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]

Bottom line is that I think that Raptor Lake and Zen 4 are competitive in cases where "normal load" is 0- ~12 threads. Once over that the e-cores start to lose to Zen 4 cores.

I'm not exactly following your logic here? Here's how I see it.

Up to 16 threads in use I think Zen 4 has the compute advantage.
Zen 4 would be using only physical Zen 4 cores for each thread.
Raptor Lake would be using 8 P cores and and then E cores for threads 9 through 16. Obviously Zen 4 physical core is more performant than an E core so that is why I think Zen 4 would be more performant here.

17 to 24 threads. This is more difficult as I think Zen 4 has the advantage to about 20 threads as 16 Zen 4 physical cores plus 4 logical cores is probably more performant than 8 physical P cores and 12 E cores. But from 21 to 24 threads Zen 4 is using more and more logical cores while Raptor is employing more physical E cores.
Zen 4 would be using 16 physical cores and 1 to 8 logical cores.
Raptor Lake would be using 8 P cores and 9 to 16 E cores.

24+ threads I seen Raptor having a slight advantage since it is probable that E cores are more performant than Zen 4 logical cores.

But really this is splitting hairs as both of these CPU's are so evenly matched is kind of crazy.

 

[nicalandia]

Intel 13900HX vs 13800HX, both true Raptor Lake, no much difference in performance.

ASUSTeK COMPUTER INC. ROG Strix G634JZ_G634JZ_000031386 vs ASUSTeK COMPUTER INC. ROG Strix G634JZ_G634JZ_000031386 - Geekbench

 

[Hitman928]

Reviewers usually don't test this, so data is limited, but Computerbase has a comparison, although limited RT titles with a 3090 Ti.

Games with RT:

View attachment 73245

Games without RT:

View attachment 73246

Of particular interest is the 7900X vs the 7600X. The effect seems to be more pronounced there.

So it's not actually well documented. . .

What you've shown isn't even really a test of what you claimed. With RT on, all of the Zen 4 results are well within margin of error and suggests some kind of bottleneck on the architecture/platform/code rather than something to do with the CCD configuration.

 

[tamz_msc]

So it's not actually well documented. . .

What you've shown isn't even really a test of what you claimed. With RT on, all of the Zen 4 results are well within margin of error and suggests some kind of bottleneck on the architecture/platform/code rather than something to do with the CCD configuration.

It is a fact. People have reported it, even if reviewers haven't tested it properly.

Dual CCDs have caused performance issues in games ever since they became a thing with AMD CPUs.

Unless the reviewer provides error bars, it is wrong to suggest the data is within margin of error.

 

[Hitman928]

It is a fact. People have reported it, even if reviewers haven't tested it properly.

Dual CCDs have caused performance issues in games ever since they became a thing with AMD CPUs.

Unless the reviewer provides error bars, it is wrong to suggest the data is within margin of error.

Testing it properly is what is required but more importantly, you made a very specific statement that dual CCDs suffer under CPU intensive scenes with RT enabled. I haven't seen any such testing or data for this and asked you for some sense you claimed it was well documented, but now are saying it hasn't been properly tested and there's not much data on it. So you came to a conclusion with extremely limited data. In fact, Carfax already posted some contradictory data to your statement. I would agree that the dual CCD configuration has a negative impact on some games and that this has been tested and there is data for this, but that is very different than your original claim.

I also don't need a reviewer to tell me that gaming results within a +- 1.3% spread are within margin of error.

 

[coercitiv]

Do we know when the locked chips are coming?

 

[nicalandia]

Do we know when the locked chips are coming?

I believe they will be announced on CES 2023 Jan 5th

 

[Hulk]

How many dies are there for RPL and what are the configurations?

 

[igor_kavinski]

Games with RT:

View attachment 73245

Games without RT:

View attachment 73246

5900X3D would have really helped AMD with more threads in RT games.

Best conclusion I can draw from these two graphs is that the best performance/$ CPU for gaming is i5-13600K.

 

[Hitman928]

5900X3D would have really helped AMD with more threads in RT games.

Best conclusion I can draw from these two graphs is that the best performance/$ CPU for gaming is i5-13600K.

I've mentioned this multiple times now, but computerbase's CPU results are not very representative of what most people will get from their systems because they stick strictly to officially supported memory speeds/timings. This has an especially negative effect for Ryzen CPUs. Unless you want to run 5200 MHz for Zen 4 and 5600 MHz for RPL memory with crap timings, computerbase's results aren't all that useful.

 

[Carfax83]

I've mentioned this multiple times now, but computerbase's CPU results are not very representative of what most people will get from their systems because they stick strictly to officially supported memory speeds/timings. This has an especially negative effect for Ryzen CPUs. Unless you want to run 5200 MHz for Zen 4 and 5600 MHz for RPL memory with crap timings, computerbase's results aren't all that useful.

One has to wonder, why Intel and AMD set their official memory standards so low relatively speaking to what the CPUs are actually capable of running. I haven't heard of a single case of anyone with a Zen 4 CPU being unable to run DDR5 6000 at a 1:1 ratio. And Raptor Lake's memory controller is far more robust than Alder Lake's and can hit much higher frequencies.

Seems to me that Alder Lake's stock memory should have been DDR5 5600 instead of DDR5 4800, and Raptor Lake's should have been DDR5 6400.

My 13900KF runs DDR5 7600 CL36 without even touching the system agent voltage.

