back
87 comments
Will be interesting to see how long this RAM insanity will last. If it doesn't calm down before Zen 6 releases, people like me on older platforms might just have to skip Zen 6 entirely and wait for the AM6 platform.
Reportedly, Zen 7 will also be on AM5.

https://overclock3d.net/news/cpu_mainboard/amd-extends-am5-l...

Can they double the memory lanes without switching socket ? If not I feel like PC is going to fall behind even further compared to Apple chips. Having ram on chip sucks for repairability but 500gb/s main ram bandwidth is insane.

They stumbled into the right direction with strix halo but I have a feeling they won't recognize the win/follow up.

The "insane" RAM bandwidth makes sense with Apple M chips and Strix Halo because it's actually "crap" VRAM bandwidth for the GPU. What makes those nice is the quantity of memory the GPU has (even though its slow), not that the CPU has tons of RAM bandwidth.

When you go to the desktop it becomes harder to justify including beefed up memory controllers just for the CPU vs putting that towards beefing some other part of the CPU up that has more of an impact in cost or performance.

No, but they can skip the socket, much like many of the mini-pcs/SFFs that include laptop chips in small desktops. Strix halo already doubled the memory channels and the next gen is supposedly going to move the memory bus from 256 bits wide to 384 bits.
Not easily, and you will need a new motherboard anyhow because each of the 2 slots you can have per lane are wired in tandem.
The socket io locks in the amount of memory channels. Some pins could be repurposed but that's pretty much a new socket anyway.

They could in theory do on package dram as faster first level memory, but I doubt we'll see that anytime soon on desktop and it probably wouldn't fit under the heat spreader

> Can they double the memory lanes without switching socket?

Sure. Keep the DIMM sockets and add HBM to the CPU package.

Actually probably the best possible architecture. You can choose to have both or only one, backward compatible and future proof.

Yes, it adds another level to the memory hierarchy but that can be fine tuned.

So Zen 6/7 will have a core design and a CCD design. But like past gens, these will be packaged into different products with different sockets and packages (everything from monolithic APUs to sprawling multi-chiplet Server cpus).

So to say that Zen 6/7 supports AM5 on desktop, doesn't necessarily exclude that Zen 6/7 product family in general doesn't support other new/interesting sockets on desktop (or mobile) also. Maybe products for AM6 and AM5 from the same zen family.

Medusa Halo and the Zen7 based 'Grimlock Halo' version might be the interesting ones to watch (if you like efficient Apple-stlyle big APUs with all the memory bandwidth)

Higher DRAM prices might mean that there is less demand from new system builders mean depressed prices so it might be more tempting to upgrade your existing AM5 CPU to Zen 6
> less demand from new system builders mean depressed prices

Only if they overestimate demand and overproduce CPUs. Otherwise it will lead to higher prices because there's less economy of scale.

I would figure the opposite. There are plenty of people like me staying on AM4 because of the RAM price increases. I will probably skip AM5 entirely.
I am a hypocrite, but there is really not that much need to upgrade CPUs anymore. Even a ten year old chip seems completely adequate for day to day use. I played with a N100 recently and those things are incredibly capable.

(Ignore my AM5 workstation with 192GB RAM in the corner)

But they are still gonna fab the Zen 6 chips. So for people already with AM5 motherboards populated with RAM but rocking a Zen 4 CPU this could be a good time to upgrade that CPU with your existing setup. You passing this generation just means less competition for those CPUs which should make them even cheaper.
and do what, buy now-hideously expensive DDR6?
Hopefully it settles down soon. DDR4 prices are climbing now as well since more people are sticking with it.

I'd love to build a new desktop soon but I couldn't justify the cost and am instead building out a used desktop that's still on ddr4 / lga1151.

Holy ram prices man!

I just checked how much the 64 Gb ddr4 in my desktop would cost now... it starts at 2.5 times what i paid in 2022.

Sorry AMD, I would maybe like a new desktop but not now.

I hope they'll release a new AM4 CPU

Something like 5900x on 2nm or 4nm

I’m sure there are a plethora of technical reasons it’s impractical - but my dream is a big, unified L3 cache across their CCD chiplets. Maybe 256mb in size for the x950 x3d chips.
There are challenges with really big monolithic caches. IBM does something sort of like your idea in their Power and Telum chips, with different approaches. Power has a non-uniform cache within each die, Telum has a way to stitch together cache even across sockets (!).

https://chipsandcheese.com/p/telum-ii-at-hot-chips-2024-main...

https://www.eecg.utoronto.ca/~moshovos/ACA07/projectsuggesti...

(if you do ML things you might recognize Doug Burger's name on the authors line of the second one)

They could bond multiple CCDs on top of a single large unified L3 die (similar to MI300C) if they wanted to. I've seen no rumors about that though.
I'm currently cache limited by my work and I share your dream
I hope for a little more PCIe lanes so I can run 2 gaming VMs on these and upgrade my old Threadripper.
There is fuck all difference between x8 and x16 for gaming. Heck with PCIe5 even dropping to x4 is borderline noticeable outside of benchmarks.
100% this

The PCI-Express bus is actually rather slow. Only ~63 GB/s, even with PCIe 5 x16!

PCIe is simply not a bottleneck for gaming. All the textures and models are loaded into the GPU once, when the game loads, then re-used from VRAM for every frame. Otherwise, a scene with a lowly 2 GB of assets would cap out at only ~30 fps.

