• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Ryzen 8000G APU Memory Sweet Spot is DDR5-6000

Joined
Apr 28, 2023
Messages
67 (0.11/day)
I hope AMD address this serious limitation. Zen5 needs to support being run at 8000 and beyond.
No it doesn't.

Zen5 and 4 CPU cores aren't anywhere near bandwidth limited. If anything, due to the high amount of cache, they're relatively bandwidth insensitive. They're actually more sensitive to latency which is why the focus on overclocking the RAM for Zen4 has mostly focused on getting the timings (tertiary and secondary in particular) down with EXPO once you get to 6000.

The APU's on the other could indeed absolutely use DDR5 8000....for the iGPU! NOT THE CPU!!

This seems oddly low.
While it'd be nifty if faster RAM made a big difference it doesn't so its worth bothering with on Zen4 or Zen5. Unless you're trying for some AIDA read/write world record of course.

Being able to be pretty fast with, relatively cheap, DDR5 6000 is a good thing. You can get 2x16GB DDR5 6000 kits for ~$90-ish. Mean while cheapest 2x16GB DDR5 8000 kits run around $200-ish on newegg. Twice the price for minor performance gains, even with the lowest possible graphical presets and low resolution, is not something to get worked up over.


Much of the time in game the difference was only 1-2fps vs DDR5 6400 vs DDR5 6000 for Zen4 guys. Sometimes it was even less. Less than 1 fps!! For actual production CPU benches the differences were usually just as minor.

This is not at all a bandwidth limited CPU and its bizarre to me that people think it is.
 

tanaka_007

New Member
Joined
Mar 29, 2022
Messages
12 (0.01/day)
iGPU is not affected by XGMI Link (FCLK), so higher MCLK are better.

[CPU 7000]
CPU die---(FCLK,XGMI link on CPU PCB)---I/O die(UCLK,iGPU)---Memory(MCLK*2=DDR)
MCLK3200:UCLK3200:FCLK2000 (DDR5-6400, Low Latency)

[APU 8000G]
CPU+I/O(FCLK,UCLK,iGPU)---Memory(MCLK*2=DDR)
MCLK4200:FCLK2400 (DDR5-8400, High Bandwith)
*Reccomended B650/X670(Support PBO/Curve Optimizer) 2Slot DIMM M/B + Hynix A-die or Next Generation hynix chip
*Hynix's next generation DDR5 features ClockDriver to achieve higher clocks. (More than A-die, DDR5-6400 1.1v Native)

[Tuesday, January 30th 2024] AMD Ryzen 7 8700G Loves Memory Overclocking, which Vastly Favors its iGPU Performance
https://www.techpowerup.com/318446/...king-which-vastly-favors-its-igpu-performance
 
Last edited:
Joined
Apr 14, 2018
Messages
775 (0.31/day)
No it doesn't.

Zen5 and 4 CPU cores aren't anywhere near bandwidth limited. If anything, due to the high amount of cache, they're relatively bandwidth insensitive. They're actually more sensitive to latency which is why the focus on overclocking the RAM for Zen4 has mostly focused on getting the timings (tertiary and secondary in particular) down with EXPO once you get to 6000.

The APU's on the other could indeed absolutely use DDR5 8000....for the iGPU! NOT THE CPU!!


While it'd be nifty if faster RAM made a big difference it doesn't so its worth bothering with on Zen4 or Zen5. Unless you're trying for some AIDA read/write world record of course.

Being able to be pretty fast with, relatively cheap, DDR5 6000 is a good thing. You can get 2x16GB DDR5 6000 kits for ~$90-ish. Mean while cheapest 2x16GB DDR5 8000 kits run around $200-ish on newegg. Twice the price for minor performance gains, even with the lowest possible graphical presets and low resolution, is not something to get worked up over.


Much of the time in game the difference was only 1-2fps vs DDR5 6400 vs DDR5 6000 for Zen4 guys. Sometimes it was even less. Less than 1 fps!! For actual production CPU benches the differences were usually just as minor.

This is not at all a bandwidth limited CPU and its bizarre to me that people think it is.

