archives

/archive 1

Welcome

Welcome to Wikichip's General Discussion!

Trying to find your way? check WikiChip:welcome.

Click here to start a new topic.

Rules

• Don't be a jerk
• Don't bite the newcomers
• Off-topic discussions are allowed but should be kept to a minimum
• If a topic has its own article, comments should be posted on its talk page

---

Discussion

Server upgrades & MediaWiki 1.27

Hey everyone, just want post here letting you know that next week I'll be taking the time to update my server to Fedora 24. Additionally I'll be updating MediaWiki 1.27 seeing as our current version (MediaWiki 1.25) has reached EOL. So..... if things stop working, there is a good chance I've caused it. Just let me know! -David (talk) 02:58, 29 July 2016 (EDT)

So, it's been a month since I actually posted it and it since been delayed because I decided to upgrade everything. Yesterday I went ahead and actually upgraded the entire server. Everything was outdated by years and had to be upgraded to latest version. This was also necessary due to the increase in web traffic we've been receiving. This is WikiChip's current setup:

• Intel Xeon E5-2680 v3
• Fedora 24
• Nginx 1.10.1
• PHP 5.6.25
• MariaDB 10.1.16
• MediaWiki 1.27.1
• Semantic MediaWiki 2.4.1

As far as I'm aware all extensions are working. There was a small hiccups with Semantic MediaWiki, but it too seems to work fine (as far as I can tell). The Extension:CustomSidebar that were were using (which was already on life support by 1.25) has completely stopped working. Looks like many of the functions that were deprecated for a very long time were finally removed. I've commented out the affected templates but this does have to be dealt with somehow. Lastly, some languages (such as our mirc section have lost their syntax highlighting). We have to manually fix this up as well somehow. --David (talk)

SSL

Effective immediately WikiChip now uses SSL for everything. All "http" links will automatically redirect to their "https" versions. SSL Certificate is provided by Let's Encrypt project. --David (talk) 18:03, 11 September 2016 (EDT)

That's awesome thanks David! --Inject (talk) 15:34, 12 September 2016 (EDT)

possibly bug with sort?

Has anyone else noticed that when we setup a semantic subquery with a sort option, if the property specified in the sort isn't available for a specific result, it gets discarded completely? This is actually causing some serious issues for processors that do not have a launch date yet. --Inject (talk) 03:33, 6 December 2016 (EST)

I did some searching and I found https://www.semantic-mediawiki.org/wiki/Help:Sorting saying

Sorting a query also influences the result of a query, because it is only possible to sort by property values that a page actually has. Therefore, if a query is ordered by a property (say Population) then SMW will usually restrict the query results to those pages that have at least one value for this property (i.e. only pages with specified population appear). Therefore, if the query does not require yet that the property is present in each query result, then SMW will silently add this condition.

This really bad news for us. Because we certainly have processors with Property:first launched blank (for obvious reason as they're not out yet). Sorting by launched date will filter those processors out. I'm not sure how to actually address this problem. --ChipIt (talk) 12:52, 6 December 2016 (EST)

I asked on the Semantic MediaWiki mailing list. Apparently if the field we're sorting by is not specified in the ask selection list, sort won't filter out NULL properties. If that's indeed the case, we can create a secondary "ldate" property that mirrors the first one - perhaps even have it set to like '3000' by default if 'future=yes'. --David (talk) 03:13, 8 December 2016 (EST)

max memory is finally in MiB

we should've done it a long time ago, but I've finally changed Property:max memory to MiB for consistency. That does mean a hell lot of stuff (max memory related, mainly in tables/queries) will be somewhat broken. So if you find something, do change it appropriately. Thanks! --David (talk) 16:58, 4 January 2017 (EST)

Yea we should have done it ages ago. I think there are a bunch of other properties we need to convert as well. --ChipIt (talk) 19:40, 4 January 2017 (EST)

Switching over to the new skin

Hey everyone. I think I'll be pushing the new skin today. If anyone notices something that's broken because of this change, please point them out for me so I can fix them.

