Question Zen 6 Speculation Thread

[511]

Today the 8-core Zen5 is better at games than the 12-core, and yet each has it's own place on the market.

It's cause the 4 core is on different CCD

 

[coercitiv]

It's cause the 4 core is on different CCD

And that's exactly what the discussion was about, single vs dual ccd:

The dual ccd 16c might be worse than single ccd 12 core in some workloads.

Extreme edge cases do not count.

Games aren't extreme edge cases as far as desktop is concerned, and I'm pretty sure some of them will run better on the mono-12c part(s).

 

[Fjodor2001]

proven wrong by 9800X3D award.

Because X3D.

 

[Tachyonism]

Games don't use 12c.

That's because 12 cores part is not what most normal gamers would buy. Plus cross-CCD is also known to increase latency and decrease Fps.

Also games go hand in hand with computer tech. 10 years ago it was 6 cores, now the meta is 8 cores. If AMD can push out 12 cores single CCD part, there will be game that ultilize 12 cores.

 

[Fjodor2001]

Zen6 launching in 2027:

"AMD Olympic Ridge “Zen 6” Ryzen Desktop CPUs Reportedly Launching In 2027"

AMD Olympic Ridge "Zen 6" Ryzen Desktop CPUs Reportedly Launching In 2027

AMD is reportedly going to launch its next-gen Ryzen Desktop CPUs based on the Zen 6 architecture, codenamed Olympic Ridge, in 2027.

wccftech.com

It is confirmed that we will not see the next generation of Ryzen processors with the Zen 6 processor architecture in 2026
The Olympic Ridge processor, based on the Zen 6 architecture, will continue to use the AM5 socket. This literally means that current AMD 800 series motherboards can continue to use it. However, whether this is actually the case remains to be seen, and we'll have to wait for further confirmation. After all, the earliest possible release date for the Olympic Ridge processor based on the Zen 6 architecture is 2027; the AMD AM5 socket has been used since the Zen 4 processor architecture in 2023.
via Benchlife

 

[gdansk]

If AMD can push out 12 cores single CCD part, there will be game that ultilize 12 cores.

Not for long. Anything embarrassingly parallel which scales with increased core count that is done on the CPU should infact be done on the GPU by any competent development team when they have the time to optimize.

The parts that stay on the CPU long term are the complicated, messy branching logic. Which unsurprisingly is difficult to thread.

 

[Fjodor2001]

Not for long. Anything embarrassingly parallel which scales with increased core count that is done on the CPU should infact be done on the GPU by any competent development team.
The parts that stay on the CPU long term are the complicated, messy branching logic. Which unsurprisingly is difficult to thread.

A lot of parallelizable code cannot be executed on a GPU.

 

[gdansk]

A lot of parallelizable code cannot be executed on a GPU.

This is true, but game code is inherently low parallelism. And always will be.
Show me what parallelizes well in modern games. It's basically doing parts of ray-tracing for the GPU and that's going to go away shortly.

Note - for example - that this average frame rate chart scales with clock rate, not core count.

 

[511]

This is true, but game code is inherently low parallelism. And always will be.
Show me what parallelizes well in modern games. It's basically doing parts of ray-tracing for the GPU and that's going to go away shortly.

Note - for example - that this average frame rate chart scales with clock rate, not core count.
View attachment 138565

Which game is this ? Also it depends on the game each game has its characteristics ....

 

[gdansk]

Which game is this ? Also it depends on the game each game has its characteristics ....

It's the average of dozens. Games have to ship, no time to wank to Cinebench scores. The few games which scale with core count (usually in a limited fashion e.g. Indiana Jones) are basically doing visual effects on the CPU and that will not survive their encounter with next generation ray accelerators and simplified APIs.

It's the same story for decades but somehow magically 12 cores are supposed to fix this. It's not a hardware problem. It's the software and it isn't changing since they can meet their frame rate targets anyway.

 

[Tachyonism]

The few games which scale with core count (usually in a limited fashion e.g. Indiana Jones) are basically doing visual effects on the CPU

Have you heard of sims games?

 

[gdansk]

Have you heard of sims games?

They're usually limited by awful, awful coding. Cities Skyline II is the poster child for unplayable trash, worse than its predecessor and multi-threaded simulation code. But sure, if we look at simulation games then there is already a game that scales well. So the 12 core part is still not special. That would be best for the 24 core part probably.

 

[Jan Olšan]

You don't understand how product segmentation works if you think that the only 6c parts they'd sell would be "rejects". You think that's true with Apple, that everything they sell with a missing GPU core is because that core was in fact defected - so that if yields later increase those cut down models will become unavailable to buy?

AMD will sell 6c parts that could have been sold as 8c for less than the 8c price because there's demand for 6c, and they will have decided on their prices for both 6c and 8c parts knowing that many 8c capable parts will be sold at 6c prices.

Exactly this.

Just remember Ryzen 3 3100 and 3300X. AMD made them because market wanted them, but they clearly weren't comfortable selling those parts, and availability was bad. Large part of such SKUs is likely using perfectly fine dies or dies that could have been 6C, because what's the odds that any defective die can be ran as 4C but not 6C?
It will be the same story with 6C Zen6 SKU. AMD may launch it so that they nominally have a price tier covered, but it may be unattractive with price set too high, or low availability, to limit the number of dies that have to be sacrificed for the product.

It's math where you have to make compromises between exploiting all the potential the dies have and what you can sell and so on. Adding the relative painful 4c or 6c SKU can gain you further volume and marketshare so it could lead to higher profit even if average selling price gets hit. Or it can be even more complex where the low core count SKU can be a requirement of some OEM so you make it to get their contracts, even if you don't intend to sell many - that OEM may mostly need it so that they can quote lower price in their "from XXX USD" advertising. Or those SKUs sell in reality, but not in great number on the biggest markets.

Note how Ryzen 3 5300G exists but is fairly hard to find.

 

[adroc_thurston]

Because X3D.

Whats the point?

That's because 12 cores part is not what most normal gamers would buy. Plus cross-CCD is also known to increase latency and decrease Fps.

No, it's because game code isn't very parallel at all.
Plus the consoles are 8c.

 

[LightningZ71]

Selling a product in that slot ( 6 cores) also allows them to push their parametric yield targets higher. If they want to, they can tighten the FMax and voltage targets they are binning for on desktop to command a higher price for their more premium bins while still allowing for a "recovery" sku. It's likely, knowing TSMC, that even with tighter bin targets, they still have to cripple good CCDs to hit volume targets, but, it can allow the rest of the stack to be priced more aggressively.

 

[Fjodor2001]

Whats the point?

No 6C X3D exists.

 

[gdansk]

No 6C X3D exists.

https://www.amd.com/en/products/processors/desktops/ryzen/7000-series/amd-ryzen-5-7600x3d.html