PSA: Intel Graphics Drivers Now Collect Telemetry (after Opt-In)

[btarunr]

Graphics cards are the most dynamic hardware components of the modern PC, in need of constant driver updates to keep them optimized for the latest games. Intel may be the newest on the block with discrete gaming GPUs, with its Arc A-series competing in the mid-range, but the company has a vast software engineering muscle that ensures a constant stream of driver updates for these GPUs regardless of their smaller market share compared to entrenched players NVIDIA and AMD. A part of keeping the drivers up-to-date and understanding the user-base to improve future generations of GPUs, involves data-collection from the existing users.

The updated installer of Intel Arc GPU Graphics Drivers lets users decide if they want the company to collect anonymous usage data from them. For those with the data-collection already opted in, the installer leaves the data-collection component untouched in the "typical" installation option, and presents it as an optional action item in the "Customize" installation option. For those that did not opt for data-collection, the "typical" installation option doesn't sneak the component in, but presents it as an unchecked optional item in the "Customize" screen. An older version of this article stated that the data-collection component, dubbed Computing Improvement Program (CIP), would install onto unsuspecting users' machines in the "typical" installation, disregarding their prior choices with the component. We have since significantly changed our article as Intel clarified many of our questions and demystified CIF, what its scope of data-collection is, and how it makes its way to your machine with Intel's driver software.



When you run the installer the very first time, you are presented with two agreements to accept. The first one of course is a standard Software License Agreement. Declining this will cause the installer to close, but accepting it will take you to the next screen, where you're presented with an agreement dedicated to data collection. This one clearly spells out the scope of data-collection of CIP, and what it does not collect. You can accept this agreement, which will install and enable CIP; or you can even choose to decline this agreement, which will cause installation to proceed without telemetry.

The first time you run the Arc GPU Graphics Software installer, it remembers whether you opted for or against CIP (whether you accepted or declined the data collection agreement); by leaving a Windows Registry key located in "HKEY_LOCAL_MACHINE\SOFTWARE\Intel\SUR\ICIP". All future instances of Arc GPU Graphics Software installer run on your machine, be it a "clean" reinstall, or a simple upgrade; will check your opt-in status by looking up this Registry key. Apparently, our machine from the original article was opted into CIP, and so the installer assumed that since we had accepted its agreement, it should simply install CIP as part of the "typical" option, and only present it as an action item in the "customize" screen, where you can choose not to install CIP, even if you had accepted its agreement.



You can remove CIP from your machine at any given time from "Add/Remove Programs" in the Windows Control Panel, since CIP is its own item in the installed software list. Uninstalling CIP from here does not disturb the installation of other Intel Software products or drivers (eg: your Chipset INF, Management Engine, Networking, or even Arc GPU Graphics drivers).

It's pertinent to note here, that since the CIP agreement is presented right after the main Intel Software License Agreement, and since it is a visually similar screen, gamers in a hurry could be tricked into thinking that the CIP agreement is as obligatory as the SLA. They could be made to think that one must accept both agreements for the installer to proceed, since the first one is mandatory. There's also the likelihood that someone in a hurry would simply keep mashing that "Accept" button below a wall of text, without bothering to read what it is.

This is similar to how some freeware programs include opt-ins for third-party software as part of the installer that they are tied up with (eg: anti-virus utilities included with Adobe Reader). So far, AMD's handling of this telemetry opt-in with the user in their Radeon Software Adrenalin drivers comes across as the least stealthy or cloak-and-dagger. You're presented with a big, centrally-located checkbox to decide if you want AMD to collect data, at the completion of the driver installation. This way you know that your drivers are installed, and that the data-collection tool is truly optional.

View at TechPowerUp Main Site

 

[Jism]

"The categories of websites you visit, but not the URL"

So how does that Categorizing exactly work then? It has to know the URL in order to categorize what your doing.

 

[thunderingroar]

I mean can you really blame them? One of the reasons why NV is so quick on driver hotfixes is because they collect a fuckton of telemetry data. As much as people cry about geforce experience forced logins and telemetry, at the end of the day they still give their money to nvidia.

