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.

 

[Exist50]

Using slower memory and using the tightest timings each CPU can get with the same memory isn't the same thing. I shouldn't have to point out something so obvious.

Why do you think those are the tightest timings each can achieve? You can buy plenty of XMP kits that do better.

Never claimed any such thing. You tend to try to put words in my mouth to discredit my arguments. Please stop.

Then give your memory criteria that doesn't involve maxing out AMD and crippling Intel. What's so magical about 6000 besides being AMD's "sweet spot"?

Claiming the same thing over and over without evidence doesn't make something true

In this case, it's true to anyone with eyes. I've had to spell out their own behavior for you, and you scour YouTube to find any crackpot with results you like.

What data do you have on Zen4 chips to make this statement?

Why don't you instead show me all these chips doing >3000 fabric?

 

[Hitman928]

It's not the only matchup with such gaps.

They basically have said ray tracing doesn't matter to justify excluding it from GPU coverage, claimed DDR5 didn't matter up until Raphael required it (and yet left DDR4 out of the price comparison I linked), etc.

If you're going to defend them, at least be familiar with their behavior.

And yet you ignored TPU's Zen 4 results entirely, and quote that hilarious TechGage graph with crippled RPL. What is that if not cherry picking?

You are completely arguing in bad faith and you know it. If you make an accusation of bias and incompetence, it is up to you to provide examples and proof. I am not supposed to be able to read your mind to determine what it was that they said that you found objectionable.

I also don't think they've ever claimed ray tracing doesn't matter. They have said that their audience doesn't care about ray tracing results as much (which is feedback they received by people who do in fact follow them and engage with them) and so they keep their focus on raster performance. They still include RT results for cards that support it, it is just not their focus because, again, no one has infinite time and so they partition their time according to what their audience is interested in. You may disagree with that, which is fine, but it is not evidence of bias or incompetence.

Also never heard them say DDR5 didn't matter. Now, I don't watch every second of every video so I don't know how to respond to that without a source for the comments. I also tried searching for your price comparison post but couldn't find it and asked if you could link me to it but you haven't done so, so I can't really comment on that either.

I never ignored TPU's results. I'm also assuming you mean TechDeals. If that's the case, then no, RPL is not crippled.

 

[Markfw]

@Exist50 , @Hitman928

Just an observation, don't put me in your arguments.

I agree with Hitman on the games outlier. I see that Exist50 will argue things forever, and never give in and admit he is wrong, he just changes the goalposts, like now its about memory. See @Det0x for AMD timings. Yes, Intel can get higher, but gear 2 has a heavy price, I know, I used to have an alder lake.

I will see myself out of this argument now......

 

[Exist50]

If you make an accusation of bias and incompetence, it is up to you to provide examples and proof.

As I've done so many times, and yet you outright ignore it.

I also tried searching for your price comparison post but couldn't find it and asked if you could link me to it but you haven't done so, so I can't really comment on that either.

#4037

Don't accuse someone of bad faith if you literally won't read their posts.

I never ignored TPU's results. I'm also assuming you mean TechDeals. If that's the case, then no, RPL is not crippled.

For TechDeals, I pointed out that no rational person would consider 6000 CL30 AMD vs 6000 CL32 Intel to be a fair comparison. You simply insist otherwise. Either you haven't been following any other reviews, and thus don't know what's typical, or you're just deliberately trolling.

As for TPU, you claimed that HWUB's Zen 4 bias was actually just memory speeds. When I pointed out that TPU's data shows otherwise, you've yet to actually acknowledge it.

 

[Exist50]

@Exist50 , @Hitman928

Just an observation, don't put me in your arguments.

I agree with Hitman on the games outlier. I see that Exist50 will argue things forever, and never give in and admit he is wrong, he just changes the goalposts, like now its about memory. See @Det0x for AMD timings. Yes, Intel can get higher, but gear 2 has a heavy price, I know, I used to have an alder lake.

I will see myself out of this argument now......

As you've called people trolls for merely reading reviews or quoting official AMD specs, forgive me for not considering your opinion in good faith.

 

[Hitman928]

Why do you think those are the tightest timings each can achieve? You can buy plenty of XMP kits that do better.

They are the tightest timings they could get, with that memory. That doesn't mean that's the best timings Intel can do with different memory.

Then give your memory criteria that doesn't involve maxing out AMD and crippling Intel. What's so magical about 6000 besides being AMD's "sweet spot"?

