AMD recommends pairing Ryzen 8000G APUs with dual-channel DDR5-6000 memory

Shawn Knight

Posts: 15,296   +192
Staff member
In brief: AMD at CES this week introduced its Ryzen 8000 G-series APUs for its AM5 platform, and PCWorld recently spent some time with AMD technical marketing manager Donny Woligroski to get an idea of what the what the new chips have to offer and how to get the most out of them.

As the executive correctly highlights, integrated GPUs rely entirely on the memory subsystem in your machine meaning your system RAM is also what powers your graphics. Cutting corners here, say with slower memory, could impact day to day tasks as well as in-game performance.

That said, Woligroski said going with dual-channel RAM is an absolute must if you want to maximize an 8000 series chip's capabilities. He also recommends going with at least DDR5-6000 as it is still reasonably affordable and you will get the best frame rates with it.

Woligroski also spoke on the benefits of an APU in a desktop versus a laptop. Namely, the desktop variant puts a lot more power at your disposal. All of them are 65W TDP parts, and he said that power is pretty linear with graphics in these parts. The desktop platform also provides a lot more headroom with regard to thermals so you will not be constrained by hot temperatures like you would with a laptop (assuming you are running a decent heatsink and good case cooling, of course).

AMD introduced four new 8000G chips at CES including the Ryzen 7 8700G, the Ryzen 5 8600G, the Ryzen 5 8500G, and the Ryzen 3 8300G. All but the Ryzen 3 will be available directly to consumers, with the Ryzen 3 being limited to OEM system builders.

Pricing starts at $180 for the 8500G and scales up to $230 for the 8600G and $330 for the 8700G, are are schedule to launch on January 31. As for whether or not an 8000 series APU is right for you, you will want to check out Tim's recent write-up for a deeper dive.

Permalink to story.

 
You can get about 60GB/s with DDR5 and AMD's Zen 4 memory controller on AIDA. That is what I am seeing. Good jump over DDR4 and write speeds in particular on Zen 3. Probably accounts for a lot of the improved graphics performance by itself.
 
You can get about 60GB/s with DDR5 and AMD's Zen 4 memory controller on AIDA. That is what I am seeing. Good jump over DDR4 and write speeds in particular on Zen 3. Probably accounts for a lot of the improved graphics performance by itself.

Agreed as the RX 6400 has the same 768 core count as the 680M and 780M but 128GB/s memory, and is usually 30% faster despite it's slower core clock.
 
Would love to see an APU generation with 3d Stacked cache or built in IC to help the GPU portion really stretch its legs.

I guess we'll see it when Intel actually poses a serious threat to AMD's APU line-up.
 
The poor thing is still memory starved. The 7840u shows scaling well into the 8000 mhz range with LPDDR5X and I'd easily assume the new 8000s with the 9500 mhz option will be even faster. These desktop chips also have higher TDPS and better cooling to maintain higher GPU speeds. They are CRYING for more bandwidth.
 
The poor thing is still memory starved. The 7840u shows scaling well into the 8000 mhz range with LPDDR5X and I'd easily assume the new 8000s with the 9500 mhz option will be even faster. These desktop chips also have higher TDPS and better cooling to maintain higher GPU speeds. They are CRYING for more bandwidth.



I wonder how the hell AMD are going to feed the Strix Halo with it's 40CU iGPU. They will never have anywhere near enough bandwidth to feed that beast.
 
I wonder how the hell AMD are going to feed the Strix Halo with it's 40CU iGPU. They will never have anywhere near enough bandwidth to feed that beast.
Just connect the iGPU to the I/O die using PCIe lanes in the infinity fabric
 
I wonder how the hell AMD are going to feed the Strix Halo with it's 40CU iGPU. They will never have anywhere near enough bandwidth to feed that beast.

From what I heard Strix Halo will use 256-bit LPDDR5X..... It has double memory bandwidth of regular APU that use 128-bit LPDDR5X

It still low for 40CU GPU though...
 
I wonder how the hell AMD are going to feed the Strix Halo with it's 40CU iGPU. They will never have anywhere near enough bandwidth to feed that beast.

It will almost certainly be on-package 256-bit bus width RAM (custom LPDDR5X?), probably with fanout packaging like RDNA3's Navi 31/32. Will simplify board designs, but kills option for user-upgradable memory after build.

Seems to be AMD's response to Apple's M3 Max (also 40 GPU "cores"), so may offer similar bandwidths or 400GB/s aggregate. This can be achieved with custom memory controllers/RAM chips and a memory-attached cache for bandwidth amplification (such as Infinity Cache). Something like MCDs containing cache with RAM chips mounted directly on top.

Just hope minimum configuration is 32GB up to 128GB (or 128-256GB in a workstation).
 
It will almost certainly be on-package 256-bit bus width RAM (custom LPDDR5X?), probably with fanout packaging like RDNA3's Navi 31/32. Will simplify board designs, but kills option for user-upgradable memory after build.

Seems to be AMD's response to Apple's M3 Max (also 40 GPU "cores"), so may offer similar bandwidths or 400GB/s aggregate. This can be achieved with custom memory controllers/RAM chips and a memory-attached cache for bandwidth amplification (such as Infinity Cache). Something like MCDs containing cache with RAM chips mounted directly on top.

Just hope minimum configuration is 32GB up to 128GB (or 128-256GB in a workstation).
If it uses LPDDR5X from micron at the new 9.6Gbps speed, on a 256 bit bus it would have just over 300GB/s of bandwidth. Most solutions for the 780m are between 102-120 depending on what speed LPDDR5X is used.

It would be a monster laptop chip. I want one.
 
Pair it with itx m'board and sfx psu, for maximum sff taste..
put mini gpu on it, for maximum console-like taste..
 
Back