Poll: Ryzen 7950X3D, 7900X3D, 7800X3D

[Grim5]

No chance.

the 5800X3D is currently AMD's best selling CPU, even outselling the 7xxx series. AMD want people to upgrade/choose AM5 based hardware as their next choice.

Extending 3D Cache to 5900 and 5950 CPU's would just keep people on the AM4 platform. There would be no incentive to go AM5 as 3D versions of the 5900 and 5950 would significantly close the gap between the 7xxx series, especially in gaming.

As a 5800X3D owner, I'd love to see it though.

And motherboard makers really hate it

As I learnt today, about 30% of the price of an AM5 motherboard is just extra margin that all manufacturers added this time around to cover future sales losses. The AIBs really hate that AM4 lasted so long, they don't like that users can buy one board and keep slotting in new CPUs for many years and they don't get any new revenue from the users. And because AM5 is doing the same thing again, supporting the next 5 years of CPUs, the AIB all added 30% price premium to AM5 so they can get more money up front due to future revenue losses.

What AIBs would really like is a way to lock compatibility behind software but AMD won't let them. For example if you want to use a Zen5 in a x690e board you can, but you pay a a fee, maybe like $50 or $100 and then you get given the bios update that enables it. Something to that affect, but AMD forces motherboard makers to offer several years of support and future CPu compatibility from day one and they can't get any money for it so now they just make the boards really expensive

 

[BS Dave]

I was thinking about the caches the other day. If AMD can get really good yields what's to stop them taking all the caches of the CPU die and lumping them on top? They could raise Level 1 2 and 3 caches and the die size could be smaller or extend the original die to 12 cores...

 

[rumple9]

But AMD have said AM5 is only supported to 2025 which is only just over 2 years. They might follow the intel model going forward new socket every 2 years

 

[Grim5]

I was thinking about the caches the other day. If AMD can get really good yields what's to stop them taking all the caches of the CPU die and lumping them on top? They could raise Level 1 2 and 3 caches and the die size could be smaller or extend the original die to 12 cores...

Heat is a problem when you stack transistors on top of each other. Intel is currently researching how to do in between layer liquid cooling, AMD probably as well but that's the issue with just adding more and more layers stacked

 

[mid_gen]

There are already games where the 5800x3d is no faster or even slower than the 5800x. Those are games where the game data already fits inside the existing cache so extra cache has no benefit. Cache doesn't provide 1:1 scaling, performance only improves as you lower latency by fitting more of the game in the cache, once everything is in there there is no further improvement from adding extra cache. The only reason you see EPYC and Xeon CPUs now moving towards hundreds and thousands of MB of cache is because those are used to multitask

Cache is also a one trick pony, once all CPUs have lots of cache there won't be any further gains to be had outside of cache getting faster and lower latency through other improvements, simply adding more cache won't do anything just like a PC with 128GB RAM doesn't run games any faster than a PC with 16GB RAM

It's nothing to do with fitting all the game data in cache.

The difference is how the game is accessing data. If it's written such that it's making random reads (usually due to lots of indirection), then it doesn't matter big your cache is....the CPU is going to sat waiting for data to be fetched from RAM all the time.

If the game is written in a decent data-oriented fashion, organising logic such that it's doing lots of sequential reads, then the prefetch engine can work effectively and keep the cache stuffed with data *before* the CPU needs it, so there's minimal wasted cycles.

Simulations (like MSFS) lend themselves to data oriented designs as you typically have lots of data that needs unconditional logic running over it, with no branches, like the weather and fluid simulations.

 

[Greebo]

And motherboard makers really hate it

As I learnt today, about 30% of the price of an AM5 motherboard is just extra margin that all manufacturers added this time around to cover future sales losses. The AIBs really hate that AM4 lasted so long, they don't like that users can buy one board and keep slotting in new CPUs for many years and they don't get any new revenue from the users. And because AM5 is doing the same thing again, supporting the next 5 years of CPUs, the AIB all added 30% price premium to AM5 so they can get more money up front due to future revenue losses.

What AIBs would really like is a way to lock compatibility behind software but AMD won't let them. For example if you want to use a Zen5 in a x690e board you can, but you pay a a fee, maybe like $50 or $100 and then you get given the bios update that enables it. Something to that affect, but AMD forces motherboard makers to offer several years of support and future CPu compatibility from day one and they can't get any money for it so now they just make the boards really expensive

AM5 is only three years though, not 5 years. So only one year longer than Intel. I suspect future sockets will be no more than 3 years as well.

And your extra margin makes no sense. The issue with AM5 boards is there are very few mid priced ones and even less budget ones. If you look at the higher end boards out and compare them to the (now eol) z790 Intel boards, the equivalent AMD board seems to be £50 to £100 cheaper than the Intel board. If board manufacturers are adding 30% because AMD are using that socket for 5 years (they arent, its 2025 only), why are they adding even more to the Intel boards when they are getting a new socket next year??

But if its true that AMD board manufacturers are adding 30% like you say, then Intel Board manufacturers are adding 40% for just no reason whatsoever and are really gauging every bodies eyes out.

To me the issues seem to be PCI-e5 and ddr5. That is what seems to hike up the prices, even between Intel boards. You want 3 pcie-5 slots and 5 M2 slots with 2 M2 pcie-5 and 3 M2 pcie-4 then its going to cost you.

