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.

 

[Carfax83]

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

If one thing is clear to me, it's that Golden Cove is a high bandwidth architecture. ADL at it's inception was bandwidth starved at both the cache and memory level. That's why RPL can scale so much better than ADL simply due to cache and memory controller upgrades despite having the same core microarchitecture.

And Raptor Cove seems to consume as much bandwidth as you can give it. Not every workload will scale with bandwidth of course, but achieving high throughput in vectorized workloads for instance requires tons of bandwidth; especially at the cache level. That's how RPL is able to compete with Zen 4 in a lot of workloads despite having less cores.

 

[Carfax83]

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

We bring balance to the Force

 

[Abwx]

Review of a low cost ADL/RL at Computerbase :

Intel Core i5-13400F im Test: Gaming-Benchmarks

Intels 13. Gen Core bringt i5-CPUs unterhalb des 13600K mehr Kerne, doch die Architektur ist die alte. Das zeigt sich in Gaming-Benchmarks.

www.computerbase.de

 

[Hulk]

If one thing is clear to me, it's that Golden Cove is a high bandwidth architecture. ADL at it's inception was bandwidth starved at both the cache and memory level. That's why RPL can scale so much better than ADL simply due to cache and memory controller upgrades despite having the same core microarchitecture.

And Raptor Cove seems to consume as much bandwidth as you can give it. Not every workload will scale with bandwidth of course, but achieving high throughput in vectorized workloads for instance requires tons of bandwidth; especially at the cache level. That's how RPL is able to compete with Zen 4 in a lot of workloads despite having less cores.

I remember reading about the deficiencies in Alder cache structure here.

Going Armchair Quarterback on Golden Cove’s Caches

Processor speed has rapidly outpaced advances in DRAM technology, so caching strategy is a huge part of CPU performance today. A couple of articles ago, we saw Intel’s decisions with Golden C…

chipsandcheese.com

 

[Henry swagger]

https://twitter.com/x/status/1616401152622428160

13900ks loves bandwidth.. with new ddr5 8600 the lead will grow

 

[Markfw]

https://twitter.com/x/status/1616401152622428160

13900ks loves bandwidth.. with new ddr5 8600 the lead will grow

While capframex is also (like you) an Intel supporter, the article does not say is blows Zen 4 out of the water. It even used words like "buying the Intel is at times questionable" and "when the 700x3d comes out, it will win".

 

[igor_kavinski]

Even saddled with slower RAM in Handbrake 4K, 7950X says hello and bye to the KS

 

[Ranulf]

DDR5 wins! At a cost. 13900k ddr5 vs ddr4 gaming benches

DDR5 at 7600mhz, he goes over all his specs. Benches start at 7m55s or so.

 

[igor_kavinski]

DDR5 wins! At a cost. 13900k ddr5 vs ddr4 gaming benches

DDR5 at 7600mhz, he goes over all his specs. Benches start at 7m55s or so.

View attachment 75006

The average performance delta is not worth the additional cost of DDR5 mobo and RAM.

 

[Mopetar]

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.

Really, any CPU will scale with faster memory. It's just that Zen has different points where the memory speed manages to synchronize better with the Infinity Fabric speed and when you get farther away from those points there's not really any recognition of the increase in RAM speed because the data getting their faster just means a longer wait to get moved around on chip. Slower RAM that gets it there just in time is essentially just as effective.

The funny thing is that while Intel will scale in step with the RAM speeds to a degree, there's eventually a point where AMD would suddenly gain a massive jump in performance when the speed lines up with the infinity fabric again and the data can arrive and be passed along sooner.

I don't know if we ever get there with DDR5 and how much such exotic RAM kits would cost, but it's not quite as clear cut as Intel scaling and AMD being stuck at a certain point.

 

[Hitman928]

https://twitter.com/x/status/1616401152622428160

13900ks loves bandwidth.. with new ddr5 8600 the lead will grow

Hmm, I thought that running only one company's product with overclocked memory and overclocked CPU while leaving the other at stock CPU and dog slow JEDEC memory was supposed to be evidence of clear bias and that their results couldn't be trusted. Seems no one has an issue with it in this case though. Strange .

 

[Markfw]

Hmm, I thought that running only one company's product with overclocked memory and overclocked CPU while leaving the other at stock CPU and dog slow JEDEC memory was supposed to be evidence of clear bias and that their results couldn't be trusted. Seems no one has an issue with it in this case though. Strange .

I missed that part. Its hard to read a translated article. Yes, this review is total crap.

