Question Raptor Lake - Official Thread

[Hulk]

Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.

 

[Hulk]

This is the plot for my 12700K. I run it at 4.6GHz, after which it's efficiency begins to decline rapidly. This is info from the BIOS, I could probably under volt but I am too lazy.

 

[Carfax83]

The 7700x already matches if not slightly outperforms a 12900K in gaming, so yes, I think a Zen4-3d CPU is the best bet to be the gaming king upon release.

The 7700x has a clock speed advantage against the 12900K that is going to be erased against Raptor Lake, plus the latter will have enhanced cache and memory performance. I have no doubts that Raptor Lake is going to beat Zen 4 in gaming, but I do agree that Zen 4 3D is a different beast.

I'm not sure that Zen 4 3D will benefit from the increased cache the way that Zen 3 did though, as Zen 4 doesn't seem to be bandwidth starved unlike Zen 3. Not saying it won't beat Raptor Lake, but I doubt we will see the huge gains that Zen 3 3D had.

 

[IntelUser2000]

Raptorlake fixes the uncore clocking much lower when E cores are disabled. They went from a 4.1GHz uncore to 5GHz. They said itself is responsible for a gain of 5% in gaming.

Also from leakers about inter-core latency, Raptorlake has a much better latency distribution among the cores. Basically only the boundary line between P and E are affected, whereas previously the P's suffered too.

Both are going to improve the scenarios where the hybrid implementation doesn't fare so hot. I can't wait to see how much better it becomes in the following years!

There is almost no IPC uplift with Raptor Cove's P cores, at least not in the leaks we had thus far. It's just clock and process improvements, plus faster E cores.

Ironically despite the expectation, the E cores have a bigger gain because of the uncore improvements benefitting them more. For example they said the prefetching improvements along with the increased L2 cache size allow the E cores to have 2-3% perf/clock improvement and in some cases up to 16%!

This also shows the promise of the hybrid because E cores should improve at a faster rate than P cores do. Maybe in 2-3 generations most of the doubts about hybrid will disappear because the E cores will be that much closer to the P cores. We need to have a long term perspective, because Alderlake is their first try.

Next major IPC jump for Intel will be Arrow Lake. Lion Cove will indeed be the king of the jungle even more likely is if Intel 3 is potent.

20A not 3.

 

[poke01]

20A not 3.

ahh yes.

Those Arrow Lake mobile (6+8) configs are gonna be super efficient with 20A and it will be powerful too.

 

[poke01]

the E cores will be that much closer to the P cores. We need to have a long term perspective, because Alderlake is their first try.

If that's the case Intel will dominate laptop and desktop sales wise, rep wise and AMD will have to go hybrid.

 

[trivik12]

If that's the case Intel will dominate laptop and desktop sales wise, rep wise and AMD will have to go hybrid.

Isn't that already in works. Zen 5 is rumored to be a hybrid cpu with zen 4c cores as well.

 

[Wolverine2349]

If that's the case Intel will dominate laptop and desktop sales wise, rep wise and AMD will have to go hybrid.

I hate the hybrid arch. It belongs in mobile space not power desktops. Nevermind all the scheduling and compatibility crap as we have been in SMP world for 30 years in X86 space.

More than 8 P cores on the same CCD/ring would be great.

Though it would be nice if Intel had an 8 P core only option with no e-cores that is well binned and can overclock so high. Cause even highly threaded games do not really benefit at all from more than 8 good cores and seems unlikely to change anytime soon as I have heard coding for more and more threads is beyond impossible for things like games.

 

[nicalandia]

This also shows the promise of the hybrid because E cores should improve at a faster rate than P cores do. Maybe in 2-3 generations most of the doubts about hybrid will disappear because the E cores will be that much closer to the P cores.

I just don't see that happening, the e cores due to their design will Always be constrained by their design(which is to max out performance/Area) they have limited FP performance, their L2 Cache is far away from the L3 and Core 1(on the cluster) has to do a round trip to comunicate with Core 2.