My guess as to why many reviewers use it is because it's relatively affordable and likely to be used by many people. It's very easy to use for both platforms with good results, going faster memory takes additional time and testing, at least on AMD, and from the tests I've seen, RPL doesn't get any benefit from going faster anyway. Seems like a very logical and reasonable speed to use for both, but if a reviewer wants to spend the extra money and get a faster kit for Intel, I wouldn't be opposed to that at all. But that doesn't mean anything less than DDR5-7200 memory with RPL immediately makes the results invalid and shows the reviewer is biased or incompetent.

Why don't you instead show me all these chips doing >3000 fabric?

You are the one who claimed the majority of Zen 4 CPUs can't go above DDR5-6000 in gear 2. It is up to you to support your claim. With that said, our own forum member, detox, has been able to push above 6000 speeds with good results. HWUB also has results up to DDR5-6400. I'm guessing you consider them to be too biased for you to care though.

In this case, it's true to anyone with eyes. I've had to spell out their own behavior for you, and you scour YouTube to find any crackpot with results you like.

Now you've really just devolved into nothing but ad hominems and straw mans. I won't waste any more time or clutter up this thread any further. Best of luck on your future endeavors.

 

[Exist50]

They are the tightest timings they could get, with that memory.

How do you know that to be the best achievable?

My guess as to why many reviewers use it is because it's relatively affordable and likely to be used by many people

5600 is even more available, nor is 6400 or 6800 hard with Hynix. So why specifically 6000?

You are the one who claimed the majority of Zen 4 CPUs can't go above DDR5-6000 in gear 2.

I was actually being generous. Without overclocking (i.e. not just EXPO), AMD will not do >6000 in gear 2.

 

[Det0x]

For the majority of Raphael chips, you can't go past 6000 in gear 2 (i.e. with a benefit), but Raptor Lake can do that easily.

Then give your memory criteria that doesn't involve maxing out AMD and crippling Intel. What's so magical about 6000 besides being AMD's "sweet spot"?
Why don't you instead show me all these chips doing >3000 fabric?

5600 is even more available, nor is 6400 or 6800 hard with Hynix. So why specifically 6000?
I was actually being generous. Without overclocking (i.e. not just EXPO), AMD will not do >6000 in gear 2.

Why are you talking about gears and above 3000 FCLK ?
As you clearly lack the knowledge/understanding of the z4 memory system, i suggest you stick to making outlandish claims gen13 gaming performance

 

[Exist50]

As you clearly lack the knowledge/understanding of the z4 memory system, i suggest you stick to making outlandish claims gen13 gaming performance

Well if Raphael is so much more capable, then feel free to post a comparison with both using 6800 or 7200. We both know why you won't.

 

[Det0x]

Well if Raphael is so much more capable, then feel free to post a comparison with both using 6800 or 7200. We both know why you won't.

Comparison in what, for what reason ?
It seems to me you're just moving the goalpost again

 

[Carfax83]

But one of your main arguments before for using computerbase is that they clearly test properly for CPU bottlenecks in games by picking the right settings and locations. So do you believe that they do this or not? Also, computerbase showed lower fps for the 13900k at 720p than what is in the video at 1080p + DLSS quality So if we're just going to go off of uncontrolled comparisons and a bunch of assumptions, it seems computerbase's situation is actually causing more of a CPU bottleneck than the videos you showed.

Well we don't know what the GPU usage was for Computerbase.de's test or where it was done. But you can clearly see in the YouTube video that the Raptor lake rig was CPU limited throughout the entire run. 1080p DLSS quality corresponds to an input resolution of 720p. Average GPU usage was in the upper 70s, but fell into the 60s during the most demanding area despite the high framerates.

Also, the Raptor Lake video had a tweaked setup with an overclocked CPU, tricked out DDR4 and an overclocked RTX 4090.....and it was still CPU bottlenecked.

Edit: I also forgot to mention that in the 2 videos you posted as "proof" there is a clear increase in characters on screen for the AMD machine. I have no idea why, but it is obvious if you pay attention to it in the videos (examples below, one is really blurry because the uploader's video quality is terrible).

Well the reason is because the crowd moves around and isn't stationary for the most part. If you watch the entire video, you'll see that the crowd density is equivalent between them; particularly in the area where the AMD rig turns around and goes back the other way (the most CPU demanding area).

 

[Exist50]

Comparison in what, for what reason ?
It seems to me you're just moving the goalpost again

You were literally just saying I don't understand Zen4's memory limits, including quoting a post where I highlighted why no one goes past 6000. I thus invited you to prove me wrong, but I predictably got this attempt to deflect.

So go on, prove me wrong if you can. We both know why you're incapable of doing anything more than handwaving.

 

[Hulk]

I don't understand why anybody would buy the 13700 for $400 over the 13700K for $380, which is how they are priced at Microcenter. And at least at my Microcenter they have 25+ of the K's and 7 of the non K's, which would seem to indicate the non K's are selling.