Intel and AMD have to work with only voluntary telemetry data and with much smaller market share which results in far fewer user crash reports

 

[NeuralNexus]

I mean can you really blame them? One of the reasons why NV is so quick on driver hotfixes is because they collect a fuckton of telemetry data. As much as people cry about geforce experience forced logins and telemetry, at the end of the day they still give their money to nvidia.

Intel and AMD have to work with only voluntary telemetry data and with much smaller market share which results in far fewer user crash reports

True, which is why I do not buy into the foolishness that AMD drivers are bad.

 

[_JP_]

It's for a beta driver, which is in development. I'm surprised people are not finding it the norm that beta drivers are to provide feedback to the developers by default, as opposed to stable WHQL releases.

 

[Nostras]

"The categories of websites you visit, but not the URL"

So how does that Categorizing exactly work then? It has to know the URL in order to categorize what your doing.

If I had to guess there's a local lookup for website to category. If found it will send it, otherwise it won't.
Doesn't explicitly state the SW won't be able to see the URL, just that it won't phone that information home.

 

[TheDeeGee]

It's really simple, can't improve without data.

Those removing telemetry with clean driver tools only make bugs take longer to fix.

 

[chrcoluk]

Not surprising, I am surprised though they didnt do this from day 1.

Think of how the bug finding normally works, unlike reviewers the typical end user has to report things to standard tech support of which there is triage process in place, so many reports might be incorrectly fobbed off or not end up at the right place, telemetry bypasses it all.

Dev's need to be upfront though, explain telemetry is in use, and provide options for the level of telemetry one is ok providing. As part of the issues there is distrust, people dont know if data is collected about PC, or just for the drivers.

I recently uninstalled google drive as an example, I discovered it was attaching a crash reporter process to all my firefox processes, all msedge processes, all notepad processes, to some games, and to some media apps. There was no information on this in google drive itself, and I couldnt find a way to adjust its settings or disable it, I could only remove it by uninstalling google drive windows app. Firefox now starts about 3x as fast when I launch it.

 

[Assimilator]

I'm very interested to understand what category of websites I visit, have anything to do with a driver not working. I don't really care though, since I only use the Intel driver on a work laptop and DGAF if they know the sites I browse there.

It's for a beta driver, which is in development. I'm surprised people are not finding it the norm that beta drivers are to provide feedback to the developers by default, as opposed to stable WHQL releases.

It's in the latest WHQL 4577.

If I had to guess there's a local lookup for website to category. If found it will send it, otherwise it won't.
Doesn't explicitly state the SW won't be able to see the URL, just that it won't phone that information home.

I very much doubt that. Rather, the telemetry collector will call out to a third-party company that does URL classification, with some sort of obfuscated/encrypted payload containing an URL, and that company will return the URL category. That way there is no way for the classifying company to know who was browsing that URL, while the driver knows the user and category but allegedly disposes of the URL.

 

[Verpal]

With the complexity of GPU driver and likely extraordinarily poor quality each support ticket provide manufacturer with information....

I am surprised Intel take this long to turn telemetry on by default.

Unlikely the days before gaming GPU from Intel, people expect different things now, back then Intel only need to concern with quicksync, and those few important producitivity software work as intended, then a couple low spec competitive game work decently well, and that's it. But now, every game had to be supported, one way or another, I hope telemetry does help Intel getting more competitive in GPU space in long run.

 

[Tahagomizer]

Good guy corporation steals your data but gives you vague promises that you *might* get some bugs fixed faster. Because, obviously, the websites you visit are an important factor for your graphics card driver functionality...
This really is a dystopian timeline where every product you buy is your enemy and corporate apologists will try to convince you it's for your own good.

 

[Vayra86]

Good guy corporation steals your data but gives you vague promises that you *might* get some bugs fixed faster. Because, obviously, the websites you visit are an important factor for your graphics card driver functionality...
This really is a dystopian timeline where every product you buy is your enemy and corporate apologists will try to convince you it's for your own good.