 

[exquisitechar]

https://twitter.com/x/status/1616401152622428160

13900ks loves bandwidth.. with new ddr5 8600 the lead will grow

That’s a massive gap in favor of Raptor Lake. I was correct to say that even Raphael X3D will struggle against it when I saw that Raphael’s memory OC potential is underwhelming. There’s little reason to get Raphael for gaming, especially because high speed DDR5 is not even that expensive anymore.

 

[DrMrLordX]

That’s a massive gap in favor of Raptor Lake. I was correct to say that even Raphael X3D will struggle against it when I saw that Raphael’s memory OC potential is underwhelming. There’s little reason to get Raphael for gaming, especially because high speed DDR5 is not even that expensive anymore.

Raphael-X will do very well. Wait and see.

 

[Markfw]

That’s a massive gap in favor of Raptor Lake. I was correct to say that even Raphael X3D will struggle against it when I saw that Raphael’s memory OC potential is underwhelming. There’s little reason to get Raphael for gaming, especially because high speed DDR5 is not even that expensive anymore.

So you missed the fact that they used fast memory on Raptor lake, and slow memory on Zen 4 ? As that invalidates any results. This review is crap.

 

[nicalandia]

That’s a massive gap in favor of Raptor Lake. I was correct to say that even Raphael X3D will struggle against it when I saw that Raphael’s memory OC potential is underwhelming. There’s little reason to get Raphael for gaming, especially because high speed DDR5 is not even that expensive anymore.

The magic of X3D parts is that it does not care what Speed of RAM you have.

 

[JoeRambo]

So you missed the fact that they used fast memory on Raptor lake, and slow memory on Zen 4 ? As that invalidates any results. This review is crap.

"Invalidates" is a strong word. In fact the review shows stock results for 13900K and 7950x, running 5600 and 5200 DDR5 speeds accordingly. Raptor Lake is ahead by 24% in 13900KS incarnation.

How much can typical Z4 guy can expect to add? 800-1000Mts + tuning, ending up with 6000-6200 speed. While Intel can add 1600-2000Mts + tuning, ending up with 7200-7600 speed. Does not take a rocket scientist to realize that Intel will scale better even if we ignore the fact that Intel scales better with faster RAM due to core being wider.
So review is perfectly fine, stronger core with better memory subsystem beats the hell out of weaker core with 1st generation DDR5 controller in gaming where those things matter the most.

So those are the objective facts, what these reviews fail to mention are, I can expand on them since I own and overclocked both systems:

1) Running XMP memory speeds and timings for AMD and Intel leaves plenty of performance on table, tightening secondary and tertiary timings is a must. In this department Intel seems to go further, while AMD is more limited and harder to push ( see 2 ).
2) Anyone tuning memory on AM5 is beta tester for AMD. Experience is horrible, safe boot after memory training failure does not work, BIOS is missing certain options. Some BIOS'es have same memory timings and settings duplicated in 2-3 vendor and AMD menus and duplicates have crazy interactions between them.
So You will be fighting for those nanoseconds. AM5 has dubiuos achievement of more ruined OS installs and BIOS resets with battery shorting than all previuos combined OC experience since ABIT BH6. Things are improving with each BIOS, but compared to Z690 + 13900KS it is still hell on earth and nothing to show for.
3) DDR4 is still going very strong on Raptor Lake. 4200-4300 speeds seem to be doable and tuned are matching DDR5 8000. So Raptor Lake does not have to be expensive, $200 Z690 + 2x16GB B-Die DR ram and after tuning You can score better than in review, since they most likely ran DDR5 XMP and 7400 speed is just 3700 IMC clock, compared to 4000+ for DDR5.
4) DDR5 will continue scaling for Intel, while Zen4 seems to be stuck hard in 6200-6400 region, there seems to be both hard wall in clock and due to them internal datapaths like Infinity Fabric don't scale beyond 2200ish clocks.

So for proper gaming machine from AMD one needs to wait for X3D parts and Zen4 based laptop chips.

 

[Hitman928]

"Invalidates" is a strong word. In fact the review shows stock results for 13900K and 7950x, running 5600 and 5200 DDR5 speeds accordingly. Raptor Lake is ahead by 24% in 13900KS incarnation.

How much can typical Z4 guy can expect to add? 800-1000Mts + tuning, ending up with 6000-6200 speed. While Intel can add 1600-2000Mts + tuning, ending up with 7200-7600 speed. Does not take a rocket scientist to realize that Intel will scale better even if we ignore the fact that Intel scales better with faster RAM due to core being wider.
So review is perfectly fine, stronger core with better memory subsystem beats the hell out of weaker core with 1st generation DDR5 controller in gaming where those things matter the most.