Also the 13900K is $10 more than the 13900 yet there are less 13900's in stock.

It's just nuts! I guess people don't know that the K's do everything the non K's do but at higher clocks and they are better binned parts? Even if you don't have a Z mobo the K's are clocked higher stock.

It just seems as though if you are buying a CPU you must have researched a little bit right? Just checked out the specs at least?

Or am I missing something that makes the non K's desirable?

 

[Hulk]

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.

Good points. But actually the E's are already a thorn in Intel's side. Without them, on a 10nm monolithic die only about 12 P's would fit and that CPU would have unconditionally yielded the high end to AMD regardless of how many watts Intel forced down it's throat! AMD would be sitting all alone at the top with a $700 top of the line part while Intel would have been under the table looking for scraps.

Now in reality we know that gaming benchmarks would have been about the same and even many applications, but heavily threaded applications would have been decisively AMD dominated if Intel had no implemented 8+16 when they did.

The bottom line fact is that there aren't a lot of applications that use more than 8 cores but less than 16, which AMD currently has an edge. So if for example Intel simply went 8+32 they would be at nearly 60,000 CB R23 points with no architectural changes. Even if they weren't going tiles they could most likely fit that on Intel 4. The AREA efficiency of Gracemont can't be underestimated.

Anyway it's a good race of it for Intel and AMD currently and good (and fun) for us to be able enjoy the parts and the discussion surrounding them. If Intel should stumble next round I'll go AMD. Since I had already gone with Alder Lake it was easy for me to pop in the 13900K so that's what I did. But I'm totally open to giving AMD a shot next time if they put up another outstanding lineup like they have done for the last 3 generations.

 

[Carfax83]

And the circle jerk continues........

I didn't say that RT is an edge case scenario. Playing video games on a 4090 with 'officially supported' mem speed/JEDEC timings at 720p is an edge case irrespective of whether or not RT is enabled. 1080p gaming on a 4090 is arguably an edge case.

Yes it's not realistic, but it's a testing environment designed to highlight CPU performance. The whole point is to isolate CPU performance as much as possible, and to do so you have to remove the GPU as a potential bottleneck.

So in other words you are speculating that they aren't testing in adequately CPU demanding areas of these games. Why is this criticism only leveled at HUB?

I'm not speculating. I own Plague Tale Requiem so I know which areas are CPU heavy, and the HZD built in benchmark explains why AMD has such a large lead.

I don't think the built in benchmark loads the A.I routines, as they're primarily for determining GPU performance and not CPU.

Since they don't provide data for the same games with and without RT we can't isolate RT as a variable. The pcgameshardware.de testing is a much better example of how testing at officially supported memory speeds biases the data against Zen 4.

Raptor Lake is also running official memory speeds as well, and can memory at much higher speeds than Zen 4.

Lets compare Plague Tale Requiem data between HUB and pcgameshardware.de.

PCGH.de normally does CPU tests at 720p, and they probably tested a CPU heavy area like when the rats show up or in the town in the later part of the game.

The early town in the beginning part of the game wasn't very CPU intensive from my recollection, which is why the scores are bunched up.

 

[igor_kavinski]

Or am I missing something that makes the non K's desirable?

Customer: I'm looking for an i7.
Employee: We have the i7-13700 and the i7-13700K.
Customer: What's the difference?
Employee: The K version is good for games and applications requiring high performance.
Customer: I just need to run Microsoft Office.
Employee: The 13700 would be an excellent choice for you.
Customer: Great! I'll take it!

Other reasons:

13700 comes with the official Intel cooler(!).
It runs cooler and consumes less power.
You don't need to worry about tinkering with it to extract more performance. You save time.
It. Just. Works.

When comparing prices for custom builds, the one with 13700 comes out slightly cheaper, due to needing a less powerful PSU, less ventilated case and no extra expense for the heatsink or liquid AIO.

 

[Det0x]

You were literally just saying I don't understand Zen4's memory limits, including quoting a post where I highlighted why no one goes past 6000. I thus invited you to prove me wrong, but I predictably got this attempt to deflect.

So go on, prove me wrong if you can. We both know why you're incapable of doing anything more than handwaving.

I'm still not sure what exactly you are asking for, proof of Zen4 running memory speeds above 6000MT/s ?
Here are a few memory benchmark ive ran on my own system, feel free to match them or whatnot your looking for here.


Z4 @ 7000MT/s (not my screenshot, just to show its possible)

 

[DrMrLordX]

You were literally just saying I don't understand Zen4's memory limits, including quoting a post where I highlighted why no one goes past 6000.

