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.

 

[Det0x]

^^ No point to post LN2 singlecore Alder lake maxbench runs as a comparison..

This 13900K is 7% faster singlecore compared to average Core i9-12900K (not even KS) based on official ranking, so not really impressive when you look at the clockspeeds it were running:

https://twitter.com/x/status/1546488389234356224

This fits with earlier CPU-Z benchmark leaks

• the 13900K scores 880 @ 5.5Ghz
• the 12900K scores 834 @ 5.2Ghz

Zero % IPC increase on the big cores compared to regular Alder Lake ? How many % faster is it compared to the 12900KS ?

*edit*

Stock 12900KS shows ~2080/19000 in GB5:
1. tech4gamers . com/intel-i9-12900k-vs-intel-i9-12900ks/#geekbench-5-multi-core
2. hothardware. com/reviews/intel-core-i9-12900ks-cpu-review-and-benchmarks?page=2

37% faster in multi-core Geekbench 5, which is one of the few ~best-case benchmark where the e-cores actually are a benefit, from the theoretical + 33% from twice as many e-cores ?
Not so impressed by the "IPC increase" from the doubling of L2 on e-cores either.. (rather increased P2 power limit we see here)

Lastly worth remembering is that the 13900k were running DDR5-6400 MT/s memory which also affect the overall score in GB5


This is my max tuned PBO CO watercooled 5950x GB5 score for a comparison:

Micro-Star International Co., Ltd. MS-7D51 - Geekbench

Benchmark results for a Micro-Star International Co., Ltd. MS-7D51 with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

My 2cent, think bestcase RaptorLake will be bearly the fastest singlecore CPU in highly synthetic benchmarks while it will get stomped in about everything else by the the 7950x3d. (including gaming)
When you consider IO, powerusage and expected lifetime these two new competing platforms it gets even worse.. But as positive note, its looking like RaptorLake will fare better in the mid to low-end where price means more than maximum performance. (oh how the tables have turned)

 

[nicalandia]

Intel Core i9-13900K 5.5 GHz Raptor Lake CPU Benchmarked, Crushes The Ryzen 9 5950X & Core i9-12900K In Single & Multi-Threaded Tests

Intel's Core i9-13900K 5.5 GHz Raptor Lake CPU has been benchmarked within Geekbench 5 and obliterates the Core i9-12900K & Ryzen 9 5950X.

wccftech.com

Beast 43% faster than 12900k in multi-core performance before final clocks.. will be even higher in final silicon 😁😃

I was boosting to 5.7 Ghz so very close to the alleged 5.8 Ghz final CPU

 

[FangBLade]

Intel Core i9-13900K 5.5 GHz Raptor Lake CPU Benchmarked, Crushes The Ryzen 9 5950X & Core i9-12900K In Single & Multi-Threaded Tests

Intel's Core i9-13900K 5.5 GHz Raptor Lake CPU has been benchmarked within Geekbench 5 and obliterates the Core i9-12900K & Ryzen 9 5950X.

wccftech.com

Beast 43% faster than 12900k in multi-core performance before final clocks.. will be even higher in final silicon 😁😃

Rocket Lake was also "beast" in leaks, we know how it ended. RPL will have very high power consumption, dead end platform, big/little complexity/bugs etc. Zen 4 at worst will have very similar MT performance while consuming much much less power, + 3dcache will have no problems destroying it in gaming, while non 3dcache will have about 7-10% less performance in synthetic benchmarks, we will see about gaming, but as i said, 3dcache will take care of everything, and second generation will work at higher clocks and probably OC allowed + ddr5 so you get boost in games that doesn't like cache, and AM5 support for at least 2 next gen cpu-s.

 

[Saylick]

So no real IPC gains for Raptorlake, just 8 more E cores and then running up the TDP to allow for 12900KS clocks, essentially. If that's all it is shaping up to be, why the heck can't Intel roll it out any faster? This sounds like what happened with Coffeelake, in that it was basically Skylake with just 2 more cores.