Which is funny to think about historically. I remember when AGP first came out, and it was advertised as making it so GPUs wouldn't need tons of memory, only enough for the frame buffers, and that they would stream texture data across AGP. Well, the demands for bandwidth couldn't keep up. And now, even if the port itself was fast enough, the system RAM wouldn't be. DDR5-6400 running in dual-channel mode is only ~102 GB/s. On the flip side the RTX 5050, a current-gen budget card, has over 3x that at 320 GB/s, and on the top end, the RTX 5090 is 1.8 TB/s.

> All the textures and models are loaded into the GPU once, when the game loads, then re-used from VRAM for every frame. Otherwise, a scene with a lowly 2 GB of assets would cap out at only ~30 fps.

Ah, not really these days, textures are loaded in/out on demand, at multiple different mipmap levels, same with model geometry and LOD's. There is texture and mesh data frequently being cached in and out during gameplay.

Not arguing with your points around bus speeds, and I suspect you knew the above and were simplyifing anyway.

You are correct that games generally are not PCIe limited. But you are incorrect that games just upload everything ones and be done. Most modern engines are most certainly streaming in and out assets all the time.
Main problem seems to be they're kinda badly utilized (IMHO) on many motherboards. Most seem to go with two x16 slots so you get x8 lanes in both.

There are some exceptions, but I haven't seen one with for example four x16 slots that support PCIe 5.0 x4 lanes with bifurcation.

You can buy add-in cards that do lane bifurcation

E.g. https://www.ebay.co.uk/itm/126656188922

Most motherboards don’t go beyond 2x8 with 2x16 physical slots because there is little actual use for it and it costs quite a bit of money.

The biggest difference for me for PCIe 5.0 has been additional bandwidth for my M2 drive.
Your comment is basically the "tl;dr" of this Techpowerup article (which is great and people should read it if they are unconvinced or curious): https://www.techpowerup.com/review/nvidia-geforce-rtx-5090-p...
You're not getting more lanes without a new socket. Or a PCIe switch, which is expensive.
This. I needed a high speed link between two PCs and bought a mellanox card, cue me surprised that a consumers PCs do not have enough PCIe lanes to handle both a thickboii GPU and a thickboii 200GBe mellanox card...
for that you need new socket and motherboard. you need to physically route those extra lanes to pcie slots or other components
And even when AMD does move their mainstream desktop processors to a new socket, there's very little reason to expect them to be trying to accommodate multi-GPU setups. SLI and Crossfire are dead, multi-GPU gaming isn't coming back for the foreseeable future, so multi-GPU is more or less a purely workstation/server feature at this point. They're not going to increase the cost of their mainstream platform for the sole purpose of cannibalizing Threadripper sales.
Had to look up what vm gaming is. What's your motivation? If you don't mind sharing.
"7 GHz clock speed"

When did the GHz race start again?

It never stopped.

Just takes backwards steps from time to time with major architectural innovations that deliver better performance at significantly lower clock speeds. Intel's last backwards step was from Pentium 4 to Core all the way back in ~2005. AMD's last backwards step was from Bulldozer (and friends) to Zen in 2017.

7GHz is ridiculous and probably just a false rumour, but IMO; Intel and AMD are probably due for another backwards step, they are exceeding the peek speeds from the P4/Bulldozer eras. And Apple has proved that you can get better performance at lower clock speeds.

Intels plan for P4 was to scale to 10Ghz. Its always been a race but plans don't always work out.
Rumors = the author just made something up
Similarly:

Leaks = the author just made something up, but now it ranks extra highly when someone searches for "[upcoming thing] leaks"

I remain quite skeptical of that. Maybe on a purpose built overclocking rig :^)
Yeah, first of all we need to get 6 GHz with Zen 6.
They should be reintroducing the 3D vcache [0] variants (X) in EPYC, with a higher cache/core ratio, that was present in EPYC4 (e.g. 9684X [1]) they for some reason wasn't available in EPYC5.

Makes a massive difference at high density and utilisation, with the standard cache/core performance can really degrade under load.

[0] https://www.amd.com/en/products/processors/technologies/3d-v...

[1] https://www.amd.com/en/products/processors/server/epyc/4th-g...

> This increases the maximum core count per chiplet from 8 to 12. Furthermore, it increases the L3 cache per CCX/CCD from 32 MB to 48 MB.

I'd say the amount of L3 is not increased but adapted/scaled to the increased core count, since per each core there is still the same amount of cache available as before.

We get faster cores, so we need to get from 5600 to e.g. 6000 DDR5. Since core count is increased by 50%, we'd need 9000... DDR5^W, well yes, we'd need actually as planed before AM6 and DDR6!

There are already DDR5 CUDIMMs at and above 8000 MT/s, and 9600 MT/s has been demonstrated but none are currently in stock. By the time AMD ships Zen 6 desktop processors, the market should be ready with memory modules that will mean even the highest core count Zen 6 parts will be at worst only slightly more bandwidth-starved than their predecessors. And the lower core count Zen 6 CPUs with a single CCD should be able to provide substantially more bandwidth than their predecessors. All without requiring DDR6 yet.
By the time Zen6 launches, there will already be RVA23 chips in the market.

x86 releases will never again be as interesting.

Well with the way ram is going my next buy may well land on zen 6/ddr6/pcie6