News AMD Unveils Three Ryzen 7000X3D V-Cache Chips, Three New 65W Non-X CPUs, Too

[bit_user]

I'm curious to know which process node the stacked SRAM dies are using. I'll bet probably TSMC N6, like the RDNA3's MCDs.

Given what's come out about poor SRAM scaling, I wonder if AMD is moving to an arrangement where they'll put far less L3 cache on their compute dies, and only their low-end CPUs won't have stacked SRAM.

 

[Elusive Ruse]

My 4x8gb kit of CL 14-15-15-36 3600mghz G skill RAM that I got for less than a comparable DDR5 kit begs to differ. I currently have it at tighter than stock timings as well.

Correct me if I'm wrong but a 32GB 3600 CL14 kit costs over $300, you can't possibly believe that's a valid argument.

 

[rDigital]

With one of those 7950X3D I could run CS:GO @ 780 FPS 1080p instead of a measly 590 FPS!

 

[Roland Of Gilead]

7800x3d boost dissapointing vs the others. I hope it is actually more than that, and it not an upsell to the higher models. I don't want 2 ccds

Yeah, seems like a big drop compared to the others! Doesn't really add up as they are already segmented enough.

 

[blacknemesist]

X3D model at 120W? Damn. No value lost on my 7700X since I locked it at 105W to keep it under 80ºC at all times
Value was lost with these CPUs because they will thermal throttle anyway forcing you to either keep it at 95ºC(+ the added fan noise) or locking it at a lower power consumption.
In real world games the 5950X probably does not lose by much at higher resolutions and with 2k, 4k and those weird ultrawides becoming more and more common the weight on the CPU is a lot less. It also does not required all the top end gear that does not do much at all for gaming.

 

[Roland Of Gilead]

Great, more hardware that goes on a high priced platform (that does not seems to go down any time) and thats not available where I live, or if its, is many times more expensive than MSRP.

At least books are still cheap, and doesn't need cooling nor power to work.

Eh, this argument only stands because Raptor supports DDR4. With Raptor being the end for that platform, do you think Intel will still support DDR4 on Metoer? My guess is no. Then both platforms will be roughly the same in cost. That of course doesn't include the longevity with AMD Zen 3/Zen 4. Which will support new CPU's up to 2025 IIRC.

I would agree that AMD shot themselves in the foot by not supporting DDR4 for Zen 4. But it will be a well established ecosystem for future CPU's.

 

[RodroX]

Eh, this argument only stands because Raptor supports DDR4. With Raptor being the end for that platform, do you think Intel will still support DDR4 on Metoer? My guess is no. Then both platforms will be roughly the same in cost. That of course doesn't include the longevity with AMD Zen 3/Zen 4. Which will support new CPU's up to 2025 IIRC.

I would agree that AMD shot themselves in the foot by not supporting DDR4 for Zen 4. But it will be a well established ecosystem for future CPU's.

Of course, once the new "lake", which I heard is delayed till 2024, comes out the DDR4 vs DDR5 will be over.
And then, perhaps, the platform cost for AMD may become more affordable. I guess it will depend on what intel does with thier new platform too.

 

[bit_user]

With Raptor being the end for that platform, do you think Intel will still support DDR4 on Metoer? My guess is no.

Agreed.

Then both platforms will be roughly the same in cost.

I'm not sure about that. Toms did a breakdown of AM5 motherboard costs, a while back.

https://www.tomshardware.com/news/why-amds-ryzen-7000-and-motherboards-cost-so-damn-much

The sense I got was that it has some features oriented more towards the high-end that raise the cost floor for everyone. Maybe that's too simplistic or I got it wrong, but I'm not sure if entry-level AM5 boards will ever be as cheap as their AM4 counterparts are.

I would agree that AMD shot themselves in the foot by not supporting DDR4 for Zen 4.

They made a bet that DDR5 would've fully ramped up, by now. That was a little ambitious, but it's getting there. They've probably learned their lesson and will provide dual-standard support, when DDR6 rolls out. If, mainstream CPUs haven't already transitioned to in-package memory, by then!

Another possibility is that AMD could've made a backwards-compatible Zen4 CPU, by mating the new compute chiplet with an AM4-compatible I/O die. Then, if you wanted to stick with DDR4, you could just buy/keep an AM4 board.

 

[Roland Of Gilead]

They've probably learned their lesson and will provide dual-standard support, when DDR6 rolls out.

Yeah, I just can't see them being caught with their pants down again. Definitely a lesson learned.

The sense I got was that it has some features oriented more towards the high-end that raise the cost floor for everyone. Maybe that's too simplistic or I got it wrong, but I'm not sure if entry-level AM5 boards will ever be as cheap as their AM4 counterparts are.

That's a really interesting breakdown of the costs. The manufacturing costs are obviously up . Does Intel having their own fabs give them an advantage?

