And that's the fundamental flaw in your statements / equations.
"No. of seconds" alone has no meaning
"Work done within No. of seconds" is what we needed.
1s and 200s has no meaning if the amount of work done is unknown.
The person could have spent 1s and get 1 work done and stop there while the other guy spent 200s and did 300 work done.
Therefore the "Amount of work done" must be put into the equation to calculate "Speed" before anyone could describe who is "Faster".
Comparing 204s vs 297s alone
You can say 204s is 31% "Smaller" or "Shorter" or "Less" or "reduction" than 297s, but you should never, never say it is "Faster" than 297s without a work done in the equation.
In this CPU case it is "1 test case done in 204 seconds" vs "1 test case done in 297 seconds"
So the "1" must be put into the equation
1/204 = 0.00490
1/297 = 0.00337
(490-337)/490 = 31% slower
(490-337)/337 = 45% Faster
Without the "1" , your equation does not represent any "Speed" element.
Please do realize we are comparing "How quickly the CPU works"
Your explanations only represents "Time reductions in 2 tests"
Replace the "Faster" word with "shorter" and your statements are totally fine.
But if you need to use the word "Faster" please include "Speed" into your equations.
Enough said
It's 7950X being 45% faster than 12900K or 12900K being 31% slower than 7950X (not 7950X being 31% faster than 12900K), it's so simple, basic logic stuff really, AMD's marketing guys probably are from financial sector measuring everything in margins (even when we are talking about mark-ups), it's better for profits for sure lol
Personally I am mostly invested in the FLCK. Hopefully it is 3000MHz (DDR5-6000) for starting. Just speculation of course before the NDAs steal my soul and I become aware of the truth. Beside the AM5 memory, I like the idea of more PCIe 5.0 Lanes. Dual chipset is a interesting approach as a solution. Will the consumer market need Gen5 NVMe drives? No way! but, it does make sense for content creators.
AMD iGPU??? I want to see a option without it. A waste of silicon in my opinion. Only time iGPU is useful is video encoding (Intel QUICKSYNC), laptops and low-end pre-builts. Give me a CPU at a lower price point and gut the iGPU for my desktop!
I would not worry at all at this stage. Now Intel may have a real surprise for RL and Zen4 may well find itself in trouble again at the end of the year but only time will tell. The one thing Intel has going for it is MT performance will be greatly enhanced due to doubling E-cores. AMD I doubt can get that crown back before Zen 5 at best.
Now seriously, the only thing that I would bet is that since 7nm CPU chiplet+14nm I/O die was better than Intel's 7nm 12th gen attempt in performance/Watt, I can't find a reason why a 5nm CPU chiplet+6nm I/O die wouldn't be better than Intel's 7nm 13th gen attempt, especially since on Intel's side we would also have double the efficiency cores and much higher cache and higher frequency also according to rumors.
As has been mentioned before in this thread, iGPU is very handy indeed in case of borked driver upgrades or early failing stages of dGPUs. I would expect the option of turning it off completely in bios as is the case already with Rocket lake K cpus.
If 12900k can render 1 image for 297s and 7950x can render the same image for 204s then Let the ryzen to render second image for 93s. That is the both CPUs will render for 297s, then the 12900k will have 1 full rendered image while 7950x for the same time will render 1.45 images or 45% faster. 7950x can finish 45% more work per unit time
Example for 45% faster rendering
Not saying this is right way to do it, I'm saying it gives some sort of information about the performance. I think you meant, real-world performance not IPC.
Using a single benchmark to indicate IPC is just as invalid as using a single benchmark to indicate the overall performance of a product. Or, arguably, even more invalid, as using the term IPC purports to speak to more fundamental architectural characteristics, which is then undermined all the more by using a single benchmark with its specific characteristics, quirks and specific performance requirements. IPC as a high-level description of real world performance per clock must be calculated across a wide range of tests in order to have any validity whatsoever.Again: these are saying the same thing. What does "the 19900% is still there" even mean? You understand that these numbers are relative representations of difference, right? That they don't exist in an absolute sense, but only exist as comparisons relative to a baseline? There is no contradiction here.....This is accounted for in my equations, as "297" or "204" isn't seconds, it's seconds to finish the work. The explicit context here is the question of "how much time does it take each of the CPUs to finish this task", not "which duration is longest". We wouldn't be talking about these numbers if they weren't the time to finish a workload, thus they can only be understood as seconds/workload, not seconds. If I was speaking of time alone, as you say I wouldn't be using terms like "faster" or "slower", I would be talking about "less" or "more" time. But I'm not. I'm talking about time to finish the work.But this is where you're reversing things. Again: you're calculating a rate: work-per-second, not seconds-per-workload. The numbers given are seconds-per-workload. You are transforming this into something that the data given is not - a rate of fractional units of work per second. The calculation for seconds per workloads - the speed in this context - is 204/1 and 297/1, which means the division by one is omitted as it is entirely redundant. You don't need to write out 204/1=204. Nobody before you here, and certainly not AMD's marketing team, has said anything about how many units of a given workload the CPU can complete per second. They compared the time spent to finish a specific workload. That's the opposite equation of what you're drawing up here.
Your base equation above is the following:
1 second/X seconds per workload = Y workload per second.
You're calculating percentages from this unit you're producing: workloads per second.
I don't have an equivalent equation, as my percentages are calculated from the unit given: secondsd per workload. You are explicitly transforming the data given into a different unit; I am not. This is where your confusion stems from. No such transformation is necessary in order to compare the speed of these processors, as we're not talking about their rate of work, but their relative speed in completing a given task.... but that's what we're comparing: the time difference between two CPUs finishing a single workload. We are not comparing "how quickly the CPU works". Not at all. If, for example, we were talking clock speeds (which are a rate), then you would be correct. But we are talking a comparison of the duration for a single workload.