Thank you! --David (talk) 17:14, 7 January 2017 (EST)

Everything looks good on my side David. I have noticed the menu icons (I think all font awesome icons?) are not loading on Edge too. But otherwise they work fine on Chrome/FireFox/Opera as far as I can tell. --ChipIt (talk) 18:38, 7 January 2017 (EST)

Kaby Lake articles

Now that Kaby Lake is fully released (except for W/X I guess) and we have most of the related articles somewhat built up. Can anyone go through the specs of the chips and double check that we have the right info everywhere? I've spent a ton of time making sure everything looks alright but a second pair of eyes wouldn't hurt. --David (talk) 11:00, 13 January 2017 (EST)

Everything looks fine for the most part. Intel has not yet released an update to their optimization manual so there could be a number of tiny arch related changes we need to document once that's out. We are also missing a bunch of articles relating to the sockets used in Kaby. Then we really need to shift our focus on Zen, before it's launched! --ChipIt (talk) 15:09, 13 January 2017 (EST)

MathML is failing

I've noticed all our math tags have stopped working probably because formulasearchengine is down for whatever reason. Anything we can do about this? --At32Hz (talk) 01:19, 6 February 2017 (EST)

You can see the errors on this page Tebibyte for example. --At32Hz (talk) 01:19, 6 February 2017 (EST)

Semantic MediaWiki 2.5.0

WikiChip has been upgraded to Semantic MediaWiki 2.5.0 (Release Notes) --David (talk) 12:05, 9 April 2017 (EDT)

WikiChip has been upgraded to MediaWiki 1.28.1 (Release Notes) --David (talk) 12:05, 10 April 2017 (EDT)

In an attempt to cut down on both bandwidth and load time we've done away with the full Font Awesome and instead switched to a handful of used SVG variants. --David (talk) 12:42, 10 April 2017 (EDT)

I think Jon mentioned it before but we need to reinstall the EastTimetine extension. Not really sure why we had a problem with it in the first place but I think we need to re-introduce it. Any thoughts? --ChipIt (talk) 01:46, 12 April 2017 (EDT)

I will see what I can do. --David (talk) 03:22, 12 April 2017 (EDT)

It's installed! --David (talk) 19:20, 15 April 2017 (EDT)

Microprocessor comparison page

So this has been talked about for a bit now so I figured I'll detail it here so everyone can see and possibly take part of the discussion if they want. As we continue to detail every individual model we've gained the ability to make use of the data in a number of unique ways. One such way is leveraging the embedded semantic properties to create a comparison table feature (similar to Intel's ark) - but we can do it on a much grander scale! Compare any model to any other model! E.g., VIA next to AMD next to an Intel model.

In our current flimsy prototype you can compare virtually any model to any model (e.g. an Intel 8086 to an i7-7700K to a Qualcomm 630 but such comparison diminish in their usefulness as their differences (especially ISA, but also date, node, market segment, etc..) grow. We will likely keep that functionality moving forward but add some kind of indicator informing the user they are comparing models that differ greatly.

Anyway. This is an initial post, I'll post updates as we continue to make progress. --ChipIt (talk) 02:18, 12 April 2017 (EDT)

Mobile site style fixes

So I've been spending some time attempting to fix some of the issues people have been having. Almost all the issues at the moment involve the mobile style. It getting clear how difficult it is to try to make our site responsively size to the smallest devices that visit our website - particularly iPhones. There's been a number of requests to try to address some of those issues so that's where my focus will be. If anyone is seeing some more problems, please drop them here. --ChipIt (talk) 00:50, 23 May 2017 (EDT)

Feel free to rework the entire front page pretty much. As it stand it looks like ass on mobile. Nothing really shifts and scales there properly. This is even a problem on low rest tables. --David (talk) 06:44, 24 May 2017 (EDT)

uarch template is now mobile-friendly - I updateed that template. Other templates will have to wait as they are way too complex for a simple switch. --David (talk) 08:22, 26 May 2017 (EDT)

The main page was switched to a responsive grid layout. It's in pretty good shape but some rough spots needs to possibly be fixed. --ChipIt (talk) 01:21, 27 May 2017 (EDT)

The main microprocessor template is now being switched over to the new template which is much more mobile friendly. --David (talk) 21:36, 7 June 2017 (EDT)

Current AVX-512 indications

Currently we just have one value for "AVX-512", but since it's not monolithic and various variations are used for Phi while other variations are used for Skylake SP and the other variations will be added in Cannonlake, we need to break it down into the various components (AVX-512F/CD/ER/PF/BW/DQ/VL et al). --ChipIt (talk) 02:24, 27 June 2017 (EDT)

Yea we need to break it down. I'll do that today or tomorrow. --David (talk) 17:20, 27 June 2017 (EDT)

Done! --David (talk) 01:44, 11 July 2017 (EDT)

SPEC CPU2017 & benchmarks

Given vendors are now rolling out the new SPEC CPU2017 scores, I've added a new template {{benchmarks main}} template we can start experiment with adding official SPEC benchmarks to the pages. --ChipIt (talk) 01:19, 5 July 2017 (EDT)

New expansion template

Using ChipIt's original {{benchmarks main}} idea, I've created {{expansions main}} which should replace the older expansions template. This new template is far more customization and scalable then what we previously had. It's also more flexible in term of the kind of information we can. It still uses the old Semantic properties, so everything else should work just fine. --David (talk) 23:35, 16 August 2017 (EDT)

HotChips 29

As you may know, a few weeks ago HotChips 29 was held. I finally finished collecting and organizing all of my notes from the event and I'll start dumping that info in the appropriate articles. So keep an eye on this in the weeks to come. I'll probably need some grammar correction and maybe some diagrams designed too, we'll see. --David (talk) 00:36, 14 September 2017 (EDT)

• Scorpio Engine Microsoft's new SoC found in their new Xbox One X consoles. --David (talk) 22:57, 20 September 2017 (EDT)

Query on citation policy

Hi David, all:

I'm intrigued by the site, but in the first few articles I've looked at I haven't seen any citations. For Intel processors it looks like much of the information is coming from ark.intel.com, but that connection isn't made explicit.

What do you think your policy will be on providing cites going forward?

Best,

Barry Rountree (talk) 23:33, 12 July 2018 (EDT)

Hey Barry, this is actually a real pickle for us. The connection isn't made explicit because the info doesn't come from ARK. Understandably, like Wikipedia article, people expect every article to reference lots of external sources. Unfortunately for us, we mostly curate our own information based on direct information (i.e., directly from the primary source - Intel, AMD, whichever) or, alternatively, in cases where we attend IEEE conferences and obtain the information from the presentations and possibly through an interview with the presenter (that's the case for some of our microarchitectural pages).

With that in mind, our processor information almost always comes the same way - the company sends us this information as part of their media information they send out to tech websites. Some parts of the specifications are derived based on the underlying architectural details of the chip (e.g., cache structure), while some parts of the specifications we actually go the extra mile to personally request - in Intel's case it can be code names, turbo frequencies, part numbers, etc.. All that info comes directly from Intel as opposed to ark (we usually have more info than ARK too). To be clear, this is not limited to Intel. This is true for AMD, ARM, and anyone else who announces chips (e.g., Cavium's recent ThunderX2 and Qualcomm's Centriq - those chips are not even listed on their website at all).

Basically, this boils down to the reference section just saying "Intel/AMD/Cavium/et.." or "Directly from Intel/AMD/etc.." which seems to confuse people more than it helps. But that's literally the reference for us. We actually do write that for non-spec pages such as Cavium's Vulcan just for the sake of having some kind of reference. But it looks just as silly.