 

[saunupe1911]

In benchmark numbers, yes. In real life gaming performance, I think it is questionable. Assuming you are using a 5950X now, you should observe in games that most are the cores are at idle. Most games don't scale well with number of cores, which is exactly why Intel Alder Lake and Raptor Lake are able to keep up with the best Zen 4 chips, even with just 8 P-cores in games. Games as I understand don't utilize the E-cores based on the logic in the Windows Thread Director.

Yeah but in this chip still has a 3D core and much higher Ghz so I believe even real world gaming will have a big advantage.

 

[lmcnabney]

Eh, this argument only stands because Raptor supports DDR4. With Raptor being the end for that platform, do you think Intel will still support DDR4 on Metoer? My guess is no. Then both platforms will be roughly the same in cost. That of course doesn't include the longevity with AMD Zen 3/Zen 4. Which will support new CPU's up to 2025 IIRC.

You are forgetting the higher motherboard costs related to including PCIe5. The consumer is paying $80+ more for equivalent Intel MB and you are paying around $70 more for equivalently performing 32GB of DDR5. That means for system price parity with Intel they will need to have their CPUs cost more than $100 less.
Currently a 13600K build is going to cost less than that 7600X build AND soundly whip it in every metric. AMD needed a X3D version of the 7600 chip running at ~5.5ghz priced against the 13600K to sell to gamers.

 

[PlaneInTheSky]

If you want to make full use of a high-performance CPU, then DDR4 is not good enough anymore.

There is barely a 1% difference between DDR4 and DDR5 in gaming.

Spending $200 on DDR5 for a 1% potential gain in performance is throwing money down the drain.

Not only is AMD forcing users into high costs by having to buy new DDR5, just from an e-waste perspective AMD's decision should be lambasted.

This idea that these new $230+ non-X CPU are "value CPU" because they are $20 cheaper than their non-X counter, when users will be forced to spend an additional $150-$200 on DDR5, is baloney. They are not value CPU at all.

Value CPU are those $100 i3 with DDR4, and AMD needs an answer to them, one that doesn't involve forcing people to throw away their DDR4.

 

[Roland Of Gilead]

AMD needed a X3D version of the 7600 chip running at ~5.5ghz priced against the 13600K to sell to gamers.

Good point. Admittedly I was hoping for a 7600X3d, as there has been a lot of rumours. Still, depending on pricing, which we don't know yet, the 7800x3d will be more of a comparison.

You are forgetting the higher motherboard costs related to including PCIe5.

Don't you get a bit more PCIe 5 with AMD than Intel at the mo? As in more lanes, storage options etc? I get what you are saying though. Still, if AM4 longevity is anything to go by, after another year, AM5 will still be rocking and beyond. Overall I think it balances out, given Intel change the socket every 2 gens.

 

[Coffee Fueled Curmudgeon]

’
Yes they have… 5800x3d with b550 mobo and ddr4 memory.

The 5800X3D isn't value by any measure or metric. You might be able to plonk it on a cheap boards with budget RAM, but that doesn't magically make the £350-400 chip budget, value or remotely an answer to a 12/13400 or any decent i3 option.

 

[Coffee Fueled Curmudgeon]

Currently a 13600K build is going to cost less than that 7600X build AND soundly whip it in every metric. AMD needed a X3D version of the 7600 chip running at ~5.5ghz priced against the 13600K to sell to gamers.

Very much my situation. Even with the pre-Christmas 7600X price drop it still didn't compete with any of the 13600K options I'd part-picked. This was true across all merchants.

I was hoping to see a 7600X3D as a final reason to decide on a new build, but based on rumoured pricing, even if there was one, I'd still really only need a 7600X or 13600K for the next three to four years at which point I'd be looking to buy a new motherboard with whatever newer technologies are leveraged.

 

[nightbird321]

"Both the 7950X3D and the 7900X3D have two four-core chiplets "

Eight-core

 

[cyrusfox]

Typo on the first table on 7900X3D Cache showing

140MB (12+132)

Which of course does not add up, pretty sure it should be 140MB (12+128)

Error is repeated on the second table as well

 

[jp7189]

I can't tell if the 7950x3d is the best of both world or the worst. One CCD with lots of cache but low clocks, and one with low cache but high clocks.

I have little faith in Windows being able to schedule the right thread on the right core. Intel Thread Director on Win11 has had a while to work out the kinks and it still doesn't get it right 100% of the time, and that's pretty straight forward. Either a thread gets priority and is a "P" core, or it doesn't and it's and "E" core. AMD's cache vs clockspeed seems like a much harder decision.

 

[lmcnabney]

Don't you get a bit more PCIe 5 with AMD than Intel at the mo? As in more lanes, storage options etc? I get what you are saying though. Still, if AM4 longevity is anything to go by, after another year, AM5 will still be rocking and beyond. Overall I think it balances out, given Intel change the socket every 2 gens.