IF Am5 socket cpus had supported ddr4 then i am sure we would have seen lots of cheap motherboards with ddr4, pci-4 slots and a couple of pci-4 M2 or even pci-3 M2. Okay, chances are over the next 3 (or 5) years that socket is in use, people would perhaps have looked at upgrading their motherboard again for the extra features.

 

[Worthy]

It's nothing to do with fitting all the game data in cache.

The difference is how the game is accessing data. If it's written such that it's making random reads (usually due to lots of indirection), then it doesn't matter big your cache is....the CPU is going to sat waiting for data to be fetched from RAM all the time.

If the game is written in a decent data-oriented fashion, organising logic such that it's doing lots of sequential reads, then the prefetch engine can work effectively and keep the cache stuffed with data *before* the CPU needs it, so there's minimal wasted cycles.

Simulations (like MSFS) lend themselves to data oriented designs as you typically have lots of data that needs unconditional logic running over it, with no branches, like the weather and fluid simulations.

That was really insightful, thanks. You should do more of these types of posts with your experience in the industry

 

[Curlyriff]

And motherboard makers really hate it

As I learnt today, about 30% of the price of an AM5 motherboard is just extra margin that all manufacturers added this time around to cover future sales losses. The AIBs really hate that AM4 lasted so long, they don't like that users can buy one board and keep slotting in new CPUs for many years and they don't get any new revenue from the users. And because AM5 is doing the same thing again, supporting the next 5 years of CPUs, the AIB all added 30% price premium to AM5 so they can get more money up front due to future revenue losses.

What AIBs would really like is a way to lock compatibility behind software but AMD won't let them. For example if you want to use a Zen5 in a x690e board you can, but you pay a a fee, maybe like $50 or $100 and then you get given the bios update that enables it. Something to that affect, but AMD forces motherboard makers to offer several years of support and future CPu compatibility from day one and they can't get any money for it so now they just make the boards really expensive

What's their excuse for the Intel 700 series boards that are similar prices?

 

[uscool]

They do say 2025 and beyond, maybe didn't wanna commit if things change ?

Still getting more than what intel would offer that's confirmed , zen4 Inc x3d and zen5 Inc x3d

 

[BS Dave]

Heat is a problem when you stack transistors on top of each other. Intel is currently researching how to do in between layer liquid cooling, AMD probably as well but that's the issue with just adding more and more layers stacked

I was just thinking about the 2 layers they have now

 

[Dg834man]

I was thinking about the caches the other day. If AMD can get really good yields what's to stop them taking all the caches of the CPU die and lumping them on top? They could raise Level 1 2 and 3 caches and the die size could be smaller or extend the original die to 12 cores...

They are planning zen4c 16 cores per chiplet with seriously depleted l3 cashe, doesn't have stacked cashe though, could do a 3d stacked zen 4 and zen 4 c chiplet 24 cores hybrid beat intel at their own game.

 

[Iron Man]

This will be AM5 I presume?

Yep

 

[Dicehunter]

Can't wait to pickup a 7950X3D on release, assuming it performs as well as we expect!

My 13900k will be donated to my VR/family rig (Nieces will be happy!), until then I'm enjoying getting the most from my 4090 with the 13900k

13900K will be a trash tier CPU once the 7950X3D comes out, Mine will be going into my HTPC where it will be used exclusively for films.

 

[Grim5]

Can we ban Dave if he doesn't show evidence of buying a 7950x3d?

 

[Dave2150]

13900K will be a trash tier CPU once the 7950X3D comes out, Mine will be going into my HTPC where it will be used exclusively for films.

I don't think the 13900k is powerful enough for HTPC, it's probably best used as a keyring

 

[Dave2150]

Can we ban Dave if he doesn't show evidence of buying a 7950x3d?

Probably best to ban those infatuated with other forum members purchasing decisions

 

[Dicehunter]

I don't think the 13900k is powerful enough for HTPC, it's probably best used as a keyring

I was thinking about relegating it to a dedicated MS Paint box !

 

[Journey]

Probably best to ban those infatuated with other forum members purchasing decisions

If it were only your purchasing decisions that were the issue, then no one would care, in fact go about your business merrily, it's the tripe you post along with those decisions that people take issue with.

Oh right, you probably can't read this though, as you put me on ignore for explaining how PCIE lanes work properly to you. Talk about putting your fingers in your ears and going la-lala-la.

 

[Tee Hee Johnson]

I was thinking about the caches the other day. If AMD can get really good yields what's to stop them taking all the caches of the CPU die and lumping them on top? They could raise Level 1 2 and 3 caches and the die size could be smaller or extend the original die to 12 cores...

The reason they are called L1 - L3 is due to proximity to the cores.

L1 being the closest and L3 being the furthest. Taking L1 and L2 and putting them above the core (like L3) defeats the purpose of super fast L1 and L2.

 

[Tee Hee Johnson]

There are already games where the 5800x3d is no faster or even slower than the 5800x. Those are games where the game data already fits inside the existing cache so extra cache has no benefit. Cache doesn't provide 1:1 scaling, performance only improves as you lower latency by fitting more of the game in the cache, once everything is in there there is no further improvement from adding extra cache. The only reason you see EPYC and Xeon CPUs now moving towards hundreds and thousands of MB of cache is because those are used to multitask

Thats not really how cache works, its somewhat like common bits of data output from commonly done instructions.

It doesnt try to 'fit' game data as such in there.