I am open to suggestions on how to address this. --David (talk) 07:35, 13 July 2018 (EDT)

Just to elaborate what David meant by "direct information". The most common case is when the company sends the information via email (he is usually assigned a PR contact) and might be in the form of press deck or press video briefing. Many times (since we are pretty technical), the briefing is not detailed enough, so we end up with a series of back-and-forth emails asking for additional information/clarifications. Sometimes a phone call is involved or he attends an actual press event. --At32Hz (talk) 07:46, 13 July 2018 (EDT)

Here is an example from today https://fuse.wikichip.org/news/1482/intel-launches-desktop-xeon-e-their-fastest-entry-level-workstation-processors/ --At32Hz (talk) 22:27, 13 July 2018 (EDT)

Hi At32Hz, David.

I'm a computer scientist at Lawrence Livermore National Laboratory and I publish a fair amount on power-constrained performance optimization in high-performance computing. For my particular needs, I have to be able to cite primary sources. That can be *pers. comm.* (personal communication) if need be, but the general practice there is providing a name and company affiliation. Trying to decorate every fact and figure with a cite, however, is overkill and not that helpful.

Instead, I'd like to see a "Bibliography" section that listed the relevant public documentation (perhaps NDA as well, noted as such), and brief discussions of which aspects of the article needed further, non-public sourcing. For Intel processors, that might be pointers to ARK, the specification updates, the datasheets, the NDA external design specifications and BIOS Writers' Guides, and conference presentations. Just telling readers that those documents exist can be a huge help, and if a particular claim (e.g., "P-state requests only set a ceiling") are based on conversations, I know not to spend my time digging through the docs looking for that particular factoid.

Adding this information would be a nontrivial amount of work, and I'd want to update most of the (say) Intel processor articles myself to see if it's doable prior to asking anyone else to follow suit. So all that said, if I were to begin adding bibliography sections, would that be seen as useful by other folks here?

Barry Rountree (talk) 23:45, 13 July 2018 (EDT)

We can definitely try to add bibliography sections that lists those documentations. We can actually create a template and automate a lot of it since a single document typically covers a whole series.

I want to point out that we've also tried that before and things got very complicated, mainly from the maintainability side. We're certainly willing to give it another shot. We started doing it for some of our early pages when all of a sudden Intel started removing old documents. If you check the Atom N270 article, you'll see what we ended up doing - archiving the data sheet and spec update because Intel decided to purge it from their site one day. I think they eventually restored some of it after we asked them about it, but it highlights the whole issue we have with old documents/datasheets. For the really old stuff, like AMD's K5, we have gone ahead and archived every datasheet ourselves because they were simply disappearing from the AMD website and they did not seem to care even after we asked. We've started to preserve Intel's optimizations manuals as well (see intel/documents) because Intel started removing the very old microarchitectures from their manuals which can now only be found in the earlier revisions and those are obviously no longer on the Intel website. This really sucks because what good is listing "Xth generation core specification update" if no one can actually find that document anymore or the document no longer covers it? So we ended up archiving it. The good thing is that for the most recent stuff, this unlikely to happen any time soon (though it's already happening with the Sandy Bridge documents), but given we have 1000s of Intel processors cataloged, there is a good chance those references will start to rot within the next 5 years.

If we try to generalize this procedure for other companies, things fall apart very quickly. Take for example LLNL's new Astra supercomputer which is based on the ThunderX2 CN9975 chip based on Vulcan. As far as we are aware, there are no public datasheets available and Cavium doesn't even list that part on their website. Actually, a quick search pretty much brings on our website and all our info came directly from a de-briefing from Cavium (and some from Broadcom). So I'm not sure what we would even reference in those cases. Keep in mind this is not a rare case either, it's very common. Take Qualcomm's Centriq 2452 server chip as another example.

