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.

 

[Timmah!]

I've heard it both ways as well. It's difficult for laymen like us to extract truth from these kinds of excerpts. When I did my research, it seemed that the CPU does the initialization and maintenance for the BVH structures, but building the BVH structures seemed to me like it was done on the GPU for the most part.

Of course this will vary a lot. I know that World of Tanks uses Intel Embree to accelerate BVH building and traversal.

Yes, thats the gist i took from what i read or was told as well, so i am willing to give you benefit of doubt you might be right and CPU does indeed plays some role in the process, that may be reflected on the performance.
That said, even if 13900k really stomped Raptor Lake in this regard, since its truly just Insomniac games that seem to show this kind of difference (and it still might be result of slow RAM used that gimps Zen significantly moreso than Intel), i dont think we should really do any broad conclusions yet. The CPUs seem to be otherwise very competitive everywhere else.

 

[Elfear]

I have to see the rest of the graph and see what the game is and what engine it's running on. But even without seeing it, the answer is likely yes, it's GPU bottlenecked, otherwise there would be a performance increase. Faster memory increases performance when the CPU itself is the bottleneck.

The pic was from the 12-game average from the same review at TPU. Faster memory only increases CPU performance if the memory is the bottleneck.

 

[nicalandia]

13900KS vs 7950X Sisoft Sandra Arithmetic Native

Details for Result ID 13th Gen Intel(R) Core(TM) i9-13900KS (8C + 16c 32T 5.58千兆赫（GHz） + 4.28千兆赫（GHz）, 4.6千兆赫（GHz）/5千兆赫（GHz） IMC, 8x 2兆字节（MB） + 4x 4兆字节（MB） L2, 36兆字节（MB） L3)

ranker.sisoftware.co.uk

 

[Carfax83]

Yes, thats the gist i took from what i read or was told as well, so i am willing to give you benefit of doubt you might be right and CPU does indeed plays some role in the process, that may be reflected on the performance.
That said, even if 13900k really stomped Raptor Lake in this regard, since its truly just Insomniac games that seem to show this kind of difference (and it still might be result of slow RAM used that gimps Zen significantly moreso than Intel), i dont think we should really do any broad conclusions yet. The CPUs seem to be otherwise very competitive everywhere else.

It's not just the Spider-Man games. Callisto Protocol and Witcher 3 Next gen also shows that behavior, and so does Cyberpunk 2077. In the PCGH.de 7600/7700/7900 review, the 13900K had a 40% lead over the 7950x in the Witcher 3 Next gen, a 37% lead over the 7950x in Callisto Protocol, and a 26% lead over the 7950x in CBP 2077.

Ryzen 7000 Non-X: Benchmarks, neue Grundlagen (pcgameshardware.de)

These games were all tested at 720p with RT enabled if the option was present. As for slow RAM gimping the Zen CPUs, yeah it definitely has an impact I agree, but it can't explain a 40% gap. I think it's a combination of different factors. Memory speed is one, memory bandwidth is a big one, and also cache bandwidth. The reason why I say the latter two are factors is because Raptor Lake also has big performance uplifts over Alder Lake as well, and Raptor Lake's memory controller and cache bandwidth are upgraded over Alder Lake. Raptor Lake's cache bandwidth is much larger core for core than Zen 4. RPL's larger ROB (512 vs 320) may also be a factor, but that's harder to determine.

 

[Carfax83]

The pic was from the 12-game average from the same review at TPU.

Then there's definitely some GPU bottlenecking going on then. The vast majority of games are GPU bottlenecked, so to provoke a CPU bottleneck certain conditions usually have to be met, ie low resolution, fastest GPU available, settings that increase CPU dependency like draw distance, ray tracing, crowd density etcetera.. Some games are also CPU bottlenecked due to bad CPU optimization like Callisto Protocol and the Witcher 3 Next gen, which only really use two cores.

It always amazes me how ignorant/incompetent many of these hardware reviewers are. Many of them will test memory kits at GPU bound settings and then conclude that there's no difference at all in performance.

 

[nicalandia]

It always amazes me how ignorant/incompetent many of these hardware reviewers are..

You of all people should not talk about Ignorance. The amount of Copium coming from a Random internet warrior like you is off the chart.