If the 5950X gets ~1800 GB5 ST score with a tuned memory profile and Zen 4 brings 10% higher clocks and 7% higher IPC, then that's already enough for Zen 4 to tie the ~2100 score of the 13900K. Once you add in V-cache, there's likely no contest in gaming workloads.

 

[coercitiv]

and Zen 4 brings 10% higher clocks

For ST it's ~15% higher clocks, Intel will push RTL like crazy for that last bit of juice.

This time around I think even memory OC will matter.

 

[Thunder 57]

Rocket Lake was also "beast" in leaks, we know how it ended. RPL will have very high power consumption, dead end platform, big/little complexity/bugs etc. Zen 4 at worst will have very similar MT performance while consuming much much less power, + 3dcache will have no problems destroying it in gaming, while non 3dcache will have about 7-10% less performance in synthetic benchmarks, we will see about gaming, but as i said, 3dcache will take care of everything, and second generation will work at higher clocks and probably OC allowed + ddr5 so you get boost in games that doesn't like cache, and AM5 support for at least 2 next gen cpu-s.

Well considering @Henry swagger expects 6GHz, maybe not. 5.7 or 5.8 seems more realistic. Also the base frequency regressed. Maybe it's not final, but I would not be surprised if it was because of the additional 8 cores. That would only be a factor when pretty much every core is used, though.

So no real IPC gains for Raptorlake, just 8 more E cores and then running up the TDP to allow for 12900KS clocks, essentially. If that's all it is shaping up to be, why the heck can't Intel roll it out any faster? This sounds like what happened with Coffeelake, in that it was basically Skylake with just 2 more cores.

If the 5950X gets ~1800 GB5 ST score with a tuned memory profile and Zen 4 brings 10% higher clocks and 7% higher IPC, then that's already enough for Zen 4 to tie the ~2100 score of the 13900K. Once you add in V-cache, there's likely no contest in gaming workloads.

I think there will be some workloads that benefit more because of the cache increases alone. Still, not a great first impression. It could be a close race between Raptor Lake and Zen 4.

 

[Saylick]

For ST it's ~15% higher clocks, Intel will push RTL like crazy for that last bit of juice.

This time around I think even memory OC will matter.

Yeah, you're right... ~5 GHz ST clocks for the 5950X increases to ~5.7 GHz for Zen 4. I also misquoted the IPC uplift at 7% instead of the 8-10% advertised by AMD. That's still around a 23-25% ST uplift and is enough to match the 13900K with ease. We'll just have to see what kind of power it pulls in comparison.

I think there will be some workloads that benefit more because of the cache increases alone. Still, not a great first impression. It could be a close race between Raptor Lake and Zen 4.

You're right. The extra L3 from the extra 8 E cores will likely help with gaming workloads, but it's nowhere near the same level of boost as adding 64 MB of cache that AMD did with the 5800X3D. I, too, expect a close race between the two... until Zen 4 with V-cache comes out at least.

 

[FangBLade]

Well considering @Henry swagger expects 6GHz, maybe not. 5.7 or 5.8 seems more realistic. Also the base frequency regressed. Maybe it's not final, but I would not be surprised if it was because of the additional 8 cores. That would only be a factor when pretty much every core is used, though.



I think there will be some workloads that benefit more because of the cache increases alone. Still, not a great first impression. It could be a close race between Raptor Lake and Zen 4.

Yeah, 8 additional cores, cache, clocks all while on the same node, and ADL already consumes too much power so RPL will be even worse, you can't negate 8 additional cores and clocks while staying on the same node.

 

[Thunder 57]

Yeah, you're right... ~5 GHz ST clocks for the 5950X increases to ~5.7 GHz for Zen 4. I also misquoted the IPC uplift at 7% instead of the 8-10% advertised by AMD. That's still around a 23-25% ST uplift and is enough to match the 13900K with ease. We'll just have to see what kind of power it pulls in comparison.