 

[Hitman928]

One has to wonder, why Intel and AMD set their official memory standards so low relatively speaking to what the CPUs are actually capable of running. I haven't heard of a single case of anyone with a Zen 4 CPU being unable to run DDR5 6000 at a 1:1 ratio. And Raptor Lake's memory controller is far more robust than Alder Lake's and can hit much higher frequencies.

Seems to me that Alder Lake's stock memory should have been DDR5 5600 instead of DDR5 4800, and Raptor Lake's should have been DDR5 6400.

My 13900KF runs DDR5 7600 CL36 without even touching the system agent voltage.

I think it mainly comes down to not wanting to "officially" support speeds beyond the JEDEC standard and the timing of CPU development to that standard.

 

[Hulk]

With my new 280 AIO I am able to push more power without throttling so I could extend these charts a bit. Might have to take the rig into the garage over the break to take it to the end at 300+ Watts for completeness. Frequencies are rounded to the nearest 100MHz as they do jump around a bit during the run.

 

[igor_kavinski]

I think it mainly comes down to not wanting to "officially" support speeds beyond the JEDEC standard and the timing of CPU development to that standard.

Intel and AMD also want the memory makers to be able to sell their low quality crap dies. Can't have all that silicon going to waste. Also much easier for memory manufacturers to provide millions of DDR5-5600 RAM chips to OEMs like Dell and HP, rather than having to bin better quality DDR5-6400 and trying to repurpose the low quality ones to be used by bottom of the barrel brands.

 

[igor_kavinski]

Might have to take the rig into the garage over the break to take it to the end at 300+ Watts for completeness.

Why take that risk? Intel official TDP is 250W or something like that for RPL. Why go over the limit and risk damaging your CPU that you actually use for your important workloads?

 

[Hulk]

Why take that risk? Intel official TDP is 250W or something like that for RPL. Why go over the limit and risk damaging your CPU that you actually use for your important workloads?

I don't think there is much risk since I'm not boosting voltage or anything like that. It's really cold here in NJ so temps would be really low (safe).

 

[Carfax83]

Raptor Lake seems to be primarily constrained by power draw, which is why lowering voltages is one of the best ways to optimize as there is a lot of headroom and is especially useful for lowering temps. I recently increased my P core clock speed to 5.3ghz instead of 5.2ghz without changing the voltage at all (-125mV) but I did raise the PL1 and PL2 to 225w, up from 215w. Temps increased from 76c to 80c max, but clock speeds are more stable and don't fluctuate with the increased power draw.

May not sound that impressive, but 5.3ghz is just 200mhz off from the stock all core frequency which is 5.5ghz with significantly less power draw in an air cooled rig. If I put everything on auto including the clock speed, power draw and the voltage, temps skyrocket close to 100c and my rig becomes unstable under load. Stock voltage is also 1.35v or something at max load which is way more than my optimized voltages which is below 1.2v.

I think it's the high voltage that is driving those temps. If I had an AiO, I would probably increase the power draw to stock 253w and leave the clock speeds at default, but try to lower the voltages as much as possible.

 

[Carfax83]

Finally got into the 40K club for CBR23. Took 5.4ghz to get there, and 235w. Voltage is still the same, which furthers my earlier point that Raptor Lake is limited by power draw and not voltage.

*Edit* Ran HWinfo during multiple CBR23 runs and apparently I wasn't hitting 5.4ghz, much less sustaining it. It was mostly at 5.3ghz. Apparently more power is needed to run 5.4ghz all core boost but with my cooling it's not worth it.

5.3ghz is definitely the sweet spot for me.

 

[nicalandia]

Intel 13900E makes its debut in GB Compute.

Insyde AlderLake - Geekbench

Benchmark results for an Insyde AlderLake with an Intel Core i9-13900E processor.

browser.geekbench.com

 

[igor_kavinski]

Finally got into the 40K club for CBR23.

Great! Now you will be able to shave a few more seconds off your Handbrake video encoding

 

[nicalandia]

Impressive Result for The 13900K In NAMD Molecular Dynamics Simulations.

13900K vs 7950X vs 12900K, now the 13900K is using Windows and 7950X is under Ubuntu. So I also included a Windows vs Linux(Windows 11 vs Ubuntu) for the 12900K to see if CPUs performed better on Linux than on Windows.



13900K vs 7950X NAMD


There is no much difference between Linux and Windows 11

WSL2 Windows 11 Vs. Ubuntu Linux Benchmarks Performance - OpenBenchmarking.org

openbenchmarking.org

 

[lopri]

Finally got into the 40K club for CBR23. Took 5.4ghz to get there, and 235w. Voltage is still the same, which furthers my earlier point that Raptor Lake is limited by power draw and not voltage.

*Edit* Ran HWinfo during multiple CBR23 runs and apparently I wasn't hitting 5.4ghz, much less sustaining it. It was mostly at 5.3ghz. Apparently more power is needed to run 5.4ghz all core boost but with my cooling it's not worth it.

5.3ghz is definitely the sweet spot for me.

Is HWiNFO reporting 235W for a 40K run?

 

[Hulk]

Is HWiNFO reporting 235W for a 40K run?

I was wondering the same thing.

 

[Hulk]

Intel 13900E makes its debut in GB Compute.

View attachment 73411

Insyde AlderLake - Geekbench

Benchmark results for an Insyde AlderLake with an Intel Core i9-13900E processor.

browser.geekbench.com

What is a 13900E? Mobile part?

 

[igor_kavinski]

What is a 13900E? Mobile part?

Embedded