Why do you keep linking this review as evidence of something it doesn’t really prove?

Before I explain further, in this case reviewers DO NOT have the required time to make proper comparisons, and those results serve more a catch-all.

Looking into the test setup there are two flaws with making proper comparisons in this review you have linked: theres no documentation on secondary and tertiary timings, and for testing 6200/6400 the flck was left at 2000 (see photo).

Without knowing what second and tertiary timings are, these can unfortunately have massive impacts on both latency and bandwidth of the tested kits. And when you desync fclk and mclk like this, you’re incurring an additional latency penalty on the 6200-6400 kits; the proper fclk would be 2033 and 2133 respectively.

Again some of this is beyond the scope of a typical review, but you cannot draw a conclusion (such as yours) when multiple variables are changing between kits aside from solely changing the frequency (whether they’re intentional or not).

So to reiterate, application dependent, there are benefits to ram frequency and additional bandwidth with dual CCD Zen 4 parts.

IMG_5256.jpeg
 
Last edited:

tanaka_007

New Member
Joined
Mar 29, 2022
Messages
12 (0.01/day)
*For APU
When purchasing a motherboard from now on,
we recommend a motherboard with 2 DIMM slots if you will only be installing 2 pieces of memory.
DDR5-8000+ can often be run on a 6-layer board with 2 slots.
If you have a 6-layer board with 4 slots, you may not be able to run DDR5-8200+. (Same for Intel systems)
*This may not always be possible due to individual differences in CPU and M/B.
*Some 4Layer M/Bs are not suitable for high clocks.
B650M-H_M.2+.png


-- manual overclocking for DDR5-5600 1.1v Hynix A-die 16GBx2 (Green PCB).
If you want to OC up to DDR5-8000+, please install a heat sink.
If DDR5-7800 or lower, it can also be executed with Hynix M-die.
A-die.jpg


AM5 tier list
 
Joined
Jul 30, 2019
Messages
3,381 (1.69/day)
System Name Still not a thread ripper but pretty good.
Processor Ryzen 9 7950x, Thermal Grizzly AM5 Offset Mounting Kit, Thermal Grizzly Extreme Paste
Motherboard ASRock B650 LiveMixer (BIOS/UEFI version P3.08, AGESA 1.2.0.2)
Cooling EK-Quantum Velocity, EK-Quantum Reflection PC-O11, D5 PWM, EK-CoolStream PE 360, XSPC TX360
Memory Micron DDR5-5600 ECC Unbuffered Memory (2 sticks, 64GB, MTC20C2085S1EC56BD1) + JONSBO NF-1
Video Card(s) XFX Radeon RX 5700 & EK-Quantum Vector Radeon RX 5700 +XT & Backplate
Storage Samsung 4TB 980 PRO, 2 x Optane 905p 1.5TB (striped), AMD Radeon RAMDisk
Display(s) 2 x 4K LG 27UL600-W (and HUANUO Dual Monitor Mount)
Case Lian Li PC-O11 Dynamic Black (original model)
Audio Device(s) Corsair Commander Pro for Fans, RGB, & Temp Sensors (x4)
Power Supply Corsair RM750x
Mouse Logitech M575
Keyboard Corsair Strafe RGB MK.2
Software Windows 10 Professional (64bit)
Benchmark Scores RIP Ryzen 9 5950x, ASRock X570 Taichi (v1.06), 128GB Micron DDR4-3200 ECC UDIMM (18ASF4G72AZ-3G2F1)

tanaka_007

New Member
Joined
Mar 29, 2022
Messages
12 (0.01/day)
FYI the FAQ says it's not a tier list.
Yes, this is just a list of hardware specifications.
For example, with a 6-phase VRM switching frequency of 800KHz and an 8-phase VRM switching frequency of 400KHz,
the latter has inferior transient response, so the ranking is meaningless.
VRM switching frequency setting in BIOS affects the stability of CPU/iGPU CurveOptimizer.
M/Bs whose settings cannot be changed are inferior even with the same VRM configuration.
*When stability is more important than conversion efficiency
 
Top