His system is at something like 6500 MT/s, what're you on about? In any case it's the bandwidth and latency that matters in the end. If Raphael is putting up good numbers within the limits of its IMC, then bully for them.

 

[Carfax83]

I just calculated it out and Eurogamer (which I guess is run by digital foundry?) actually has the 7950x as 2.4% faster on average across their game tests than the 13900k when both are using DDR6000 memory. So there's that. Keep in mind that it's only 8 games though.

The Eurogamer Digital Foundry test used an RTX 3090, which means they were GPU bottlenecked. That's why the gap between both CPUs lessened when run with DDR5 6000.

This can be mitigated somewhat by testing at 720p, but it appears the lowest resolution they used was 1080p.

The most accurate tests will always use an RTX 4090. These CPUs are so fast that it's very easy to bump into a GPU bottleneck with them, especially if the game is well optimized for CPU utilization.

 

[Elfear]

Then give your memory criteria that doesn't involve maxing out AMD and crippling Intel. What's so magical about 6000 besides being AMD's "sweet spot"?

They were both run at DDR5-6000 in the Tech Deals review and the AMD was run at CL30 and the Intel at CL32. You are using rather inflammatory language ("crippling Intel") for a difference that is miniscule.

13900k at DDR5-6000 (CL36) vs DDR5-7400 (CL34). Source <1% difference...

 

[Exist50]

I'm still not sure what exactly you are asking for, proof of Zen4 running memory speeds above 6000MT/s ?

With improved gaming numbers. That's the only workload discussed so far.

If Raphael is putting up good numbers within the limits of its IMC, then bully for them.

Sure, but I'm responding to someone claiming that despite all evidence (e.g. TPU's review), the only reason other reviews don't look like HWUB's is JEDEC memory. Moreover, that user was claiming that Raptor Lake can't even match Raphael's memory support, which is absurd.

They were both run at DDR5-6000 in the Tech Deals review and the AMD was run at CL30 and the Intel at CL32. You are using rather inflammatory language ("crippling Intel") for a difference that is miniscule.

The person I was responding to disputes those very numbers from TPU. My argument is that if you're going to blame memory for huge performance discrepancies, you can't just turn around and push reviews with some objectively poor methodology. After all, if it makes no difference, why the different setups to begin with?

 

[Det0x]

With improved gaming numbers. That's the only workload discussed so far.

k, here is one of my game benchmark that scales extremely well with memory performance.
Average CPU game numbers are used for benchmark comptition threads both at overclock.net and techpowerup.net forums

 

[Exist50]

k, here is one of my game benchmark that scales extremely well with memory performance.
Average CPU game numbers are used for benchmark comptition threads both at overclock.net and techpowerup.net forums
View attachment 74640

Ok, do you have a comparison point with the same everything else, minus DRAM speed? And a comparison to Intel at different DRAM speeds?

Also, correct me if I'm wrong, but that looks like a CPU overclock, not just EXPO/memory. Part of the premise here was that testing with >stock DRAM speeds doesn't really count as overclocking. I tend to agree with that take, but anything more than just toggling memory speeds should be explicitly an overclocked vs overclocked comparison, none of which applies to the discussion above.

 

[Mopetar]

I don't understand why anybody would buy the 13700 for $400 over the 13700K for $380, which is how they are priced at Microcenter. And at least at my Microcenter they have 25+ of the K's and 7 of the non K's, which would seem to indicate the non K's are selling.

I'm sure it seems odd to people who post on tech forums, but most people aren't going to overclock their CPU, so the K is an extra $20 for a feature that they won't use.

Frankly most people would be better off dropping the power settings regardless of whether they have a new Intel or AMD CPU. Both are using a lot of extra power for minimal performance gain.

 

[Hulk]

Customer: I'm looking for an i7.
Employee: We have the i7-13700 and the i7-13700K.
Customer: What's the difference?
Employee: The K version is good for games and applications requiring high performance.
Customer: I just need to run Microsoft Office.
Employee: The 13700 would be an excellent choice for you.
Customer: Great! I'll take it!

Other reasons:

13700 comes with the official Intel cooler(!).
It runs cooler and consumes less power.
You don't need to worry about tinkering with it to extract more performance. You save time.
It. Just. Works.

When comparing prices for custom builds, the one with 13700 comes out slightly cheaper, due to needing a less powerful PSU, less ventilated case and no extra expense for the heatsink or liquid AIO.

Good points. Or they could just set the same power level as the non K version and get better performance due to the better binned part. There are lots of good coolers that are better than stock Intel on the cheap. But yeah I get it. Those people don't want to deal with all that!