You're right. The extra L3 from the extra 8 E cores will likely help with gaming workloads, but it's nowhere near the same level of boost as adding 64 MB of cache that AMD did with the 5800X3D. I, too, expect a close race between the two... until Zen 4 with V-cache comes out at least.

I'm curios as to what latencies the new caches will have. If these numbers are real that increase in L2 and L3 doesn't have much impact, at least in this benchmark. Surely there is a hit in cycles for L2. Maybe Intel can get away with it by clocking it to the moon though.

I think Zen 4 and Raptor lake may be close in overall performance, but I expect AMD to be vastly more efficient. Sort of ironic considering it is Intel that has the efficiency cores. I expect Zen 4+3D will be the overall winner whenever it comes out.

 

[pakotlar]

I'm curios as to what latencies the new caches will have. If these numbers are real that increase in L2 and L3 doesn't have much impact, at least in this benchmark. Surely there is a hit in cycles for L2. Maybe Intel can get away with it by clocking it to the moon though.

I think Zen 4 and Raptor lake may be close in overall performance, but I expect AMD to be vastly more efficient. Sort of ironic considering it is Intel that has the efficiency cores. I expect Zen 4+3D will be the overall winner whenever it comes out.

Yeah this seems very likely. Both MT and ST seem likely to go to Intel at a PPW disadvantage, though closer than 12900K vs 5950x.

 

[poke01]

many PC folks think geekbench is bad but for bursty/single core tasks its a very GOOD indicator. Not real world but it gives a base line to compare with other chips.

 

[pakotlar]

many PC folks think geekbench is bad but for bursty/single core tasks its a very GOOD indicator. Not real world but it gives a base line to compare with other chips.

Agreed. Also, Geekbench seems to not benefit much from cache:https://www.tomshardware.com/news/ryzen-7-5800x3d-beats-ryzen-7-5800x-by-9-in-geekbench-5. If thats the case, the Geekbench 5 result is likely a worst case for Raptor Lake, since Raptor Lake’s IPC gains are likely entirely L2 cache driven (1.25MB->2MB in Raptor Cove). Probably similar to 5800x 3D vs 5800x, at best.

 

[Saylick]

I think Zen 4 and Raptor lake may be close in overall performance, but I expect AMD to be vastly more efficient. Sort of ironic considering it is Intel that has the efficiency cores. I expect Zen 4+3D will be the overall winner whenever it comes out.

Efficiency can be measured in multiple ways. The E cores aren't worth writing home about in terms of perf/W, but they really do help Intel add a ton of perf/mm2.

 

[pakotlar]

Efficiency can be measured in multiple ways. The E cores aren't worth writing home about in terms of perf/W, but they really do help Intel add a ton of perf/mm2.

Yeah. One clarifying point, according to Chips and Cheese Gracemont is fairly efficient below 15W/4 core cluster (~6W for uncore), at least compared to Golden Cove. So it has both power and space efficiency benefits, at least when clocked correctly (aka not what they did with Alder Lake desktop).

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption. For example, Anandtech measured 272 W of package power during a POV-Ray run. Our o…

chipsandcheese.com

edit: fix.

 

[Thunder 57]

Efficiency can be measured in multiple ways. The E cores aren't worth writing home about in terms of perf/W, but they really do help Intel add a ton of perf/mm2.

Very true. When looking at from an area perspective, the E cores are tough to beat.

 

[Saylick]

Yeah. One clarifying point, according to Chips and Cheese Gracemont is fairly efficient below 15W/4 core cluster (~6W for uncore), at least compared to Golden Cove. So it has both power and space efficiency benefits, at least when clocked correctly (aka not what they did with Alder Lake desktop).

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption. For example, Anandtech measured 272 W of package power during a POV-Ray run. Our o…

chipsandcheese.com

edit: fix.