The e core design for MTL and ARL have the same foundation as the gracemont cores, they will tweak the design as physically possible to extract every bit of IPC, but I will be surprice if by Arrow Lake they have reach sunny cove performance

 

[deasd]

Isn't that already in works. Zen 5 is rumored to be a hybrid cpu with zen 4c cores as well.

'Zen5 hybrid' is just a BS rumor from MLID in 2021:

Upcoming AMD Ryzen with 8 Zen 5 cores and 16 Zen 4D cores to potentially thwart Intel's hybrid advances, Zen 4D to offer massive multi-core gains at Zen 3 IPC levels

AMD could be looking to introduce a Zen 4D contemporary to Zen 4 Raphael that would focus more on multi-core performance even if it means reduced single-core gains. Zen 4D is slated to launch with Bergamo EPYC parts but are likely to be available on the desktop or even consoles such as the PS5...

www.notebookcheck.net

but im not going to talk about Zen5 here....

 

[ondma]

Just read a test of the 4090 on Techspot. Seems like it puts an much larger cpu stress on gaming than current cards. They tested with a 5800x 3D and a lot of games were cpu limited even at 1440. Will be interesting to see if finally more cores will lead to better gaming performance, e.g. 7700x vs 7950x.

Edit: I thought this was Zen 4 thread (duh). My real question is if Raptor Lake with only 8 P cores is going to be sufficient to feed the newer GPUs like the 4090, whereas AMD has up to 16 "real" cores. Not sure if the E cores can somehow be set to take over some of the increased cpu load allowed by the generational GPU leap.

 

[Wolverine2349]

Just read a test of the 4090 on Techspot. Seems like it puts an much larger cpu stress on gaming than current cards. They tested with a 5800x 3D and a lot of games were cpu limited even at 1440. Will be interesting to see if finally more cores will lead to better gaming performance, e.g. 7700x vs 7950x.

Don't games have to be coded to take advantage of more cores though?? I mean games while they are more threaded than in past aren't they still limited to how many threads they can scale to and fewer faster cores are better than more and more cores??

And yeah more than 8 cores for gaming?? Though with the 7900X and 7950X, will there be a big penalty because of the fact there are 2 CCDs threads have to communicate with and crossing them incurs a big latency hit which is bad for games and could drop 1% lows a lot at those times?? And on Intel's case if a thread gets caught ion an e-core ouch as well as the e-cores are bad for game threads. On AMD's side the other cores are still strong but the cross latency penalty. We need more than 8 P cores on a single ring/CCD if games are really going to benefit from more than 8 cores 16 threads. If only Intel would make a 10 P core Raptor Lake or Alder Lake

 

[Mopetar]

Do you think Zen 4 with 3D cache will handily beat a manually well tuned Raptor Lake 13900K for gaming or trade blows or be very close??

Yes. We saw how well the 5800X3D did in comparison, and that was with lower clocks than the regular Zen 3 CPUs. Given how soon after the launch of Zen 4 we're getting the 3D version it's pretty safe to assume that AMD has things well in hand.

I expect RL to be more competitive in terms of productivity with respect to AMD this time around, but even Intel's own figures suggest they'll lose to v-cache Zen 4 more often than not.

 

[nicalandia]

Even Intel's own figures suggest they'll lose to v-cache Zen 4 more often than not.

True, here is the 5800X3D vs 12900K and 13900K


That is a 4.5 Ghz Zen3 CPU with V-Cache. The 7800X3D will Likely boost to 5.5 That is a 22% Speed boost plus the IPC boost will likely be enough to keep the gaming crown until late 2024

 

[Wolverine2349]

Yes. We saw how well the 5800X3D did in comparison, and that was with lower clocks than the regular Zen 3 CPUs. Given how soon after the launch of Zen 4 we're getting the 3D version it's pretty safe to assume that AMD has things well in hand.

