AMD Ryzen 7 8700G and Ryzen 5 8600G Review: Zen 4 APUs with RDNA3 Graphics
by Gavin Bonshor on January 29, 2024 9:00 AM EST- Posted in
- CPUs
- AMD
- APUs
- Phoenix
- 4nm
- Zen 4
- RDNA3
- AM5
- Ryzen 8000G
- Ryzen 7 8700G
- Ryzen 5 8600G
Core-to-Core Latency
As the core count of modern CPUs is growing, we are reaching a time when the time to access each core from a different core is no longer a constant. Even before the advent of heterogeneous SoC designs, processors built on large rings or meshes can have different latencies to access the nearest core compared to the furthest core. This rings true especially in multi-socket server environments.
But modern CPUs, even desktop and consumer CPUs, can have variable access latency to get to another core. For example, in the first generation Threadripper CPUs, we had four chips on the package, each with 8 threads, and each with a different core-to-core latency depending on if it was on-die or off-die. This gets more complex with products like Lakefield, which has two different communication buses depending on which core is talking to which.
If you are a regular reader of AnandTech’s CPU reviews, you will recognize our Core-to-Core latency test. It’s a great way to show exactly how groups of cores are laid out on the silicon. This is a custom in-house test, and we know there are competing tests out there, but we feel ours is the most accurate to how quick an access between two cores can happen.
Looking at core-to-core latencies of the AMD Ryzen 7 8700G, as this is a monolithic Phoenix die, we can see good inter-core latencies between each of the eight individual Zen 4 cores. Going within the core, we can see solid latencies of 7ns, while things inter-core range between 17 and 21ns, showing that the Ryzen 7 8700G uses a single core cluster of eight cores.
Similar to what we've seen on previous iterations of Zen 4 and Zen 3, albeit on processors with multiple core complex (CCXs) such as the Ryzen 9 7950 and Ryzen 9 5950X, inter-core latencies are strong and low. In contrast, the Ryzen 7 8700G and other Ryzen 8000G monolithic chips on a single die remove the complications and penalties of connecting through AMD's Infinity Fabric interconnect. The Ryzen 7 8700G uses TSMC's refined 4nm manufacturing process, exactly the same as the Ryzen 7040 mobile, which is coincidentally the exact same design as the 8700G, given that AMD has repurposed Phoenix for use on AMD's AM5 desktop platform.
The core-to-core latency performance is inherently strong on the Ryzen 7 8700G, with low inter-core latencies. As expected, latency degrades a little going across the entire complex, but certainly not within the range where we would expect these penalties to cause latency issues when cores have to communicate with each other.
111 Comments
View All Comments
ToTTenTranz - Monday, January 29, 2024 - link
It might be worth mentioning the 8700G's larger iGPU gains almost 25% performance when paired with DDR5-6000 instead of the slower DDR5-5200:https://www.techspot.com/review/2796-amd-ryzen-870...
I don't know what advantages there are for DIY clients to hold on to JEDEC-approved speeds instead of just going with XMP and EXPO "pre-validated 1-click overclocks". Especially considering how it's only called "overclock" because JEDEC seems to take their sweet time to validate new clock/timing setups.
Slash3 - Monday, January 29, 2024 - link
The gap is likely even larger than that, as I believe HWU's 5200MT entry is still using faster XMP/Expo timings, rather than CL44 base spec. Steve simply scaled the speed for the other entries (the exception being 7200MT, which is also at Gear 2).meacupla - Monday, January 29, 2024 - link
The DDR5 RAM has to run at 1.1V to get JEDEC validation. It's not that JEDEC takes their sweet time, it is about DRAM makers not being able to produce fast DRAM.This is especially true when you want low latency.
AFAIK, the fastest JEDEC compliant DDR5 available is 6400, but with awful latency.
More so, I am surprised techspot got their chip to work at DDR5-7200. I didn't think it was possible to hit more than 6000 with any stability on the AM5 platform.
Slash3 - Monday, January 29, 2024 - link
Recent AGESA updates have made Gear 2 speeds in the 7200-8000MT range quite possible, but improvement over Gear 1 6000MT operation is fairly small due to the I/O die fabric speed being a limiting factor.These APUs are monolithic, though, so higher FCLK and MCLK/UCLK speeds are likely possible. Hopefully some reviewers and users will have time to dig into it now that the chips are officially out. I suspect we'll see some people running 6600MT in Gear 1 or 8000MT+ in Gear 2.
nandnandnand - Tuesday, January 30, 2024 - link
These APUs are bandwidth starved when it comes to iGPU performance. This review is already an outlier for using DDR5-5200, although pairing more expensive memory with 8700G/8600G doesn't make it a better choice than a CPU+discrete combo for budget users.FWhitTrampoline - Tuesday, January 30, 2024 - link
The main reason that the FCLK can get to higher rates/better FCLK:MCLK/UCLK ratios and with the proper stability on the Ryzen 8000G series APUs is because the Memory Controllers are on the same monolithic slab of silicon as the CPU cores. So there's no Cross-Module(Die to Die)/cross-clock-domain SerDes hops required there on Monolothic APUs to slow things down there. So the path from the Infinity fabric to the Memory Controllers on any Monolithic APU die based design is simpler there for APUs.thestryker - Monday, January 29, 2024 - link
JEDEC did something odd with DDR5 in that there are multiple sets of compliant latency and every maker has gone with the middle set. The lowest set is identical to DDR3/4 JEDEC compliant memory with the middle being about 14% higher. Still worse than XMP/EXPO kits as those just throw voltage through to lower latency.I'm hoping that with the 8500 DRAM Micron mentioned in a press release means maybe there will be some lower latency parts I can get for my server box.
is4u2p - Wednesday, January 31, 2024 - link
The legion Go runs 7500 RAM with the Z1 Extreme.is4u2p - Wednesday, January 31, 2024 - link
Yep, they peak around DDR5-6400 speeds, which is why you're seeing all these handhelds with 6500 or higher.barich - Wednesday, January 31, 2024 - link
DIY clients, sure. But these are going to end up in a lot of Dells and HPs, probably more by far than people who build their own. It's good to have representative benchmarks of that sort of configuration, and they'll definitely only ship JEDEC-specced RAM.Personally, I actually used Crucial's JEDEC DDR5-5600 in my 13th gen Intel build because stability is of paramount importance to me and anything higher is technically overclocking. Intel doesn't guarantee that any given CPU's memory controller will function properly faster than that. They pretty much all do, to a greater or lesser degree, but it's still a guess as to how far you can go.