The problem with the dystopia is that if you bought the product, it really is for your own good because then shit might actually get fixed proper

The solution therefore is: 'don't buy' or 'buy something that works the right way'.

Its why I switched to AMD after decades of Nvidia now. Nvidia is a software company now before a Geforce GPU manufacturer. They just stick Geforce on an enterprise cut down and plaster endless waves of software on top to make it good (or, sell). Ada is the first most blatant example, selling almost exclusively on DLSS3, because RT isn't flying after 3 iterations of enterprise cut downs.

 

[b1k3rdude]

Same response to the nvidia driver.... that is a hard F**k, NOPE.

So if and when I ever install an arq card, I may need an intel Arq version of "NV clean install" to strip out ant Geforce Exp. like crap crap and then block the driver from talking to the internet via windows firewall (via WFC) as I do for the nvidia driver.

But give props to intel, atleast (for now... dont forget this is Intel and all the s**t they have done in the past) you can disable it in the driver.

 

[Jism]

More and more privacy evasive stuff. I think Linux in the near future will be mature enough to offer a full replacement of Windows and all it's telemetry.

We got more bugs, chrashes and even data loss since we've allowed telemetry in the first place - updates where never getting better. Only worse.

 

[Assimilator]

Nvidia is a software company now before a Geforce GPU manufacturer. They just stick Geforce on an enterprise cut down and plaster endless waves of software on top to make it good (or, sell). Ada is the first most blatant example, selling almost exclusively on DLSS3, because RT isn't flying after 3 iterations of enterprise cut downs.

What absolute rubbish.

Ada is unquestionably both the most powerful and efficient GPU architecture NVIDIA has ever produced. To quote W1zz's review of the 4090:

+45% vs RTX 3090 Ti. Yup, 45% faster than last generation's flagship—this is probably the largest jump Gen-over-Gen for many years. Compared to RTX 3080 the uplift is 89%—wow!—almost twice as fast

Energy efficiency is almost doubled compared to cards like RTX 3090 / 3090 Ti

As for its ray-tracing:

While previously enabling RT at 4K always meant some compromises—either upscaling or reduced settings—the RTX 4090 will give you 60 FPS with RT active in nearly all titles.

That is not software, that is pure brute force hardware engineering.

 

[Vayra86]

What absolute rubbish.

Ada is unquestionably both the most powerful and efficient GPU architecture NVIDIA has ever produced. To quote W1zz's review of the 4090:



As for its ray-tracing:


That is not software, that is pure brute force hardware engineering.

Its also a 4090, which is not the rest of the Ada stack, and definitely not where the market share is.

Let's compare an x60 to last gen shall we
And also, 'most powerful and efficient' architecture ever applies to every gen, again and again, and when it doesn't, boy is it shit... that's the kind of architecture that generally doesn't even get released Perhaps that applies to AMD's midrange update huh...
Remove the Kool Aid for a minute and look at where we are now with current gen GPUs. Its not great and the 4090 is the unicorn.

 

[Assimilator]

Its also a 4090, which is not the rest of the Ada stack, and definitely not where the market share is.

Let's compare an x60 to last gen shall we
And also, 'most powerful and efficient' architecture ever applies to every gen, again and again, and when it doesn't, boy is it shit... that's the kind of architecture that generally doesn't even get released Perhaps that applies to AMD's midrange update huh...
Remove the Kool Aid for a minute and look at where we are now with current gen GPUs. Its not great and the 4090 is the unicorn.

Allow me to be more precise: most powerful and efficient architecture compared to the previous generation, in multiple generations.

And don't go moving the goalposts. Your claim was that "NVIDIA is a software company" which is quite obviously nonsense. They are a hardware company and like it or not, very good at what they do.

 

[Vayra86]

Allow me to be more precise: most powerful and efficient architecture compared to the previous generation, in multiple generations.

And don't go moving the goalposts. Your claim was that "NVIDIA is a software company" which is quite obviously nonsense. They are a hardware company and like it or not, very good at what they do.

They're diversifying into a software oriented company too, let's put it that way then And its already been true for quite a while that you dó buy a Geforce for its better software suite/feature set, its CUDA support that works in software, etc. DLSS3 is another such item, as much as Shadowplay, Ansel, etc. An Nvidia GPU without it loses a big chunk of its attractiveness.