I expect RL to be more competitive in terms of productivity with respect to AMD this time around, but even Intel's own figures suggest they'll lose to v-cache Zen 4 more often than not.

Do you think it will be a bloodbath for Zen 4 3D vs Raptor Lake in gaming or a close win for AMD more often than not assuming paired with GeForce RTX 4090 at 1440p.

 

[inf64]

Do you think it will be a bloodbath for Zen 4 3D vs Raptor Lake in gaming or a close win for AMD more often than not assuming paired with GeForce RTX 4090 at 1440p.

I'm afraid nobody can answer that question right now. We still have no CPU scaling reviews for RTX 4090, that will give us some clues at least.

 

[Mopetar]

Do you think it will be a bloodbath for Zen 4 3D vs Raptor Lake in gaming or a close win for AMD more often than not assuming paired with GeForce RTX 4090 at 1440p.

Outside of a few titles where the extra cache doesn't do anything for performance, a Zen 4D (I'm too lazy to write Zen 4 3D repeatedly, so I'm just going to call it that despite its lack of temporal die stacking) is going to slaughter anything else. We already know that the 4090 is CPU bound in several titles at 1440p and that's where the v-cache will shine most.

If AMD figured out how to alleviate the voltage limitations that restricted the 5800X3D's clock speed then it's even worse. There will probably still be a few titles where Intel does better, but I don't expect the wins to be a big as where Zen 4D does better.

There's also the question of cost. Zen 3D offered some of the best results for less than a top end AMD or Intel chip. AMD left an obvious price hole in their lineup for Zen 4D and it's likely going to be cheaper than whatever souped up i9 Intel releases that offers their best performance. Never mind that AM5 had an upgrade path so you won't need a new board if you want to get a Zen 5D at some point.

 

[Wolverine2349]

Outside of a few titles where the extra cache doesn't do anything for performance, a Zen 4D (I'm too lazy to write Zen 4 3D repeatedly, so I'm just going to call it that despite its lack of temporal die stacking) is going to slaughter anything else. We already know that the 4090 is CPU bound in several titles at 1440p and that's where the v-cache will shine most.

If AMD figured out how to alleviate the voltage limitations that restricted the 5800X3D's clock speed then it's even worse. There will probably still be a few titles where Intel does better, but I don't expect the wins to be a big as where Zen 4D does better.

There's also the question of cost. Zen 3D offered some of the best results for less than a top end AMD or Intel chip. AMD left an obvious price hole in their lineup for Zen 4D and it's likely going to be cheaper than whatever souped up i9 Intel releases that offers their best performance. Never mind that AM5 had an upgrade path so you won't need a new board if you want to get a Zen 5D at some point.

Yeah if true it makes me wonder why oh why is AMD not releasing Zen 4 3D cache chip right now or within a couple of weeks given the RTX 4090 is hitting retail tomorrow.

I mean it almost seems like any CPU will bottleneck an RTX 4090 even at 1440P or possibly even 4K except the Zen 4 3d Cache versions. Why is NVIDIA even releasing there super GPU if every CPU currently out is going to bottleneck it and Intel and AMD cannot come up with anything better at time of launch to not bottleneck that awesome RTX 4090 at 1440P.

 

[Carfax83]

Yes. We saw how well the 5800X3D did in comparison, and that was with lower clocks than the regular Zen 3 CPUs. Given how soon after the launch of Zen 4 we're getting the 3D version it's pretty safe to assume that AMD has things well in hand.

I think it's erroneous to assume that Zen 4 3d is going to scale with V-cache the way that Zen 3D did. Zen 4 has more cache as a baseline, plus it uses DDR5 which provides much more memory level parallelism and bandwidth.

That is going to dampen the impact of the V-cache.

 

[Kryohi]

Yeah if true it makes me wonder why oh why is AMD not releasing Zen 4 3D cache chip right now or within a couple of weeks given the RTX 4090 is hitting retail tomorrow.

