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.

 

[Markfw]

Funny that we know more about Raptor Lake than about Zen4.

Actually, all we have is smoke and mirrors for Raptor lake. No REAL benchmarks, just a lot of leaks and wishing. I said it before, and I will say it again, I will wait until we have real benchmarks to compare, except that Raptorlake will use more power than Zen 4.

 

[inf64]

Both are small upgrades. The noteworthy thing here is probably that Raptor Lake fixes some low hanging issues with the hybrid architecture (hopefully ring bus not being down-clocked when E cores enabled, better voltage regulation, more L2 cache to offset the poor L3 performance relative to Zen 3, more E cores hopefully allowing lower E core clocks and better PPW), and AM5 brings to AMD cpus DDR5 and PCIE5. Performance wise both are boring.

Actually Zen3->Zen4 should be a bigger upgrade than Zen2->Zen3. We are talking about aggregate performance uplifts and Zen4 will bring more ST and more MT gains than what Zen3 brought.

Zen3 brought 24% higher ST performance and 15% higher MT performance (vs 3950X) : source

If Zen4 clocks ~5.7-5.8Ghz for ST (which I believe it will), with ~10% IPC uplift it will perform roughly ~30% better than 5950X (better than Zen2->Zen3).
If Zen4 clocks at ~5.3Ghz in MT workloads (which I believe it will), with ~10% IPC uplift it will perform ~38% better than 5950 (much bigger uplift than Zen2->Zen3).

 

[pakotlar]

RPL have more cache yes, but Zen 4 too, and every architecture have different sweet spot, 2mb l2 of RPL doesn't mean it is better than 1mb l2 of Zen 4, they are 2 very different architectures, but what is important that they both have increased cache, + Zen 4 will have 3d versions with more l3 combined with l2 so i'm curious how will that translate to gaming + also ddr5.

Yeah, all true; will be interesting, though I’m not comparing 13900K L2 to 7950X, rather 13900K to 12900K; 13900K doubles L2 cache per core, what they call Game Cache. I’m curious about your last point: in particular, whether Zen 4 IMC (DDR5 only) will be better than Intel’s (DDR4 + 5 and may prioritise compatibility with DDR5 over sheer performance).

I’m also really hoping that competition with Intel will spur AMD to release higher core count AM5 parts, say 24 core to get the PR parity with Intel’s 24 cores in 13900K. Probably will have to go for Genoa though, which I don’t want to do for cost and clockspeed reasons. Probably won’t get my wish in 2023, but maybe 16 core Zen4 3D will be attractive enough (and with AVX512 may be extra nice).

 

[pakotlar]

Actually, all we have is smoke and mirrors for Raptor lake. No REAL benchmarks, just a lot of leaks and wishing. I said it before, and I will say it again, I will wait until we have real benchmarks to compare, except that Raptorlake will use more power than Zen 4.

We unambiguously have more info on 13900K than 7950X. There are ES1 and ES3 parts in the wild, with an ES1 actually sold and in some random person’s hands.

 

[Markfw]

We unambiguously have more info on 13900K than 7950X. There are ES1 and ES3 parts in the wild, with an ES1 actually sold and in some random person’s hands.

So ES parts, and you trust their facts. Read what I said, I don't believe it. Some truth (the power usage) but all the rest, we have no real facts.

 

[pakotlar]

Actually Zen3->Zen4 should be a bigger upgrade than Zen2->Zen3. We are talking about aggregate performance uplifts and Zen4 will bring more ST and more MT gains than what Zen3 brought.

Zen3 brought 24% higher ST performance and 15% higher MT performance (vs 3950X) : source

If Zen4 clocks ~5.7-5.8Ghz for ST (which I believe it will), with ~10% IPC uplift it will perform roughly ~30% better than 5950X (better than Zen2->Zen3).
If Zen4 clocks at ~5.3Ghz in MT workloads (which I believe it will), with ~10% IPC uplift it will perform ~38% better than 5950 (much bigger uplift than Zen2->Zen3).

AMD quoted 15% ST uplift for Zen 4 ( likely less compared to Zen3D), so smaller than the 24% for Zen 3 vs Zen 2, but even if that were not the case (say AMD is sandbagging) Zen 3’s real advantage was perceptual; it finally beat Intel ST performance, and did so at lower power levels, so there was no longer any reason to go with Intel.

Zen 4 brings no such seismic change; in fact it may give quite small ST benefits, and lose some of its power efficiency advantage (PPT going up 80%, maybe). Anyway, it will still be a great part, just not as important as Zen 3 in terms of non-vector compute core performance; all the stuff surrounding it though (AVX512, DDR5, PCIE5) is important, and AM5, if supported for as long as AM4 is probably the platform to invest in. I would have a hard time jumping on Raptor Lake even if it were much faster (which it won’t be), because that is the last upgrade possible. Intel really needs to support their mobo chipsets for more than 1 tick + 1 tock.