 

[Carfax83]

You of all people should not talk about Ignorance. The amount of Copium coming from a Random internet warrior like you is off the chart.

You're trying to make this about me. Sounds like you're the one with the "copium."

Highlighting benchmarks that are unfavorable to your favorite team is construed as an attack. That's how you think and operate.

Too bad for you those benchmarks aren't going to go away.

 

[Kaluan]

You're trying to make this about me. Sounds like you're the one with the "copium."

Highlighting benchmarks that are unfavorable to your favorite team is construed as an attack. That's how you think and operate.

Too bad for you those benchmarks aren't going to go away.

Pot, meet kettle...

 

[Timmah!]

It's not just the Spider-Man games. Callisto Protocol and Witcher 3 Next gen also shows that behavior, and so does Cyberpunk 2077. In the PCGH.de 7600/7700/7900 review, the 13900K had a 40% lead over the 7950x in the Witcher 3 Next gen, a 37% lead over the 7950x in Callisto Protocol, and a 26% lead over the 7950x in CBP 2077.

Ryzen 7000 Non-X: Benchmarks, neue Grundlagen (pcgameshardware.de)

These games were all tested at 720p with RT enabled if the option was present. As for slow RAM gimping the Zen CPUs, yeah it definitely has an impact I agree, but it can't explain a 40% gap. I think it's a combination of different factors. Memory speed is one, memory bandwidth is a big one, and also cache bandwidth. The reason why I say the latter two are factors is because Raptor Lake also has big performance uplifts over Alder Lake as well, and Raptor Lake's memory controller and cache bandwidth are upgraded over Alder Lake. Raptor Lake's cache bandwidth is much larger core for core than Zen 4. RPL's larger ROB (512 vs 320) may also be a factor, but that's harder to determine.

In Callisto Protocol 13900k has almost 20 percent jump above 13700k. I dont know about their frequency differences, but if i had to guess, i think the 13900k dominance can be explained here the best by number of cores. It is after all the only 24 core CPU in the test. Perhaps whatever is being done there concerning RT, if your theory is true, is flogged to E-cores and the sheer number of them is more helpful than 8 bigger cores on AMD dual chiplet CPU. 13700k is then only 15 percent above 7950x, its still more, but more in line with other results.

I dont know how much does RAM gimp the Zen4, but i think its undeniable that Zen CPUs gain more from faster RAM than Intel, given its direct influence on Infinite Fabric...thats still the thing, right? I recall it was back in Zen 1 days, but i had Intel up until now, so i did not pay that much attention.

Additionally, 13900k is tested with 5600 RAM, while Zen with 5200 - in situation, where Zen suffers more. And finally, its on Windows 11, i dont know what is the state of things now, but was it not so long before that AMD stuff did better on WIn10 than 11?

If both were tested at same higher RAM frequency, and AMD on Win 10, perhaps that 37 percent difference would fall down to like 10 percent. Which is still significant, but more or less expected, given the fact Raptor Lake is still faster core than Zen4 and its monolithic CPU compared to likes of 7950x, which is always going to suit games better.

EDIT: Wasnt 7950x or any of the AMDs Zen4 CPUs tested elsewhere in any of these games, with 5600 RAM and 4090 in 720p and max settings, so we can make a direct comparison? Or even better, did not someone test any of these games with both AMD and Intel at same RAM speed above 6000 MHz?
EDIT: Nvm, seems it wasnt.

 

[Carfax83]

In Callisto Protocol 13900k has almost 20 percent jump above 13700k. I dont know about their frequency differences, but if i had to guess, i think the 13900k dominance can be explained here the best by number of cores. It is after all the only 24 core CPU in the test. Perhaps whatever is being done there concerning RT, if your theory is true, is flogged to E-cores and the sheer number of them is more helpful than 8 bigger cores on AMD dual chiplet CPU. 13700k is then only 15 percent above 7950x, its still more, but more in line with other results.

Callisto Protocol is poorly optimized for the CPU. Lots of issues, and turning on RT makes the problem worse due to the higher CPU overhead that isn't balanced with proper multithreading. There's lots of benchmarks on YouTube demonstrating this if you care to take a look.

The gap between the 13900K and the 13700K is more likely due to the increased L3 cache combined with the clock speed. 13900K will boost to 5.8ghz under certain conditions, and from what I understand, this game is predominantly single threaded so it's likely the CPU was boosting to 5.8ghz.