So those are the objective facts, what these reviews fail to mention are, I can expand on them since I own and overclocked both systems:

1) Running XMP memory speeds and timings for AMD and Intel leaves plenty of performance on table, tightening secondary and tertiary timings is a must. In this department Intel seems to go further, while AMD is more limited and harder to push ( see 2 ).
2) Anyone tuning memory on AM5 is beta tester for AMD. Experience is horrible, safe boot after memory training failure does not work, BIOS is missing certain options. Some BIOS'es have same memory timings and settings duplicated in 2-3 vendor and AMD menus and duplicates have crazy interactions between them.
So You will be fighting for those nanoseconds. AM5 has dubiuos achievement of more ruined OS installs and BIOS resets with battery shorting than all previuos combined OC experience since ABIT BH6. Things are improving with each BIOS, but compared to Z690 + 13900KS it is still hell on earth and nothing to show for.
3) DDR4 is still going very strong on Raptor Lake. 4200-4300 speeds seem to be doable and tuned are matching DDR5 8000. So Raptor Lake does not have to be expensive, $200 Z690 + 2x16GB B-Die DR ram and after tuning You can score better than in review, since they most likely ran DDR5 XMP and 7400 speed is just 3700 IMC clock, compared to 4000+ for DDR5.
4) DDR5 will continue scaling for Intel, while Zen4 seems to be stuck hard in 6200-6400 region, there seems to be both hard wall in clock and due to them internal datapaths like Infinity Fabric don't scale beyond 2200ish clocks.

So for proper gaming machine from AMD one needs to wait for X3D parts and Zen4 based laptop chips.

Techpowerup showed a 1.5% performance increase on average for a 13900K going from DDR5-6000 to DDR5-7200. This led to a 13.8% lead over a 7950x running DDR5-6000 memory, a far cry from the lead PCGH showed with Zen4 running with DDR5-5200 memory. Other outlets have shown RPL and Zen4 being even closer in performance when both are using fast memory. It doesn't seem like faster and faster memory will help RPL much whereas Zen4-3d should see a significant performance increase and be able to do so with much cheaper memory. I don't expect the 7950x3d to have a real lead on a 13900k running fast memory, but I expect the reverse will also be true (i.e. they'll be basically even). The difference will be in the cost and power usage (if 5800x3d is anything to go by, Zen4-3d CPUs should be extremely power efficient while gaming).

 

[Harry_Wild]

Still Intel not release in the channel for the non K 13600! Everywhere I try to buy it, they sales guy tell me to just buy the 13600K and setup the motherboard bios to 65W!

 

[AdamK47]

So you missed the fact that they used fast memory on Raptor lake, and slow memory on Zen 4 ? As that invalidates any results. This review is crap.

How does that invalidate results? Zen 4 is incapable of higher memory speeds. Raptor Lake is capable of 7600+ easily. It's fair.

 

[Mopetar]

How does that invalidate results? Zen 4 is incapable of higher memory speeds. Raptor Lake is capable of 7600+ easily. It's fair.

It doesn't change the absolute result for the Intel CPU, but it makes any percentage gap quoted rather pointless.

May as well just pull the power cable out from the AMD box and declare the 13900K infinity percentage better because you're only doing it to feel good at that point. Why not feel better?

 

[AdamK47]

It doesn't change the absolute result for the Intel CPU, but it makes any percentage gap quoted rather pointless.

May as well just pull the power cable out from the AMD box and declare the 13900K infinity percentage better because you're only doing it to feel good at that point. Why not feel better?

A fair comparison would be Zen 4 at 6200 and Raptor Lake at 7600 since those are their respective sweet spots.

 

[Markfw]

A fair comparison would be Zen 4 at 6200 and Raptor Lake at 7600 since those are their respective sweet spots.

Exactly !

 

[Abwx]

Hmm, I thought that running only one company's product with overclocked memory and overclocked CPU while leaving the other at stock CPU and dog slow JEDEC memory was supposed to be evidence of clear bias and that their results couldn't be trusted. Seems no one has an issue with it in this case though. Strange .

What is curious is that they display the frequencies for the CPUs, so their 7600X/7700X can hit 5.5GHz in thoses games but the 7950X is listed as 5.3GHz, dunno what it means...

That being said Computerbase put the difference at 12% for stock settings using comparable games, so 20% difference is definitly some kind of biased set testing, wich is somewhat unsurprising given the source.

 

[Mopetar]

Definitely, some kind of biased set testing, wich is somewhat unsurprising given the source.

So you're saying he should change his name to CarpFrameX given how fishy his numbers are?