 

[inf64]

AMD quoted 15% ST uplift, but Zen 3’s real advantage was mostly perceptual; it finally beat Intel ST performance, and did so at lower power levels, so there was no longer any reason to go with Intel. Zen 4 brings no such seismic change; in fact it may give quite small ST benefits, and lose some of its power efficiency advantage (PLT going up 80%, maybe). Anyway, it will still be a great part, just not as important as Zen 3 in terms of non-vector compute core performance; all the stuff surrounding it though (AVX512, DDR5, PCIE5) is important, and AM5, if supported for as long as AM4 is probably the platform to invest in. Hwould have a hard time jumping on Raptor Lake even if it were much faster (which it won’t be), because that is the last upgrade possible. Intel really needs to support their mobo chipsets for more than 1 tick + 1 tock.

Could you tell me, with which part of my post you are disagreeing? What part of 10% IPC x 5.7/4.9 (ST clocks) is not clear enough?

 

[pakotlar]

So ES parts, and you trust their facts. Read what I said, I don't believe it. Some truth (the power usage) but all the rest, we have no real facts.

Yeah definitely not final figures. I think those ES figures undersell 13900K. We have more info on Raptor Lake than Zen 4, but no more production part results, agree with you.

 

[pakotlar]

Could you tell me, with which part of my post you are disagreeing? What part of 10% IPC x 5.7/4.9 (ST clocks) is not clear enough?

Sure, I disagreed in that the reason that Zen 3 was so important was not so much because of the strict % uplift, but because it completely invalidated Intel CPUs for the first time in 15 years, by erasing Intel’s ST advantage, while maintaining a strong MT advantage. I also pointed out that you quoted a 24% ST uplift for Zen 3->2, which is larger than the purported Zen 4->Zen 3, though admittedly AMD could easily be setting expectations low, as they have in the past.

I do agree that on the basis of MT uplift Zen 4 will probably be better than Zen 3, if you ignore gaming and Zen3D. It also brings AVX512 which may be nice for some stuff I do.

 

[inf64]

Sure, I disagreed in that the reason that Zen 3 was so important was not so much because of the strict % uplift, but because it completely invalidated Intel CPUs for the first time in 15 years, by erasing Intel’s ST advantage, while maintaining a strong MT advantage. I also pointed out that you quoted a 24% ST uplift for Zen 3->2, which is larger than the purported Zen 4->Zen 3, though admittedly AMD could easily be setting expectations low, as they have in the past.

I do agree that on the basis of MT uplift Zen 4 will probably be better than Zen 3, if you ignore gaming and Zen3D. It also brings AVX512 which may be nice for some stuff I do.

Well for Zen3 we have benchmarks that show 24% is the end result going from Zen2. For Zen4 we only have >15%, hence the math experiment I did.

AMD demoed Zen4 ES running at 5.5Ghz on some threads in a gaming workload. I bet they can hit 5.7Ghz fairly easily on one threaded workloads. Couple that with their own quoted 10% for IPC uplift and you get to ~30% higher ST uplift versus Zen3. This would put top Zen4 (no Vcache) neck at neck with 13900K.

If you add Vcache, then I guess they might lose some clocks but the gaming will be a clean sweep with massive (>20% advantage) over 13900K.

 

[eek2121]

I suspect we are going to see an influx in new forum accounts and discontinuation of old forum accounts when Zen 4 and Raptor Lake finally drop.

 

[pakotlar]

Well for Zen3 we have benchmarks that show 24% is the end result going from Zen2. For Zen4 we only have >15%, hence the math experiment I did.

AMD demoed Zen4 ES running at 5.5Ghz on some threads in a gaming workload. I bet they can hit 5.7Ghz fairly easily on one threaded workloads. Couple that with their own quoted 10% for IPC uplift and you get to ~30% higher ST uplift versus Zen3. This would put top Zen4 (no Vcache) neck at neck with 13900K.

If you add Vcache, then I guess they might lose some clocks but the gaming will be a clean sweep with massive (>20% advantage) over 13900K.

Yeah, hopefully you're right. 7950X + Vcache would be very, very nice, not only for gaming, but some other workloads as well. Hopefully doesn't come at the loss of clock speed; the idealist in me think that 5800X 3D clocks were low due to power limits, and that Zen 4 won't need the full 230W PPT until Zen 4 + Vcache, so 7950X 3D would keep those 5.5+ Ghz clocks. I would buy that along with some tasty 6400+ MT DDR5 memory, say this stuff: https://www.tweaktown.com/news/8213...ction-7200-memory-up-to-768gb-kits/index.html. If clocks drop a bunch and we lose overclocking support like in Zen 3D, I probably won't be excited, at all, because gaming performance is already fine with my 5950X.

 

[Abwx]

If Zen4 clocks at ~5.3Ghz in MT workloads (which I believe it will), with ~10% IPC uplift it will perform ~38% better than 5950 (much bigger uplift than Zen2->Zen3).