I dont know how much does RAM gimp the Zen4, but i think its undeniable that Zen CPUs gain more from faster RAM than Intel, given its direct influence on Infinite Fabric...thats still the thing, right? I recall it was back in Zen 1 days, but i had Intel up until now, so i did not pay that much attention.

Up to a certain point, it's likely that Zen 4 gains more from faster RAM than Intel. But that cap is limited due to the necessity of needing a 1:1 ratio; usually around 6000/6200 MT/s unless you're a master overclocker like @Det0x with excellent cooling.

Intel CPUs however will just continue to scale with memory frequency if the workload is CPU limited.

Additionally, 13900k is tested with 5600 RAM, while Zen with 5200 - in situation, where Zen suffers more. And finally, its on Windows 11, i dont know what is the state of things now, but was it not so long before that AMD stuff did better on WIn10 than 11?

5200 vs 5600 MT/s can't explain such big performance gaps though (7% difference). As for Windows 11, from what I understand, those issues affected just one game (Metro Exodus) and AMD refuted the claims that Windows 11 had performance issues on Zen 4:

AMD Refutes Reports of Ryzen 7000 Performance Inconsistencies Between Windows 11 and Windows 10 | TechPowerUp

If both were tested at same higher RAM frequency, and AMD on Win 10, perhaps that 37 percent difference would fall down to like 10 percent. Which is still significant, but more or less expected, given the fact Raptor Lake is still faster core than Zen4 and its monolithic CPU compared to likes of 7950x, which is always going to suit games better.

As I said above, there isn't much of a frequency difference between 5600 and 5200, just 7.7%. That alone can't explain these gaps. That's why I said memory bandwidth and cache bandwidth play bigger roles.

EDIT: Wasnt 7950x or any of the AMDs Zen4 CPUs tested elsewhere in any of these games, with 5600 RAM and 4090 in 720p and max settings, so we can make a direct comparison? Or even better, did not someone test any of these games with both AMD and Intel at same RAM speed above 6000 MHz?

Unfortunately no. I honestly do wish that PCGH.de and Computerbase.de would test with faster memory, but I also understand why they don't and I don't fault them for it.

 

[Hitman928]

In Callisto Protocol 13900k has almost 20 percent jump above 13700k. I dont know about their frequency differences, but if i had to guess, i think the 13900k dominance can be explained here the best by number of cores. It is after all the only 24 core CPU in the test. Perhaps whatever is being done there concerning RT, if your theory is true, is flogged to E-cores and the sheer number of them is more helpful than 8 bigger cores on AMD dual chiplet CPU. 13700k is then only 15 percent above 7950x, its still more, but more in line with other results.

I dont know how much does RAM gimp the Zen4, but i think its undeniable that Zen CPUs gain more from faster RAM than Intel, given its direct influence on Infinite Fabric...thats still the thing, right? I recall it was back in Zen 1 days, but i had Intel up until now, so i did not pay that much attention.

Additionally, 13900k is tested with 5600 RAM, while Zen with 5200 - in situation, where Zen suffers more. And finally, its on Windows 11, i dont know what is the state of things now, but was it not so long before that AMD stuff did better on WIn10 than 11?

If both were tested at same higher RAM frequency, and AMD on Win 10, perhaps that 37 percent difference would fall down to like 10 percent. Which is still significant, but more or less expected, given the fact Raptor Lake is still faster core than Zen4 and its monolithic CPU compared to likes of 7950x, which is always going to suit games better.

EDIT: Wasnt 7950x or any of the AMDs Zen4 CPUs tested elsewhere in any of these games, with 5600 RAM and 4090 in 720p and max settings, so we can make a direct comparison? Or even better, did not someone test any of these games with both AMD and Intel at same RAM speed above 6000 MHz?
EDIT: Nvm, seems it wasnt.

Here you go. Of course, some will completely ignore this, but it seems like you are actually open to more data, even if it doesn't fit a pre-coneived notion.

Edit: Eurogamer/Digital Foundry also did some CPU tests with RT enabled that show Zen4 being very competitive with RPL once both use faster memory (https://www.eurogamer.net/digitalfoundry-2022-intel-core-i9-13900k-core-i5-13600k-review?page=3). Zen4 is especially strong in Metro Exodus Enhanced where it shows a significant advantage over RPL (up to 25% between the top SKUs).

