• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Zen 6 rumours

alec said:
In light of
a) Zen6 probably being delayed to end of 2026 and
b) rumors of reworked memory controller for higher speeds
there will definitely be a new chipset and a generation of motherboards
that probably allow DDR5 to hit higher speeds up to 10000
Click to expand...
I wish they would go quad channel already.
 
jigger said:
They did and then some. TR is now upto 8 channels.
Click to expand...
I think you known I was talking about desktop, Threadripper is not that. Even Strix Halo is quad channel. I would like to see HEDT's return, quad channel RAM(not RDIMM's), more PCIE off the CPU(48-64 lanes), priced so normal desktop users can actually get them. Pushing RAM frequency increases power/heat and cost, my old X99(10 year+) had simillar bandwidth and more PCIe to a modern mainstreem desktop and it did not cost £5K+.
 
Last edited:
FredFlint said:
I think you known I was talking about desktop, Threadripper is not that. Even Strix Halo is quad channel. I would like to see HEDT's return, quad channel RAM(not RDIMM's), more PCIE off the CPU(48-64 lanes), priced so normal desktop users can actually get them. Pushing RAM frequency increases power/heat and cost, my old X99(10 year+) had simillar bandwidth and more PCIe to a modern mainstreem desktop and it did not cost £5K+.
Click to expand...

Yes I was aware, but if AMD added a pair of memory channels to AM5 we end up at Soc TR5 and 90% of the costs that come with it

So, X99 wasn’t priced to the desktop. The lowest end X99 system part was 10x the lowest Soc 115X system.

If you look at the topology of the Stix parts, you’ll not the memory system is not there to help the CPU but the GPU portion of the APU. The access the CPU cores have to main memory is limited to that of the GPU side.
 
Some of the ThreadRipper/Xeon boards are not that expensive considering the spec vs the high end desktop boards. The main problem is the cost of the CPU's and RAM. Doing a CPU with quad channel desktop RAM and 48-64 PCIe lanes (from the CPU) would fill the gap. I paid £450 for a 5930K and £300 for a ASUS X99 Deluxe, I don't remember how much the normal gear was but I think X99 was ~50% more for CPU+Mobo, think the 5820K was ~£300. Some X99 boards cost < £200. Anyway, I don't think it will happen anytime soon, it would be nice though.
 
Perfect_Chaos said:
Though I have to wonder what will be restrictions on RAM speed? If Zen 6 can take 8000+ and current boards can run that speed then will it?

And if not, how much will it impact performance?
Click to expand...

So you can run PC8000 on most AM5 boards for some time now, it’s just been pointless outside of the APU’s as the CPUs don’t gain much performance after a point. The 24c48t parts will likely show gains from faster memory.
 
jigger said:
So you can run PC8000 on most AM5 boards for some time now, it’s just been pointless outside of the APU’s as the CPUs don’t gain much performance after a point. The 24c48t parts will likely show gains from faster memory.
Click to expand...
Is the lack of extra performance due to interconnects/infinite fabric/latency though? If they improve all that, extra memory speed will make more of a difference no?
 
Last edited:
BS Dave said:
Is the lack of extra performance due to interconnects/infinite fabric/latency though? If they improve all that, extra memory speed will make more of a difference no?
Click to expand...

You can always tweak latency, but from what I’ve seen memory bandwidth (which is essentially what we’re talking about) isn’t a great factor until you have enough processing demand from cores. You see the difference in memory speed on the Soc SP/TR parts where running faster memory can shows gains, but those parts use a pretty different SOC configuration to Ryzen.

I suppose a lot will depend on how much quicker the next Zen architecture is and how far AMD increases its tread count.
 
TSMC’s 2nm fab. process is looking good. It will be used in Zen 6 desktop CPUs (due for release late in 2026) according to the latest roadmap (already used for EPYC CPUs). The transistor density is higher than Intel’s 18a process, which is being used in Panther Lake CPUs, and presumably Nova Lake as well:

2 nm process - Wikipedia

en.wikipedia.org en.wikipedia.org

The transistor density of TSMC’s 2nm fab. process is roughly double that of their previous 5nm and 4nm fab. processes:

5 nm process - Wikipedia

en.wikipedia.org en.wikipedia.org
 
Last edited:
There isn't much improvement from 3nm to 2nm, so might be better than intel but dont expect too much if you already have a 3nm zen

But its worth keeping in mind CPU's arent monolithic anymore, so not the whole CPU is 2nm. AMD can claim its a 2nm CPU but not the while CPU is 2nm. same as Intel can claim a CPU is 18a, but not the whole CPU is
 
Last edited:
The part that counts, will presumably be 2nm.

Zen 4 used TSMC 5nm, Zen 5 was a small improvement on that I think.

It’s a step up from 4/5nm, anyway.

The best that can be built with current technology.
 
Last edited:
20% would be great

Zen5 was called "Zen5%" for a reason, because it was about 5% faster than Zen4, so 20% sounds great compared to 5%
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom