Talk:Raptor Lake

"Intel 7"

The body of the article correctly points the Intel 7 link to 10 nm process, but the infobox link points to 7 nm process, which is incorrect per the references associated with the link in the body of the article. 2601:1C0:6F00:D035:F013:8211:C195:111C (talk) 16:41, 26 September 2022 (UTC)

Performance data

On average 15% faster in ST, 41% in MT workloads. Artem S. Tashkinov (talk) 22:09, 26 September 2022 (UTC)

References

13900K perf = 12900K @ 25% TDP

I raised this issue at User_talk:Artem_S._Tashkinov who contributed this claim, but further discussion, if any, should better be here.

Hi, I have tagged an unsourced statement which I traced to your edit and want to ask you about the source as well why you contribute unsourced material, which obviously requires a citation, in clear violation of WP:V – something an editor with 16 years of experience should be well aware of? 188.66.33.68 (talk) 14:13, 25 November 2022 (UTC)

"Violation", really? You really could use a stronger wording to crap on me as if I made up that bloody statement. I've provided the source which is already on the page in too many places, hope you're happy. Artem S. Tashkinov (talk) 14:10, 26 November 2022 (UTC)

Thanks for the source, I examined it and I'm going to remove this statement from the article as it's misleading: from the slide it's clear this "equal perf @ 25% tdp" is not proper comparison of technology, but total and utter marketing, which doesn't belong in Wikipedia – feel free to check the article's talk page for a more detailed explanation which I will post there shortly.

And, by the way, why such tone? You're surely well aware of what needs to be supported with a citation and should have done so in the first place, so it's only fair to ask for the source.188.66.35.247 (talk) 19:08, 27 November 2022 (UTC)

Despite rather uncivil comments, I'll make some remarks regarding this statement by Intel for anyone interested. I personally don't put in question Intel's claim per se, it may well be true, the point is a bit more subtle than that. It's not a very good idea in general to use vendor's marketing materials in WP as they tend to show products in the best – rather than neutral – light, and WP:NPOV remains one of the core policies, and cases like this one is exactly why secondary sources are preferred. I see two problems with this slide by Intel when it comes to citability in WP:
a) they use not apps (prebuilt binaries) but SPEC scores (compiled source), and they don't disclose any details – do they use identical compilers and switches or are they different? Their claims can't even be reproduced and verified independently and
b) for some reason they don't show 12900K's power-performance numbers alongside 13900K's. Guess why? Guess because 12900K's power-perf curve is in fact similar to 13900K's and power efficiency gains of the former at reduced voltage and clock are as impressive. Or perhaps more impressive – or why would Intel shy away from transparent power-perf vs power-perf comparison otherwise?

In fact, given the clear uplift in clock rate with essentially unchanged uarch, it's entirely possible that for Raptor Lake, or at least RPL desktop SKUs, Intel tuned the process and/or physical design towards higher performance, *not* higher power efficiency, and/or employs tighter binning at highest performance for RPL, so it's a big question, in fact, how well 13900K fares at scaling down compared to 12900K in a proper apples-to-apples comparison as well as at power efficiency vs 12900K. I wouldn't be too surprized to see it perform worse, at least on a number of workloads – opposite of the impression this slide creates.

Some info on how 13900K's power efficiency relates to that of 12900K is just fine for this article, but we clearly need a better source than this marketing presentation slide. Time permitting, I'll try to find something worth mentioning here. 188.66.35.247 (talk) 19:28, 27 November 2022 (UTC)

Not going to argue that marketing slides shouldn't be used as citations for WP but multiple reviews show that power limiting RPL CPUs is enough to make them very power efficient and a lot more so than ADL, so Intel did something to optimize power consumption. Of course, it only applies to the situations when the user actually bothers with changing PL1/PL2 limits in the BIOS which is not something most people will ever do, so I will agree that the point was worth removing. I've replaced it with something more subtle. Sorry for the tone of my message but I perceived your edit on my talk page as overtly aggressive. Artem S. Tashkinov (talk) 07:17, 28 November 2022 (UTC)

Major changes in february 2023

Let's discuss everything first and invite other editors.

• Release dates: Wikipedia has a WP:CBALL policy thus tables only contain released or soon to be released products. There's no need to specify the release dates as all the CPUs have long been available. This makes tables even wider for no reasons. Then what's the encyclopedic value in release dates exactly? Besides, the article already contains a ton of historical information.
• RAM support. For desktop CPUs it's already in the bullet list before the table. For laptop CPUs: laptops come with RAM preinstalled or in a lot of cases it's even soldered. You buy based off OEM specs, not based on the information from Intel. Again, no point in including duplicated data.
• L2 cache: it's per CPU (or per 4 CPUs for E-cores), it's not shared. Listing the L2 cache for all CPUs simultaneously is either misleading or outright wrong. Intel lists Smart Cache for a reason.

I'm all for minor fixes when the information is wrong. Please abstain from including everything just because you feel like it. Artem S. Tashkinov (talk) 23:32, 5 February 2023 (UTC)

Honestly I think that having the release date column makes more sense from an encyclopedic perspective—gives a precise timeline for when the products were released, and what time they came to compete with products from other companies (e.g. did this CPU come before or after that model of competing AMD CPU, and by how many months away).

But I agree with the other points. When a piece of data is exactly the same for all 25 CPU models, there's no need to have it in the table taking up so much space and creating lots of blank space (an eyesore), when it can be easily presented in the simple form of a bulletpoint above the table. Unless it's needed for consistency with other tables in the same article where data varies.

AP 499D25 (talk) 13:51, 6 February 2023 (UTC)

Speaking of "precise timeline for when the products were released, and what time they came to compete with products from other companies" - in the article intro all this information is already available, as well as in the informational box on the right - i.e. already two places. Is this really necessary to have it in multiple different places? :-) Secondly, WP is normally used by people to get valuable information and that's the product features, not when it was released exactly since WP normally lists already released products.

Would be great if WP had collapsible rows/columns, alas, that's not the case. In this case I guess we could include everything people desire. Artem S. Tashkinov (talk) 15:00, 6 February 2023 (UTC)

In response to "in the article intro all this information is already available, as well as in the informational box on the right": Sometimes it's good to duplicate information for better/easier accessibility of it, so that the reader doesn't have to scroll up and scan throughout the article to find it. My little recommendation, on the most common desktop resolution for websites, which as of now is 1920x1080 (source), if I have to scroll up a bit to find a specific piece of information, that info ought to be repeated down the article. You see I didn't even notice the release dates in the intro until you mentioned them at all, since I was focused on the tables of the articles anyways. What definitely doesn't make sense here is repeating the cache sizes and memory support in the tables since they are both in the bulletpoint list above the table, which takes barely any scrolling to locate.

"Would be great if WP had collapsible rows/columns, alas, that's not the case." - I agree. If such a feature existed, I would have put all the less important info (the release dates, prices) inside the collapsible columns.

AP 499D25 (talk) 07:49, 12 February 2023 (UTC)

P.s. : it was 176.9.113.53 who added all the additional columns to the table, while @Visite fortuitement prolongée wanted to note that dual-channel DDR5 is considered "quad-channel", as each RAM module has two 32-bit memory buses (whereas before with DDR4 it's single 64-bit bus).

It's worth noting pinging other users doesn't work in the title. To notify IP editors and bring them in, you'll need to leave a note on their user talk page.

On regard of memory support, in my opinion dual-channel DDR5 should be written as dual-channel, not quad-channel or "4 x 32" just because the 64-bit bus is split into two like that. Both dual-channel DDR4 and dual-channel DDR5 provide 128 bits wide worth of bus. Take a look at this LTT forum post. "Each integrated channel in DDR5 has half as many data lines as the single channel in DDR4. So they have the same bandwidth at the same speed ..." indeed, if you set both a 2ch DDR4 and DDR5 setup to both the same memory frequency, let's say 4000 MT/s for example, you get overall the same bandwidth. True quad-channel memory would give you double the bandwidth at same frequencies. It's highly misleading to actually call it quad-channel in my opinion. Furthermore, the sources themselves say it is 2 channels, not 4 channels. I have to agree with 176.9.113.53 on this one, the Intel Ark page says it is 2ch, not 4ch!

If dual-channel DDR5 really counts as quad-channel, then I guess ganged mode on AMD Phenom platform from 2008 counts as "single-channel", even though you are using the same dual-channel memory setup as you would with unganged, and regardless of if it's ganged or unganged you get more bandwidth with dual-channel.

AP 499D25 (talk) 09:09, 9 March 2023 (UTC)

«True quad-channel memory would give you double the bandwidth at same frequencies.» => and at same channel width. Modern RAM buses can use various channel width (16, 32, 64), see for example LPDDR#Bus width. Visite fortuitement prolongée (talk) 13:35, 11 March 2023 (UTC)

«Furthermore, the sources themselves say it is 2 channels, not 4 channels.» => The DDR5 SDRAM article say in wikivoice that «DDR5 supports a speed of 51.2 GB/s per module and 2 memory channels per module.» and that «Each DIMM has two independent channels.», so implicitly that it is 4 channels. Visite fortuitement prolongée (talk) 13:35, 11 March 2023 (UTC)

100% of motherboard and CPU vendors continue to call DDR5 for consumer platforms dual-channel. WP is not going to be exclusive and do something which serious tech companies disregard. Artem S. Tashkinov (talk) 04:36, 12 March 2023 (UTC)

I have since realised that on a technical level, DDR5 memory actually contains two channels per DIMM, so what is called a 'dual-channel' setup is technically quad-channel.