 

[Carfax83]

Here you go. Of course, some will completely ignore this, but it seems like you are actually open to more data, even if it doesn't fit a pre-coneived notion.

That chart from HWUB actually jives with what I've been saying. That Raptor Lake's cache bandwidth must be a factor because Raptor Lake and Alder Lake have essentially the same core architecture, except that Raptor Lake has more L2 cache and much higher cache bandwidth. Raptor Lake also has more intelligent prefetching, but that's nothing compared to the cache bandwidth advantage that it enjoys over Alder Lake.

Edit: Eurogamer/Digital Foundry also did some CPU tests with RT enabled that show Zen4 being very competitive with RPL once both use faster memory (https://www.eurogamer.net/digitalfoundry-2022-intel-core-i9-13900k-core-i5-13600k-review?page=3). Zen4 is especially strong in Metro Exodus Enhanced where it shows a significant advantage over RPL (up to 25% between the top SKUs).

That review was done with an RTX 3090, so it's heavily GPU bottlenecked.

 

[nicalandia]

No OC By any means allowed on the non-K SKUs

https://twitter.com/x/status/1615514745989402624

 

[Mopetar]

Anyone doing a BCLK overclock these days is probably going to run an unlocked CPU anyways. Pretty much a non-issue for 99.99% of the population.

 

[lightmanek]

Anyone doing a BCLK overclock these days is probably going to run an unlocked CPU anyways. Pretty much a non-issue for 99.99% of the population.

That's still more than 80000000 people who would like to BCLK their new non-K processor (taking world population and discounting 99.99% of them )

[math is dangerous]

 

[Hulk]

That's still more than 80000000 people who would like to BCLK their new non-K processor (taking world population and discounting 99.99% of them )

[math is dangerous]

Actually 800,000. You are off by a factor of 100. This is of course assuming 8 billion people presently on the planet.

 

[scannall]

That was always a bad idea. 3% tops, and risking data corruption and system problems was never worth it.

 

[Exist50]

That was always a bad idea. 3% tops, and risking data corruption and system problems was never worth it.

IIRC, they decoupled the clock used for PCIe and such, so there wasn't any particular risk vs multiplier overclocking.

 

[coercitiv]

IIRC, they decoupled the clock used for PCIe and such

They sure did, the OC done with early Alder Lake CPUs was in the range of 30%+.

 

[Ranulf]

Unfortunately no. I honestly do wish that PCGH.de and Computerbase.de would test with faster memory, but I also understand why they don't and I don't fault them for it.

AMD says the "sweet spot" for AM5 is DDR5-6000 remember. Is there a best ram speed for Intel?

https://www.tomshardware.com/news/amd-confirms-ddr5-6000-ram-is-the-sweet-spot-for-ryzen-7000-cpus

 

[igor_kavinski]

Is there a best ram speed for Intel?

That's the crazy thing. Intel's architecture just impressively scales and scales with faster RAM. We don't have a limit yet. Granted that you won't see that scaling in every workload but the benefit is apparent in bandwidth starved workloads.

 

[DooKey]

Pot, meet kettle...

I agree that both of them are two sides of the same coin at times.

 

[alcoholbob]

AMD says the "sweet spot" for AM5 is DDR5-6000 remember. Is there a best ram speed for Intel?

https://www.tomshardware.com/news/amd-confirms-ddr5-6000-ram-is-the-sweet-spot-for-ryzen-7000-cpus

The only time I've seen a difference in benchmarks for RPL is with ray tracing enabled, because for whatever reason RT seems to often offload work onto system RAM instead of VRAM.

 

[igor_kavinski]

The only time I've seen a difference in benchmarks for RPL is with ray tracing enabled, because for whatever reason RT seems to often offload work onto system RAM instead of VRAM.

Maybe due to latency reasons? VRAM latency is much higher than that of system RAM.

 

[lopri]

No OC By any means allowed on the non-K SKUs
View attachment 74849

https://twitter.com/x/status/1615514745989402624

You can't undervolt those on B660/760, either, without tanking performance. Need a Z690/790 for that.