Question Zen 6 Speculation Thread

[511]

We are all waiting with intense curiosity to see how much latency 144MB of bLLC adds for Nova Lake as well...

this won't add much unlike the mesh in EMR/SPR/GNR there are more cores and more stops in NVL the stops are likely the same but the cache per stop is increased there will be increase in latency but not this much

 

[LightningZ71]

I thought NVL was restructured, with pairs of P cores sharing an L2 and L3 ring interface? There should only be 8 compute stops on the ring and a 9th for the connection to the I/O and Main Memory controller. No telling how the shared L2 will act though.

 

[igor_kavinski]

I would love free 1950X. It's a historically significant part.

For a while, it was available for $25 from my local Amazon. I was like, why bother spending that much for a souvenir? I just looked now and the price has risen to $163. Must be the pricing idiots thinking every Ryzen no matter how old, accelerates AI.

 

[511]

For a while, it was available for $25 from my local Amazon. I was like, why bother spending that much for a souvenir? I just looked now and the price has risen to $163. Must be the pricing idiots thinking every Ryzen no matter how old, accelerates AI.

You are never good with price tbh Arrow to the Knee xDd

 

[511]

I thought NVL was restructured, with pairs of P cores sharing an L2 and L3 ring interface? There should only be 8 compute stops on the ring and a 9th for the connection to the I/O and Main Memory controller. No telling how the shared L2 will act though.

Yes that's how it is 4MB Shared L2 between 2 Cores

 

[igor_kavinski]

You are never good with price tbh Arrow to the Knee xDd

Yeah if anyone wants stock market manipulation, they just need to give me money to buy the product for the company they wish to tank the stock price of

Intel would be in a much better position if I hadn't gotten a 12700K and 245KF, along with almost 10 other Intel CPUs.

 

[Win2012R2]

Yes that's how it is 4MB Shared L2 between 2 Cores

I wonder how would that work in games that likely to utilise 6-8 cores, shifting some to another "chiplet" would bring same issues as with AMD, so having 2 threads competing for same L2 cache, this must be introducing extra latency since it is shared, looks like a problem area to me. Single thread meaningless benchmarks should be ok tho.

 

[Markfw]

I would love free 1950X. It's a historically significant part.

I have one, but for less than $1000, the MB+CPU + NVME are not worth less. and not saleable at that price, but thats what its worth to me.

 

[LightningZ71]

Yeah if anyone wants stock market manipulation, they just need to give me money to buy the product for the company they wish to tank the stock price of

Intel would be in a much better position if I hadn't gotten a 12700K and 245KF, along with almost 10 other Intel CPUs.

I feel you there. I refuse to ever buy into a broad market ETF like VOO or SCHD, lest I cause the next great depression by absolutely tanking the whole US stock market...

 

[eek2121]

I would love free 1950X. It's a historically significant part.

That is why I'm not getting rid of mine. Although I need to finish rebuilding it so I can boot it up again. I had to salvage it for RAM and the PSU.

For a while, it was available for $25 from my local Amazon. I was like, why bother spending that much for a souvenir? I just looked now and the price has risen to $163. Must be the pricing idiots thinking every Ryzen no matter how old, accelerates AI.

I am insulted. I paid $999 for it. 🤣

 

[Timmah!]

That is why I'm not getting rid of mine. Although I need to finish rebuilding it so I can boot it up again. I had to salvage it for RAM and the PSU.

I am insulted. I paid $999 for it. 🤣

I paid 1400 for Intel 7940x. But its still powering my backup machine. Though i use that one rarely.
Not sure i would be willing to pay that much for CPU again. It would really have to that kind of jump like those early TRs/Skylake-X i9s were over previous 6C/8C cpus.

 

[OneEng2]

They give the 9900 series a dual X3D and yet they don't even bother on threadripper.

Amd really doesn't want us gamers to have PCI-E lanes.

I would swap platforms so fast if the new dual X3D 9900's had 64PCI-E lanes.

I really don't need the quad chanel DDR5, nor do i need the more cores then i can fit on my windows task manager screen.
I just need those PCI-E lanes and a X3D instruction set on my cache, and i would be that happy little fat kid that had that unlimited supply of cotton candy.

What are you doing with your rig?

I don't think I will be able to resist the 12 core Zen 6 at 7 GHz with 2 layers of V-Cache. 😉

It is an interesting thought. I doubt very much Zen 6 will be released at anywhere near 7Ghz; however, I can absolutely see a single CCD dual layer X3D (would that make it "X4D"?). That could be a massive trump card in gaming.

 

[adroc_thurston]

That could be a massive trump card in gaming.

It won't do anything.

 

[OneEng2]

It won't do anything.

Not sure why you so delight in responding with no explanation for your statements what-so-ever.

How should I respond? "Yes it will"?

 

[igor_kavinski]

Not sure why you so delight in responding with no explanation for your statements what-so-ever.

Shouldn't you be used to his modus operandi by now?

He's not going to change. Probably filthy rich. Owes no one any explanation. It's a privilege if he shares any info

 

[RnR_au]

Not sure why you so delight in responding with no explanation for your statements what-so-ever.

I think @adroc_thurston is referring to AMD's explanations back when they released the first iteration of X3D. AMD kinda said that gaming gains beyond the L3 cache amounts provided by X3D were iffy and very much game dependent. Its not like L3 cache is an infinite gaming boost that can be triggered by a doubling of cache whenever AMD needs it.

 

[Kaffeekenan]

Not sure why you so delight in responding with no explanation for your statements what-so-ever.

How should I respond? "Yes it will"?

Could not agree more.

 

[MS_AT]

Not sure why you so delight in responding with no explanation for your statements what-so-ever.

To be fair, in which way your post is different? You say it could be a massive trump card in gaming. But you did not bother to explain why. He also did not bother to explain why he disagrees with you, but content wise, how is that different? In both cases there are no arguments to support the case of the author.

 

[Chicken76]

To be fair, in which way your post is different? You say it could be a massive trump card in gaming. But you did not bother to explain why. He also did not bother to explain why he disagrees with you, but content wise, how is that different? In both cases there are no arguments to support the case of the author.

In that particular case he didn't need to.
He was talking about 2-hi X3D cache and it's well known that increasing the L3 cache size has significant performance implications in games, even if the latencies are increased a little.
Zen2 -> Zen3 : double the L3 available for the critical thread
Zen3 -> Zen3X3D: triple the L3 cache
Each increase had a positive impact on gaming performance.

 

[MS_AT]

In that particular case he didn't need to.
He was talking about 2-hi X3D cache and it's well known that increasing the L3 cache size has significant performance implications in games, even if the latencies are increased a little.
Zen2 -> Zen3 : double the L3 available for the critical thread
Zen3 -> Zen3X3D: triple the L3 cache
Each increase had a positive impact on gaming performance.

And what makes you believe that further increase of L3 will not hit diminishing returns? After all not all games benefit from x3D. And therefore not all of them have to benefit from even more L3 cache.

 

[LightningZ71]

Not sure why you so delight in responding with no explanation for your statements what-so-ever.

How should I respond? "Yes it will"?

The current X3D cache AMD products already have extraordinarily high L3 cache hit rates. Even doubling the size barely increases that hit rate by a few percent, meaning that it's effect on game performance will be vanishingly small. Also consider, each extra layer will add a cycle or two extra latency, reducing performance as well.

Extra cache has massively diminishing returns.

 

[dr1337]

And what makes you believe that further increase of L3 will not hit diminishing returns?

the fact that ram speed and latency still affects gaming performance is a big one


Look at this graph from chips and cheese and I think it can become more clear. There is literally 4x more latency from DRAM compared to L3. If AMD can move that blue line over another segment, cache hit-rates will still continue to go up. Maybe games will finally hit diminishing returns (I doubt it), but overall efficiency would still be greater for any task that would have gone to memory.

 

[basix]

According to the rumors, Zen 6 should reduce memory latency. So +50% LLC, decently higher IPC and core clocks and on top lower memory latency are in the ring. Looks good regarding to gaming.

If that all happens:
To be seen.

 

[OneEng2]

Shouldn't you be used to his modus operandi by now?

[chuckles] Yes, but I still feel the need (not sure exactly why) once in a while to point out how useless such commentary is. I know I shouldn't scratch it, but I can't help myself .

I think @adroc_thurston is referring to AMD's explanations back when they released the first iteration of X3D. AMD kinda said that gaming gains beyond the L3 cache amounts provided by X3D were iffy and very much game dependent. Its not like L3 cache is an infinite gaming boost that can be triggered by a doubling of cache whenever AMD needs it.

... and this is a decent line of argument IMO.

To be fair, in which way your post is different? You say it could be a massive trump card in gaming. But you did not bother to explain why. He also did not bother to explain why he disagrees with you, but content wise, how is that different? In both cases there are no arguments to support the case of the author.

Fair point. My assumption (as well as others) is that lowering latency improves gaming performance AND that doubling the L3 would further decrease the latency overall for games.

I actually really like the discussion both in defense of why doubling will help, and why it may not. Both are useful. "No it wont" .... less useful .

In that particular case he didn't need to.
He was talking about 2-hi X3D cache and it's well known that increasing the L3 cache size has significant performance implications in games, even if the latencies are increased a little.
Zen2 -> Zen3 : double the L3 available for the critical thread
Zen3 -> Zen3X3D: triple the L3 cache
Each increase had a positive impact on gaming performance.

... and this was my (admittedly) unspoken assumption.

My opinion on the matter?

I think it will help a great deal on a small number of games, it will help some on most games, and it will have little effect on a small number of games.

 

[CouncilorIrissa]

the fact that ram speed and latency still affects gaming performance is a big one

View attachment 128731
Look at this graph from chips and cheese and I think it can become more clear. There is literally 4x more latency from DRAM compared to L3. If AMD can move that blue line over another segment, cache hit-rates will still continue to go up. Maybe games will finally hit diminishing returns (I doubt it), but overall efficiency would still be greater for any task that would have gone to memory.

This chart can't be used as proof that games will see an uplift. It all depends on the size of the working set of the particular game. Depending on the exact figure, additional hitrate may not actually compensate for additional latency of a larger L3.