Intel likely only clocked the E cores so high because they needed to catch up to Zen 3 in terms of MT performance, and the power budget was better spent on making the E cores run at 4 GHz rather than eking out another 100-200 MHz from the P cores. In a perfect world where Intel didn't have such a lofty target to aim for, they would have kept the E cores limited to 3.2 GHz where, on paper, the design would have been very elegant, but alas Alderlake became an imperfect product that was stretched beyond its comfort zone due to competitive pressures. If Raptorlake puts a limit on the E cores max frequency so that they stay within their efficient operating range, it would not surprise me if Raptorlake was what Alderlake should have been.

 

[pakotlar]

Intel likely only clocked the E cores so high because they needed to catch up to Zen 3 in terms of MT performance, and the power budget was better spent on making the E cores run at 4 GHz rather than eking out another 100-200 MHz from the P cores. In a perfect world where Intel didn't have such a lofty target to aim for, they would have kept the E cores limited to 3.2 GHz where, on paper, the design would have been very elegant, but alas Alderlake became an imperfect product that was stretched beyond its comfort zone due to competitive pressures. If Raptorlake puts a limit on the E cores max frequency so that they stay within their efficient operating range, it would not surprise me if Raptorlake was what Alderlake should have been.

Correct, and I’m also hopeful that uncore efficiency will be better in Raptor Lake; that seems to hurt low power mobile Alder Lake power efficiency. To your last point, Intel’s emphasis on increasing L2 caches in Raptor Lake (1.28MB->2MB in Raptor Cove and 2->4MB in Gracemont) may have significant efficiency benefits, given how much of a penalty they take for L3 access compared to Zen 3.

See Chips and Cheese recent analysis on memory subsystem in recent Intel and AMD parts: https://chipsandcheese.com/2022/07/07/alder-lakes-caching-and-power-efficiency/

 

[DrMrLordX]

many PC folks think geekbench is bad but for bursty/single core tasks its a very GOOD indicator. Not real world but it gives a base line to compare with other chips.

tbh too many people rely on GB5, especially when comparing different microarchitectures. Comparing Raptor Lake to Alder Lake, though, seems like fair game. Same basic uarch, same process.

 

[jpiniero]

So no real IPC gains for Raptorlake, just 8 more E cores and then running up the TDP to allow for 12900KS clocks, essentially. If that's all it is shaping up to be, why the heck can't Intel roll it out any faster? This sounds like what happened with Coffeelake, in that it was basically Skylake with just 2 more cores.

Alder Lake K has only been available 8 months. I think they want to keep it closer to 10-12, especially when who knows what will be released after Raptor Lake if Intel 4 is going to get another year delay for all practical purposes.

 

[pakotlar]

tbh too many people rely on GB5, especially when comparing different microarchitectures. Comparing Raptor Lake to Alder Lake, though, seems like fair game. Same basic uarch, same process.

Fair game, but GB5 doesn’t appear to benefit from larger caches, see 5800X 3D. So it may serve as a lower bound for IPC gains (here ~0), much like we saw with 5800X 3D.

 

[dullard]

Intel likely only clocked the E cores so high because they needed to catch up to Zen 3 in terms of MT performance, and the power budget was better spent on making the E cores run at 4 GHz rather than eking out another 100-200 MHz from the P cores. In a perfect world where Intel didn't have such a lofty target to aim for, they would have kept the E cores limited to 3.2 GHz where, on paper, the design would have been very elegant, but alas Alderlake became an imperfect product that was stretched beyond its comfort zone due to competitive pressures. If Raptorlake puts a limit on the E cores max frequency so that they stay within their efficient operating range, it would not surprise me if Raptorlake was what Alderlake should have been.

I basically agree with you. But, I think it is a bit more complex than you laid out. Intel is competing against AMD, but Intel is also competing against Intel. That is, Intel needs to give people with 3 year to 5 year old computers a reason to upgrade. Think of a 9700K user (Coffee Lake). They already have 8 cores running 3.6 GHz base and 4.9 GHz turbo. The 12700K has 8 P cores, 3.6 GHz base and 5.0 GHz turbo. If Intel lowered the power levels from what Alder Lake used, then the 12700K would have slower base and turbo speeds than what the 9700K user already has. Adding in a measly 4 E cores and hyperthreading wouldn't really be sufficient to get the upgrade itch going if the frequencies also drop.