Only some Intel boards have PCIe5 while all AM5 boards have at least one. I have read somewhere that they are working on a cheap board that doesn't to go with the non-X chips, but haven't heard any details. Having DDR5 and PCIe5 drastically increases the board cost.

I was hoping to see a 7600X3D as a final reason to decide on a new build, but based on rumoured pricing, even if there was one, I'd still really only need a 7600X or 13600K for the next three to four years at which point I'd be looking to buy a new motherboard with whatever newer technologies are leveraged.

Apparently the X3D line starts at 7800X3D and goes up and that CPU is low-speed compared to the others. Exclusively targeting sales over $500 has essentially ceded 90% of the market to Intel.

 

[lmcnabney]

There is barely a 1% difference between DDR4 and DDR5 in gaming.

That chart shows crap timings. If you are putting the recommended DDR5-6000 into AM5 you shouldn't be using CL36. CL30 minimum. That is about 5% faster. Not a big bump, but bigger than that chart shows.

 

[Kamen Rider Blade]

I can't tell if the 7950x3d is the best of both world or the worst. One CCD with lots of cache but low clocks, and one with low cache but high clocks.

I have little faith in Windows being able to schedule the right thread on the right core. Intel Thread Director on Win11 has had a while to work out the kinks and it still doesn't get it right 100% of the time, and that's pretty straight forward. Either a thread gets priority and is a "P" core, or it doesn't and it's and "E" core. AMD's cache vs clockspeed seems like a much harder decision.

I guess we're going to have to rely on Process Lasso to gurantee which games goes to which cores.

 

[PlaneInTheSky]

That chart shows crap timings. If you are putting the recommended DDR5-6000 into AM5 you shouldn't be using CL36. CL30 minimum. That is about 5% faster. Not a big bump, but bigger than that chart shows.

Just stop, no one looking for a value option is going to shell out $300+ on "CL30" DDR5 for a minuscule gain, when they can just keep using their DDR4 with Intel instead.

If you are going to spend an extra $300 on hardware as a value customer, it better be something worthwhile like a better GPU that doubles your framerate. Instead of wasting $300 on top of the line DDR5 for 2 fps more.

AMD made a really dumb decision by forcing people to buy DDR5.

 

[thestryker]

I can't tell if the 7950x3d is the best of both world or the worst. One CCD with lots of cache but low clocks, and one with low cache but high clocks.

I have little faith in Windows being able to schedule the right thread on the right core. Intel Thread Director on Win11 has had a while to work out the kinks and it still doesn't get it right 100% of the time, and that's pretty straight forward. Either a thread gets priority and is a "P" core, or it doesn't and it's and "E" core. AMD's cache vs clockspeed seems like a much harder decision.

This is my concern with these X3D parts. The 7800X3D shows us how much lower these clocks are going to be and it's not insignificant. AMD appears to be bringing an even bigger problem than Intel did with P/E. In the case of Intel those P cores will always be better than the E cores, but in AMD's case that isn't so cut and dried when your difference is significant clock speed versus significant cache increase as the cores are identical.

 

[-Fran-]

This is my concern with these X3D parts. The 7800X3D shows us how much lower these clocks are going to be and it's not insignificant. AMD appears to be bringing an even bigger problem than Intel did with P/E. In the case of Intel those P cores will always be better than the E cores, but in AMD's case that isn't so cut and dried when your difference is significant clock speed versus significant cache increase as the cores are identical.

Uh... It is the complete opposite of what you're saying. If, by any chance, you move a game (or any software) to run exclusively in the E-cores, you WILL notice. It will most definitely be a big enough drop in performance for you to say "ok, something went wrong here". In the case of AMD's VCache CCD vs normal CCD, the jump in performance if the software is moved across will be less noticeable, even with games considering how ludicrous the FPS's are at the high end anyway. And if you don't have a top-of-the-line GPU, then the GPU will be the bottleneck and you won't even notice.

I'm not saying I like AMD putting VCache on only one of the CCDs, but your argument reads completely backwards to reality; specially considering the multiple tests done by running games on the E-cores exclusively.

Regards.

 

[thestryker]

Uh... It is the complete opposite of what you're saying. If, by any chance, you move a game (or any software) to run exclusively in the E-cores, you WILL notice. It will most definitely be a big enough drop in performance for you to say "ok, something went wrong here". In the case of AMD's VCache CCD vs normal CCD, the jump in performance if the software is moved across will be less noticeable, even with games considering how ludicrous the FPS's are at the high end anyway. And if you don't have a top-of-the-line GPU, then the GPU will be the bottleneck and you won't even notice.

I'm not saying I like AMD putting VCache on only one of the CCDs, but your argument reads completely backwards to reality; specially considering the multiple tests done by running games on the E-cores exclusively.

Regards.

You completely missed the point I'm making: you know the P cores are always better than the E cores and can avoid using the E-cores entirely and easily. This is very much not the case for AMD's design here you'll have to know which workloads prefer which and there's no way the scheduler is going to know which one is always better for the use case.