"DDR5 supports a speed of 51.2 GB/s per module" - clock speed is a direct factor of the bandwidth of memory. So for DDR5-4800 it's 38.4 GB/s per DIMM if I'm not mistaken. That bandwidth suggests to me it is running at a clock rate of 6400 MT/s.

"and at same channel width." - this does not make sense to me. What I mean here, is the total number of bus bits, resulting from width of the memory channels (32-bit), multiplied by number of channels (4). That results in essentially 128-bits wide bus. The total bus width is the same as DDR4 (64-bits per channel * 2 channels). So at the same clock speeds, you get the same bandwidth. The reason why DDR5 is faster is because it runs at higher clock rates (e.g. 6400 MT/s instead of 3200 MT/s).

I have realised that the reason why tech companies still refer to it as 'dual-channel', despite there being actually four, or even eight channels, is for simplicity, as well as to avoid confusion for the less tech-informed people. The companies view a memory 'channel' as being 64-bits wide essentially. Actually calling it quad- or octa-channel would suggest to the average consumer that their computer memory is two or four times faster than it actually is. That's what I meant by the "true quad-channel" statement. Someone might think that their 'quad-channel' setup is providing 204.8 GB/s of total bandwidth instead of the 102.4 GB/s bandwidth it actually provides.

I just learned soon after the posting of the previous message that dual-channel and double data rate (DDR) are two separate things.

I'm still on the stance that we should present the memory support as dual-channel here, as that's what the companies call it, together with the reasoning behind it, stated above. Now, what we could do, is add an explanatory footnote (EFN), to say that the dual-channel setup technically contains four channels, but they call it dual-channel for ease of mind for the average consumer.

AP 499D25 (talk) 05:15, 15 March 2023 (UTC)

Supported via PCH on desktop processors

For Thunderbolt 3/4. What does it mean? Does it mean Titan/Maple Ridge is not needed? Does it mean there are pins to support Thunderbolt directly like on new Asus Apex Encore? Also some articles say 80 gbit/s will not work in the future with 14th gen. Also, there is now this. https://www.techpowerup.com/318041/asus-reveals-its-usb4-add-in-card-with-60-watt-usb-power-delivery-support Valery Zapolodov (talk) 22:34, 8 March 2024 (UTC)

Apparently, that USB4/Thunderbolt 4 header is just used for things like GPIO, LPC, SMBus, I2C etc. So the extra pin-header is not just about hot swapping, but it's also related to suspend and sleep, which won't work correctly without the extra connector. See in comments: https://www.techpowerup.com/318041/asus-reveals-its-usb4-add-in-card-with-60-watt-usb-power-delivery-support Valery Zapolodov (talk) 15:19, 10 March 2024 (UTC)

Intel CPUs suffixes

I really like the suffixes tidbit right before the table in https://en.wikipedia.org/wiki/Raptor_Lake#Raptor_Lake-S

Here's what I don't like: it's only for the first table, and it is impossible to spot/find for all other tables. It also takes quite a lot of vertical space and results in more scrolling. Maybe we could turn it into a template and include it via {{ }}? And then not just that, maybe list the items on a single line and put it under each relevant table, so it'll look something like that:

[ TABLE ]

• Suffixes: T - low power (reduced TDP); F - no iGPU; K - unlocked OC; KF - unlocked OC without iGPU; KS - special edition of unlocked OC

What do people think? This could be handy for other articles on Wikipedia. @AP 499D25 Artem S. Tashkinov (talk) 08:41, 30 March 2024 (UTC)

Sounds good! I like it put out all on one line, reduces the amount of space it takes up vertically a lot.

I think the only issue with creating one template to use on pretty much all generations of Intel CPUs, both desktop and mobile, is that it'd be pretty long, as there are many different mobile processor suffixes, alongside some old gen desktop processors had suffixes like "S" meaning reduced TDP (65W), "P" meaning no integrated graphics and whatnot. — AP 499D25 (talk) 09:27, 30 March 2024 (UTC)

There could be different templates for different families/generations/line ups. Artem S. Tashkinov (talk) 10:32, 30 March 2024 (UTC)

Okay that could work too!

I'll probably come to it later though after I finish the (still incomplete) merge and rework of tables to List of Intel Core processors. I'm thinking about moving those tables into templates as well like how the Ryzen lists are set up. But anyways, I agree with your proposal. — AP 499D25 (talk) 11:47, 30 March 2024 (UTC)