AMD to unveil Zen 4 CPUs at CES 2022

[PaulC2K]

I think it will be worth it for people with slow DDR4 RAM kits. Also, for system builders who haven't upgraded in a long time.

I think it will probably make sense just to buy a mid end CPU, then just upgrade when the next gen arrives on AM5. It looks like AMD could have the best DDR5 support also, possibly supporting DDR5 frequencies of 7000-8000mhz.

Thats the angle im coming at. I dont think im in a bad spot (3600x & 1080) but i need to pick up a new GPU this year and this feels like a good time to make a complete switch. Ive been wanting more ram for a while, but it feels a little wasteful when the next CPU will want ddr5, and my current ram can go into my server.
I just wonder whether i'd rather hold out a cycle longer and wait for the 2nd gen, or x3D release.

The timing seems like a good time to do major upgrades, with the new socket, pcie5 & ddr5. Just need to wait and see whether the performance aligns with that.

From you end or just because of the way the market seems to be right now?

I'm going to buy something from the next generation, RTX 4070 / RX 7700XT, hopefully if pricing isn't daft, and then i'm done for a few years, i don't buy many, if any AAA games these days, its all the same crap regurgitated with fried cheese on top.

The only thing that matters to me is a long tern good for Star Citizen GPU, once have that that's me done.

What res are you running at, and how does your current system handle in SC?
Im not gaming much these days, but just picked up a Samsung G9 on MM the other week and my fps is understandably laughable, but stable 30fps at 2560x1080. I'd play other games, but like you said, AAA has been underwhelming for a while, and while i dont play SC often at all, at least thats one benchmark thats relevant to me going forward.
The 4070 comment interests me, cos i feel like the G9 is gonna need a 4090, so maybe im under estimating existing GPU performance, and just need something like a 4080 and a CPU that isnt bottlenecking it.

 

[Purgatory]

It could be the other way round, at least on release. AMD have a higher tdp allowance with zen5 so it's entirely possible that they're literally maxing out the silicon etc to get those high boost clock speeds (basically what intel does) leaving very little if any room for the average home overclocker.

But then at the same time they have got the extreme boards so it could be that some chips won't have any headroom and others might have more... suppose we'll have to wait.

They have more power use due to the IGPU inside the cpus now, that's why the higher TDP.

 

[doodah]

What are people on OG x370/AM4 doing? I might do one more jump (3700X > 5800 X3D) and ride that for a couple more years before jumping to AM5.

 

[Journey]

What are people on OG x370/AM4 doing?

I've done stacks of upgrades for people, a lot who have been on 1xxx parts, and almost as many who were on 2xxx parts. The new B350/X370 BIOS's for 5xxx and the lower cost parts really has worked in AMD's favour for prolonging the life span of current systems while avoiding them from jumping to a competitor platform, or even waiting for AM5 which is not going to be offering as much direct value immediately.

5800X3D has driven a lot of interest, but it's the reduction in the 5600X, and the new 5700X that have been the most popular, followed by the ~£125 R5 5500, which although got slated by a lot of reviews, is a huge jump from an R5 1600 etc. and cost about £60 after selling the current CPU.

 

[Troezar]

I've done stacks of upgrades for people, a lot who have been on 1xxx parts, and almost as many who were on 2xxx parts. The new B350/X370 BIOS's for 5xxx and the lower cost parts really has worked in AMD's favour for prolonging the life span of current systems while avoiding them from jumping to a competitor platform, or even waiting for AM5 which is not going to be offering as much direct value immediately.

5800X3D has driven a lot of interest, but it's the reduction in the 5600X, and the new 5700X that have been the most popular, followed by the ~£125 R5 5500, which although got slated by a lot of reviews, is a huge jump from an R5 1600 etc. and cost about £60 after selling the current CPU.

The upgrade route is very underestimated, there are lots of users that don't want to do a big upgrade and just dropping in a new CPU is much less hassle and a lot cheaper.

 

[Journey]

The upgrade route is very underestimated, there are lots of users that don't want to do a big upgrade and just dropping in a new CPU is much less hassle and a lot cheaper.

Yes, and the other popular upgrade has been adding M.2 drives while doing the CPU, and sometimes some cheap/fast RAM as a replacement or an upgrade to 32GB. It's been certainly interesting seeing how happy people are, and the fact the is little to no disruption in continuity for their systems. I'd imagine a lot of the upgrades will be in use until 2025 and beyond, some 7-8 years after launch, and build.

It's great to see really, we create so much e-waste as it is, and if people can drop in a CPU vs. a building a whole new system then its a win-win IMO.

 

[welshrat]

The upgrade route is very underestimated, there are lots of users that don't want to do a big upgrade and just dropping in a new CPU is much less hassle and a lot cheaper.

Quite. It's one of the reasons I will stick with AMD for socket AM5 even if its not king to begin with. Motherboard and socket changes are a pain in the **** and if we get another 5 years or so support for it then its a no brainer.

I know others like to rebuild their machine's often however as I use mine for all my work and have so much installed I don't want the hassle, still rocking the same windows build I had with my Ryzen 2600 from 4 years ago and with a little TLC occasionally (dism and sfc) its still running 100% fine for me.

 

[Robert896r1]

anyone know if bclk will finally be decoupled now?

 

[timorous]

Those 8 E cores are worth 7,000 points.

12900K: 27,472
5950X: 28,577

27,472 + 7,000 = 34,472 + 10% = 37,919

28,577 + 31% = 37,435.