Because they can't. IIRC TSMC said that production of 6N stacked on 5N chips would start Q4 2022, so it's likely starting only now.

 

[Wolverine2349]

I think it's erroneous to assume that Zen 4 3d is going to scale with V-cache the way that Zen 3D did. Zen 4 has more cache as a baseline, plus it uses DDR5 which provides much more memory level parallelism and bandwidth.

That is going to dampen the impact of the V-cache.

Moore's Law is Dead seems to disagree. How would AMD do it if the problem with Zen 3 that the 3D cache boosted was memory bandwidth and that was solved by extra cache?? Where as Zen 4 has much better DDR5 bandwidth and RAM controller so you think it will not scale nearly as well as Zen 3 3D cache did.

 

[Wolverine2349]

Because they can't. IIRC TSMC said that production of 6N stacked on 5N chips would start Q4 2022, so it's likely starting only now.

Yeah kind of sucks cause if Zen 4 3d Cache is as good as we think, I oh so want t o put a system together with it right as 4090 hits and not wait lol.

If only Intel also had a 3D cache version of Raptor Lake.

 

[Carfax83]

Moore's Law is Dead seems to disagree. How would AMD do it if the problem with Zen 3 that the 3D cache boosted was memory bandwidth and that was solved by extra cache?? Where as Zen 4 has much better DDR5 bandwidth and RAM controller so you think it will not scale nearly as well as Zen 3 3D cache did.

MLID is hardly a reliable source, especially when it comes to AMD related leaks. That said, the purpose of cache is to mitigate the latency and bandwidth penalties associated with accessing main memory. CPUs with slower main memory should theoretically benefit more from increased cache than a CPU with faster memory.

I know that Zen 3 3D didn't benefit from memory overclocking as much as standard Zen 3 did, because it had a much bigger L3 cache.

Similarly, Raptor Lake's ability to utilize much higher DDR5 memory speeds than Zen 4 3D should help to compensate somewhat for the smaller L3 cache.

DDR5 memory kits with 7ghz+ speeds are already going to be available later this month.

 

[biostud]

I think it's erroneous to assume that Zen 4 3d is going to scale with V-cache the way that Zen 3D did. Zen 4 has more cache as a baseline, plus it uses DDR5 which provides much more memory level parallelism and bandwidth.

That is going to dampen the impact of the V-cache.

the vcache on 5800X3D gave between 0-40% increased fps over the regular 5800X (with lower clocks) and varied a lot depending on the game. So maybe we will see something like 0-30% compared to the 7700X as I think it will be running at least the same clock as the 7700X. But the average of many games might be lower than the ~15% the 5800X3D was promoted with. It all depends on which games you play.

 

[Shivansps]

I know no one is interested in it for the IGP, but...
DDR5-5200

Stock

2100mhz (the max allowed in the bios)

Its not that bad of a result, definately better than what AMD has on Ryzen 7000X. DDR4-3600 was just 1fps less. It is around 3200G stock performance i belive for Shadow, and W3 is around 5600G perf.
I dont know whats up with W3 1C usage

 

[maddie]

MLID is hardly a reliable source, especially when it comes to AMD related leaks. That said, the purpose of cache is to mitigate the latency and bandwidth penalties associated with accessing main memory. CPUs with slower main memory should theoretically benefit more from increased cache than a CPU with faster memory.

I know that Zen 3 3D didn't benefit from memory overclocking as much as standard Zen 3 did, because it had a much bigger L3 cache.

Similarly, Raptor Lake's ability to utilize much higher DDR5 memory speeds than Zen 4 3D should help to compensate somewhat for the smaller L3 cache.

DDR5 memory kits with 7ghz+ speeds are already going to be available later this month.

Is it memory speed or latency that matters when applied to V-cache improvements in performance? Faster memory generally do not improve latency, as faster has higher C#, leading to little to none absolute improvements in latency.