Also, the more precise comment still only and exclusively applies to the 4090.

 

[Deleted member 185088]

They lost lots of money on Ark, they need to make some of it back.

 

[chrcoluk]

More and more privacy evasive stuff. I think Linux in the near future will be mature enough to offer a full replacement of Windows and all it's telemetry.

We got more bugs, chrashes and even data loss since we've allowed telemetry in the first place - updates where never getting better. Only worse.

A thing that telemetry can make worse is a developer finds it if a feature has low usage, then it can mean the feature gets dropped as then easier to maintain code, then users of said feature become unhappy. I think it also led to nagware prompts, developer makes new feature wants everyone to use, gets low take up, so software then keeps nagging you to use the feature. It becomes really obvious the desperation when the nag prompts dont allow you to say no as a permanent answer but rather "not now", common place on android apps.

 

[Luke357]

Same response to the nvidia driver.... that is a hard F**k, NOPE.

So if and when I ever install an arq card, I may need an intel Arq version of "NV clean install" to strip out ant Geforce Exp. like crap crap and then block the driver from talking to the internet via windows firewall (via WFC) as I do for the nvidia driver.

But give props to intel, atleast (for now... dont forget this is Intel and all the s**t they have done in the past) you can disable it in the driver.

The telemetry in the ARC driver can be turned off right in the installer. I don't like telemetry either but they give you the option to not have it even before you install the driver so it's not like it's always running.

 

[john_]

OK, let's do an on topic question then. How does browsing behavior helps the development of the drivers? Are future Intel GPUs going to be optimized for Instagram if people spend too much time on Instagram? Maybe a custom accelerator?

The telemetry in the ARC driver can be turned off right in the installer. I don't like telemetry either but they give you the option to not have it even before you install the driver so it's not like it's always running.

Most people will press "Yes, next, next, next, yes".

 

[Fouquin]

I'm surprised people are not finding it the norm that beta drivers are to provide feedback to the developers by default, as opposed to stable WHQL releases.

If beta builds were even remotely optional than perhaps it would be excusable to shoehorn such features in. The problem right now is Intel is shipping beta drivers that are required to even get a game to run correctly. Remnant II for example was a hot mess with half the screen obscured by artifacts for the first week and with performance dips into the single digits. Solution? Intel dropped a beta driver that allows the game to render correctly in conjunction with a beta build of the game itself. So sure, data collection on beta drivers would be okay for debugging purposes if the WHQL drivers allowed the GPU to function correctly in games.

Your claim was that "NVIDIA is a software company"

Because they are, and have been for a long time. Their predominant retail customer isn't one that wants to buy GeForce, their predominant retail customer is one that needs CUDA. CUDA is a software ecosystem with a hardware lock around it.

What's the difference between a class RTX card and a consumer GeForce class RTX card with the same chip? Software. What allows one card to support 10-bit precision and another 12-bit despite being the same architecture? Software. NVIDIA creates more proprietary software ecosystems than physical hardware generations.

Take a moment to browse their open job listings and see that for every one hardware/network/ASIC engineer position they post, there are 15-20 software engineer positions open. This isn't new.

 

[Luke357]

OK, let's do an on topic question then. How does browsing behavior helps the development of the drivers? Are future Intel GPUs going to be optimized for Instagram if people spend too much time on Instagram? Maybe a custom accelerator?


Most people will press "Yes, next, next, next, yes".

Most people will just skip through it yes, but anyone with half a brain should atleast check the check boxes.

 

[MarsM4N]

Just a heads up: the Intel "Compute Improvement Program" (CIP) is also bundled in the Intel "Chipset INF Utility", a tool to auto update all your Intel drivers! No option to de-select present.

The data collection program is starting randomly and is eating quite some system resources, which could lead to lags in games. It is also sending out tons of data, which is quite bad if you're on a data cap. Luckily you can de-install the backpacked program in the Windows control panel, which you have to repeat each time when you update the tool.