I wouldnt count on a 5.3GHz all 16 cores, to be within the >35% MT uplift 5GHz is enough, so that should land by there.

 

[pakotlar]

I wouldnt count on a 5.3GHz all 16 cores, to be within the >35% MT uplift 5GHz is enough, so that should land by there.

In any case the performance uplift this upcoming generation isn’t impressive outside of server parts, going by rumors. Much better than we had in Intel dominating days, but I’ve gotten spoiled by AMD in recent years and want more high-IPC cores in consumer space. There’s only so much they can do keeping core counts low, or using efficiency cores with circa Skylake IPC.

 

[carrotmania]

AMD quoted 15% ST uplift for Zen 4...

No... they didn't...

 

[pakotlar]

No... they didn't...

https://www.tomshardware.com/news/amd-intros-zen-4-ryzen-7000-cpus-and-600-series-chipset-up-to-55-ghz-15-performance-rdna-2-igpus-pcie-5-ddr5

May want to read that.

 

[carrotmania]

https://www.tomshardware.com/news/amd-intros-zen-4-ryzen-7000-cpus-and-600-series-chipset-up-to-55-ghz-15-performance-rdna-2-igpus-pcie-5-ddr5

May want to read that.

Read the bit AMD themselves state, ie. the image. See if you can tell where you've gone wrong...

 

[pakotlar]

Read the bit AMD themselves state, ie. the image. See if you can tell where you've gone wrong...

You can pretend to know the uplift, but until we have benchmarks, I’ll read “>15%” as 15% average, either mode or mean, with outliers that are a lot higher and a lot lower. If you think that’s a min, ok.

 

[pakotlar]

Not sure here... While its nice to get more MT performance, it all depends on the task. Case in point (and I am sure there are other real world examples) when I do DC work, it creates 32 threads (32 individual tasks working on 32 different problems) for a 5950x. if 16 of those were gracemont, it would not do as well. Also, there is scheduling, which I am sure will not get totally worked out for years.

I think that there will be a lot of changes in both camps, and alder lake is certainly a nice step forward, I am just not convinced its the future. CB23 is one application, you need to compare a lot more applications to evaluate.

What do you think of Apple’s approach? They seem to be pushing efficiency cores for“pro” uses, but throw a bunch of accelerators in for vector and matrix workloads, and their e-cores are fairly close in performance to their p-cores IIRC. I’m pretty leery of efficiency cores in general, unless the density of ecores is so much higher that it overwhelms the IPC decrease, but people seem to pretty happy with M1 Pro/Max.

 

[Markfw]

What do you think of Apple’s approach? They seem to be pushing efficiency cores for“pro” uses, but throw a bunch of accelerators in for vector and matrix workloads, and their e-cores are fairly close in performance to their p-cores IIRC. I’m pretty leery of efficiency cores in general, unless the density of ecores is so much higher that it overwhelms the IPC decrease, but people seem to pretty happy with M1 Pro/Max.

Until apple hits "my world" which includes windows and linux, I don't care to compare. They seems to be 2 different worlds right now. At least for me.

 

[cortexa99]

Actually Zen3->Zen4 should be a bigger upgrade than Zen2->Zen3. We are talking about aggregate performance uplifts and Zen4 will bring more ST and more MT gains than what Zen3 brought.

Zen3 brought 24% higher ST performance and 15% higher MT performance (vs 3950X) : source

If Zen4 clocks ~5.7-5.8Ghz for ST (which I believe it will), with ~10% IPC uplift it will perform roughly ~30% better than 5950X (better than Zen2->Zen3).
If Zen4 clocks at ~5.3Ghz in MT workloads (which I believe it will), with ~10% IPC uplift it will perform ~38% better than 5950 (much bigger uplift than Zen2->Zen3).

Some dudes said 7600X ≈ 5800X, no mention of clocks and power. You can re-calculate it. but since this is intel thread I won't talk too much. sound like the impact is tremendous. Also rumor.

 

[pakotlar]

Until apple hits "my world" which includes windows and linux, I don't care to compare. They seems to be 2 different worlds right now. At least for me.

Same

 

[Hulk]

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.

You do have a point. But even forgetting the E's. Non-HT Coffee Lake to HT Golden Cove is a pretty significant upgrade.

 

[Thunder 57]

Well this thread (de?)escalated quickly? What some people seem to be forgetting is that Intel themselves claim "up to a double digit performance increase". Why are some shocked that Raptor Lake shows almost no IPC boost while the rest seems to be gained by frequency?

I just want these chips to launch already so we can still argue needlessly but at least have real numbers.

 

[Markfw]

You do have a point. But even forgetting the E's. Non-HT Coffee Lake to HT Golden Cove is a pretty significant upgrade.

I certainly agree that the golden cove cores are quite powerful. But not good at the electrical power they use.