Intel's other problem was tying the E core and P core voltages. Pushing to get the P core speeds up to older generation speeds therefore meant the E cores were going to use way too much power. Might as well get the frequency out of the E cores since the power is being used. There are two possible fixes for that. (1) break the voltage link or (2) adding more E cores. Either one would let the E cores run much closer to the power region where they actually are efficient in power and efficient in die area. Raptor Lake will at least do item #2; item #1 is still up for debate.

So, yes, Raptor Lake will be what Alder Lake should have been. But the real benefits of E cores won't be seen until there are even more of them to truly run them in their efficient part of the power curve. Think Arrow Lake.

 

[AdamK47]

^^ No point to post LN2 singlecore Alder lake maxbench runs as a comparison..

This 13900K is 7% faster singlecore compared to average Core i9-12900K (not even KS) based on official ranking, so not really impressive when you look at the clockspeeds it were running:

https://twitter.com/x/status/1546488389234356224

This fits with earlier CPU-Z benchmark leaks

• the 13900K scores 880 @ 5.5Ghz
• the 12900K scores 834 @ 5.2Ghz

Zero % IPC increase on the big cores compared to regular Alder Lake ? How many % faster is it compared to the 12900KS ?

*edit*


37% faster in multi-core Geekbench 5, which is one of the few ~best-case benchmark where the e-cores actually are a benefit, from the theoretical + 33% from twice as many e-cores ?
Not so impressed by the "IPC increase" from the doubling of L2 on e-cores either.. (rather increased P2 power limit we see here)

Lastly worth remembering is that the 13900k were running DDR5-6400 MT/s memory which also affect the overall score in GB5
View attachment 64330

This is my max tuned PBO CO watercooled 5950x GB5 score for a comparison:
View attachment 64333

Micro-Star International Co., Ltd. MS-7D51 - Geekbench

Benchmark results for a Micro-Star International Co., Ltd. MS-7D51 with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

My 2cent, think bestcase RaptorLake will be bearly the fastest singlecore CPU in highly synthetic benchmarks while it will get stomped in about everything else by the the 7950x3d. (including gaming)
When you consider IO, powerusage and expected lifetime these two new competing platforms it gets even worse.. But as positive note, its looking like RaptorLake will fare better in the mid to low-end where price means more than maximum performance. (oh how the tables have turned)

Doesn't seem all that impressive right now. Could improve for retail samples.

My stable 12900KS results for comparison. 5.5GHz single core 5.4/5.3GHz P-Core and 4.2GHz E-Core.

Gigabyte Technology Co., Ltd. Z690 AORUS MASTER - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. Z690 AORUS MASTER with an Intel Core i9-12900KS processor.

browser.geekbench.com

 

[dullard]

Doesn't seem all that impressive right now. Could improve for retail samples.

My stable 12900KS results for comparison. 5.5GHz single core 5.4/5.3GHz P-Core and 4.2GHz E-Core.

Gigabyte Technology Co., Ltd. Z690 AORUS MASTER - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. Z690 AORUS MASTER with an Intel Core i9-12900KS processor.

browser.geekbench.com

An ES chip slightly beating an overclocked chip that costs ~$150 more (since the final price is still not known, I just used the 12900K price here). Not a night and day difference, but it is certainly a move in the right direction.

 

[Asterox]

Well, it is expected.

https://twitter.com/x/status/1547225143025356800

 

[Hitman928]

Well, it is expected.

https://twitter.com/x/status/1547225143025356800

The Cinebench R20 results are, um, a bit suspicious. It also seems like they are doing a static overclock of both chips so stock performance will be different.