13900K: 37,919 (+1%)
7950X: 37,435

Where are you getting +31% from for the 7950X? It did blender in 31% less time than the 12900 but if you want to convert that to a score then if the 12900K scores 100 points the 7000 series part would score 145 points. (100 / 0.69 = 145) making it 45% faster. Since this was a short blender test the 12900K is slightly faster than the 5950X in that scenario so the uplift over the 5950X is closer to 50%.

IF that flows to CB MT then the 7950X would score 28,577 * 1.5 = 42,865.

Even if it does not quite flow for longer tests you could expect a 40% uplift for a similar style of workload making the 7950X around 40,000 points.

On the >15% CB ST increase the Zen 2 to Zen 3 increase for ST CB R23 was just +13% but the average was 19% so this looks like it could be on the low end of the ST gains if Zen 4 follows the same pattern as Zen 2 to Zen 3.

 

[Hook]

With all the I/O we'll hopefully see some "cheap" workstation motherboards again, with support for more than 1 card in PCIe 16x.

 

[ltron]

Where are you getting +31% from for the 7950X? It did blender in 31% less time than the 12900 but if you want to convert that to a score then if the 12900K scores 100 points the 7000 series part would score 145 points. (100 / 0.69 = 145) making it 45% faster. Since this was a short blender test the 12900K is slightly faster than the 5950X in that scenario so the uplift over the 5950X is closer to 50%.

IF that flows to CB MT then the 7950X would score 28,577 * 1.5 = 42,865.

Even if it does not quite flow for longer tests you could expect a 40% uplift for a similar style of workload making the 7950X around 40,000 points.

On the >15% CB ST increase the Zen 2 to Zen 3 increase for ST CB R23 was just +13% but the average was 19% so this looks like it could be on the low end of the ST gains if Zen 4 follows the same pattern as Zen 2 to Zen 3.

I got the 31% from what other people were saying AMD had said, but you are right it's actually over 45% faster and 31% less time according to the footnotes which is even more impressive.

 

[Grim5]

The presentation screen behind Lisa said "31% faster" not "31% less time"

But if the footnotes are saying something else that it's actually 45% faster than that's good.

It would make more sense given the 5950x stock runs around 3.8ghz and zen4 is supposed to do about 5ghz all core so multithread performance should be significantly improved even if single thread isn't. Using PBO on a 5950x bumps its score by 20% and it does 4.4ghz so going to 5ghz would provide between 30% and 40% more performance then just add zen 4 IPC and voila, 45% higher multithread performance than 12900k and 40% higher than 5950x

 

[g67575]

The upgrade route is very underestimated, there are lots of users that don't want to do a big upgrade and just dropping in a new CPU is much less hassle and a lot cheaper.

If PC builders have already got a good powersupply and SSD, it should be possible to get DDR5 16GB kits for £100-£110, and a B650 board for about another hundred. So, the cost will mostly depend on how much people decide to spend on RAM.

It looks like the B560 boards come with with PCIE5 (storage drives only) for NVME SSDs also.

DDR5 5200mhz 8gb modules cost as little as £53 now.

 

[Troezar]

If PC builders have already got a good powersupply and SSD, it should be possible to get DDR5 16GB kits for £100-£110, and a B650 board for about another hundred. So, the cost will mostly depend on how much people decide to spend on RAM.

Agreed, it depends on your setup. To replace my motherboard (X570 Aorus Master) and 32GB of 8 Pack RAM would be expensive, especially to get similar latency with DDR5. Add to that if you have a decent AM4 CPU you aren't going to need an upgrade for some time. Don't tell Gibbo I said that though

 

[Armageus]

With all the I/O we'll hopefully see some "cheap" workstation motherboards again, with support for more than 1 card in PCIe 16x.

All what I/O? Doesn't Zen4 still only has 24 PCI-E lanes?

 

[Troezar]

All what I/O? Doesn't Zen4 still only has 24 PCI-E lanes?

Aren't 24 PCIE 5 lanes the same bandwidth as 48 PCIE 4 lanes?

 

[Journey]

Aren't 24 PCIE 5 lanes the same bandwidth as 48 PCIE 4 lanes?

Or 96x PCI-E 3.0 lanes.

 

[Armageus]

Aren't 24 PCIE 5 lanes the same bandwidth as 48 PCIE 4 lanes?

They are, but you've still only got so many to go around

4x for the PCI-E 5.0 NVME is mandated I believe.

Leaving:
16x PCI-E 5.0 for the Graphics card

which still only leaves you with 4x PCI 5.0 lanes (or 8x PCI-E 4.0 or 16x PCI-E 3.0)

Even if lanes end up being split/reduced to lower speeds, then it's doubtful they'll be many boards that do, as it'll incur additional cost with splitter chips and the like

 

[Troezar]

It seems like only 5 minutes ago many were saying X570 and PCIE 4.0 was unnecessary and to be honest I am nowhere near saturating that. I don't see the vast majority needing it until we are steaming assets from storage for gaming.

 

[Troezar]

They are, but you've still only got so many to go around

4x for the PCI-E 5.0 NVME is mandated I believe.

Leaving:
16x PCI-E 5.0 for the Graphics card

which still only leaves you with 4x PCI 5.0 lanes (or 8x PCI-E 4.0 or 16x PCI-E 3.0)

Even if lanes end up being split/reduced to lower speeds, then it's doubtful they'll be many boards that do, as it'll incur additional cost with splitter chips and the like

Ok makes sense, you get 4 lanes regardless of whether the bandwidth is overkill. Seems to be a flaw in the idea of having more and more bandwidth but only the same number of lanes...