--At32Hz (talk) 09:22, 14 July 2018 (EDT)

The whole bibliography section sort of partially work, but only for Intel. We might be able to put some kind of an expiration date check (using a template or something) to make sure stuff isn't outdated or possibly just request permission to host those docs ourself. Cavium, Qualcomm, AMD, et al are entirely a different problem. AMD has no public documents, no specs update document, and no ark. All their info comes directly from the press deck they send us and via their pre-briefing. I wouldn't even know what we could reference. --David (talk) 11:15, 14 July 2018 (EDT)

With the understanding that this may be an Intel-specific solution, would both of you be interested in extension of what you're doing with the optimization manual page (which I like very much, btw) to datasheets and specification updates? That would solve a problem I need to address in my other work, and I'd rather the documents be archived here rather than in my git repo. I'm happy to postpone the discussion of bibliography sections until after there's a stable document archive.

Barry Rountree (talk) 00:35, 15 July 2018 (EDT)

Is your git repo public? It might be a good starting point for us. --David (talk) 09:14, 21 July 2018 (EDT)

Category

What is difference between these categories: Category:microprocessors, Category:Microprocessors? Oleg3280 (talk) 08:25, 21 July 2018 (EDT)

All pages on the site are supposed lowercase (with a few exceptions). The second category with uppercase 'M' should not really exist. I think it was created a very long time ago and we simply forgot to correct it (redirect it to the second one after correcting the links). --David (talk) 09:14, 21 July 2018 (EDT)

I changes pages that had the uppercase M version and redirected the category, unfortunately there are a bunch of category that also need to be fixed. --David (talk) 09:19, 21 July 2018 (EDT)

Change "designer" wikidata mapping

Currently designer property has wikidata id=P287, but I think that should probably be P178 instead. P287 is "designer" which has range "Human" while P178 is "developer" and has range "Human" or "organization".

This also raises the question if the property should be named "developer" too, it is what the ChipWiki templates use (and the wikidata property).

I can do the changes, I just wanted to get an approval first.

--Trygvis (talk) 03:47, 23 August 2018 (EDT)

There is currently no real implications to using either. It's just an early attempt to map our major properties to WikiData for programming purposes. I think the main reason we've gone with 'designer' is that it sounds more descriptive to what we're trying to define since we have a stronger distinction between 'designer' (the IP designer) and 'manufacturer' (the actual fab). To that end, developer sounds slightly ambiguous. But you can change the wikidata property if you think the later one is better. --At32Hz (talk) 10:34, 23 August 2018 (EDT)

Ok, I did the change. Thanks for the information. --Trygvis (talk) 08:34, 24 August 2018 (EDT)

More semantic queries?

I was looking at the list of 10 nm Microprocessors and noticed that the table is created manually but is really a good match for an ask query.

As an experiment I made a query that more or less shows the same data (click on "edit query" to see the details). Similar serach for 10 nm microarchitectures. For the second search it might be useful to split it up into two tables, but it shows more of the possibilities.

As a side note, I see that the designer property often have several values, for this table to be prettier it would probably be nice to have a "main designer"/"owner" or something similar. Or perhaps a single designer property and use a "contributor" field to list the other parties.

--Trygvis (talk) 05:53, 23 August 2018 (EDT)

You can actually just do something like this. I just realized we are also missing a 'name' property on the ic family template (the whole template needs to be switched to our new style too). An additional 'main designer' property really makes not only here but for all other templates in order to pick the one name the subject is associated with most closely. We might just want to always make it equal the same value as the first defined 'designer' to simplify this (and pretend the main designer is always 1 value). --At32Hz (talk) 11:02, 23 August 2018 (EDT)

Yeah I was thinking of just picking the first designer too and use that as a main designer. Is "primary designer" a better term for it? I can try to adjust the template if that is ok. I'm new here so I'm just trying to figure out how much stuff I can do without stepping on someones toes.

I'm happy to improve the 10 nm page and friends by replacing the manually made lists with ask queries if that is ok.

So I've changed Template:ic family to set {{developer}} as main developer (diff) and changed the 10nm page. Compare the queries without main developer to with main developer

However, the list would be prettier if the pages DISPLAYTITLE was not set. I'm not sure if it is possible to change how the list is rendered without having a special template for it. Ideally I think it would be nice to use the tree format, but I couldn't figure out how to nest the pages. I tried to set parent=main designer but that didn't work.