Discussion AMD failure in notebooks really was due to giving igps priority over cpu?

[Shivansps]

Ok i wanted to make this thread to discuss about this matter ever since people started chearing when AMD cut down the IGP in Renoir with some people thinking that the idea was giving "CPU higher priority because IGP priority gave nothing to AMD".

Im going to go back several years to the first notebook i ever had, the MSI U230, this was the time that in small factors your choices were a increible slow Atom CPU+useless IGP, the ION platform (Intel Atom + Nvidia chipset with a OK igp), later replaced with the ION2 platform when Intel did not allowed Nvidia to make chipsets anymore, ION2 was Intel Atom+Intel chipset+Nvidia ION2 gpu in a x1 PCI-E connection used via optimus, and finally AMD Yukon platform, with AMD Neo CPUs (it were the same CPUs but with lower clocks and voltage).
The MSI U230 as one of the best you could get in the Yukon platform, with a dual core L335 and a RS780 chipset with a HD3200 IGP, AMD was having far more CPU perf than the ATOMs, and the HD3200 was hand to hand with the IONs, because the HD3200 was a little slower than Nvidia ION GPUs(specially ION2 that had VRAM), but the CPU perf difference was way too high, so in gaming the HD3200 was better, specially with dual channel, only the ION2 was at top a few times.

Then AMD came up with i think it is the worst they ever did, the small cores. The first APU was the E-350 that was a Dual Core with Bobcat cores and a integrated HD6310 (it was a HD5450 in a IGP), but with single channel ram.
I went ahead and brought one, a HP DM1z, i was planning to give my U230 away and replace it the DM1z... You guys have no idea how dissapointed i was with the DM1z, the Bobcat cores were slower than the ones in the L335, and the IGP was held back by the single channel ram, resulting only in a very small jump gaming in performace. Only batery life were about 1 hour longer.
And thats was the last notebook i owned, by that time i was already working in a computer distrubutor with access to a lot of notebooks and general hardware.

Everything went downhill there, Intel did a far better job in keeping OEMs from using the worthless Atom off the mainstream notebooks(until Baytrail), but AMD small cores were not restricted to small factors and started to go in every type of notebook, this damaged AMD image A LOT, because performance was BAD, really, really bad, even in IGP has Intel HD3000 was outperforming the 80CU IGP in the small core APUs. The small cores are a problem even today as unsold Carrizo-L and Stoney Ridge notebooks models are still around, even with a SSD these things feel slow. And SSD arent common on those models.

And that was not the only problem i saw, the number of small core notebooks was hidding the big core AMD APU models, the big core APU models became rare, but they did a lot better in performance vs the Intel notebooks, but even big cores APUs had both CPU and GPU performance problems vs Intel, as well as using more power. At the end AMD did not had a clear GPU lead over Intel in big cores in notebooks either.

This changed only with the first Ryzen Mobile APUs, but they still had turbo and power efficiency issues that AMD fixed in Renoir.
So what i think is was never about giving the IGP priority, it was about the mistakes they did with the small cores, and they never had a clear lead on the mobile big core APUs either.

 

[yuri69]

In the past, AMD had horrible CPU performance combined with horrible battery life. This effectively limited them to the bottom-of-the-barrel laptops. Would you base a premium product on a horribly outdated, underperforming, and hot APU?

Bobcat/Jaguar aimed only at "netttops", Llano put K10 cores vs Sandy Bridge, Trinity featured super low IPC Piledriver cores vs Ivy Bridge, Kaveri got broken firmware, etc. The battery life finally got better with Carrizo and Bristol - these got an advanced power management.

Raven was meh with the battery life, but Renoir perfected that vs 14 and even 10nm Intel. So yea, AMD could finally go to premium notebooks since Renoir was way better than Intel.

 

[ElFenix]

In the past, AMD had horrible CPU performance combined with horrible battery life. This effectively limited them to the bottom-of-the-barrel laptops. Would you base a premium product on a horribly outdated, underperforming, and hot APU?

Bobcat/Jaguar aimed only at "netttops", Llano put K10 cores vs Sandy Bridge, Trinity featured super low IPC Piledriver cores vs Ivy Bridge, Kaveri got broken firmware, etc. The battery life finally got better with Carrizo and Bristol - these got an advanced power management.

Raven was meh with the battery life, but Renoir perfected that vs 14 and even 10nm Intel. So yea, AMD could finally go to premium notebooks since Renoir was way better than Intel.

/thread

 

[Hitman928]

When the asia cyber cafe issue came out, Valve said it was once per year, and to me it has been once per year so far.

Valve does not give the exact information on how the system works but they said once per year to each user, and each month a random numbers of users are asked to participate.

What i dont know is what happens if you refuse or if you miss the window, if you asked again in what time or never again.

If you have no idea what the methodology is, how can you put any trust in the survey? The once per year per user is definitely not accurate or at least does not work the way they say it does, I've seen way too many people comment on their survey experience that contradicts this, myself included.

 

[Markfw]

If you have no idea what the methodology is, how can you put any trust in the survey? The once per year per user is definitely not accurate or at least does not work the way they say it does, I've seen way too many people comment on their survey experience that contradicts this, myself included.

See my post, last one on the previous page. I am agreeing with you, but look at the statistical evaluation.

 

[Shivansps]

If you have no idea what the methodology is, how can you put any trust in the survey? The once per year per user is definitely not accurate or at least does not work the way they say it does, I've seen way too many people comment on their survey experience that contradicts this, myself included.

I think you want a data source that is fully representative of the users hardware, the steam survey is not that, it is to give developers an idea of the hardware avalible in the platform. Im petty sure they dont care that much about it.

You need to take as what it is, a poll that gives you an idea of what is going on.


But i really dont want to discuss more about it anymore, if people want to dismiss the only source of information avalible that is backed up by the fact Nvidia always made low end mobile gpus thats fine by me.

 

[scannall]

Also not sure why everyone is dismissing the big core mobile APUs, AMD had mobile versions for every one of the big core APU they did, this is a mobile Llano, the first big core APU:

The AMD Llano Notebook Review: Competing in the Mobile Market

www.anandtech.com

This was a few months after Bobcat launch, as everyone can see AMD Llano was a good product for notebooks at its time. So what exactly happen after this? Because everyone started using Bobcat instead.

I still use a laptop with the Trinity A8-4500m. It still works great really. From the factory it was pretty gutted. 1 slow, small DIMM for RAM, and a slow as dirt hard drive. The out of the box experience was pretty bad.

But, being a packrat I had RAM to fill the slots and bring it to 16 gig, and a very nice SSD to replace the spinning rust. It was like a completely different laptop after those two changes.

 

[tamz_msc]

Most surveys that require the most accurate answers have preemptive questions that filter out bad results. This is the norm when taking a survey, at least in higher education. This is called a "screener."

This survey doesn't ask a set of complicated questions that requires the respondent to answer truthfully and sincerely. It just collects data that's already available and just requires consent to send it over.

Power users typically care about privacy. While data is likely anonymized, people who are more technically inclined prefer to remain as anonymous as they can, and will choose to limit this data.

A fraction of the users not opting to send the data doesn't change anything - it just adds another layer of randomness to the data.

It makes a world of a difference. The survey doesn't see a mobile variant of a processor. It sees cores and threads. If the distinguation was not relevent to the premise of the thread neither was your bringing up the use of laptops in casual gaming.

The OP mentioned the Steam HW survey in the context of the widespread adoption of Intel HD 4000 graphics compared to AMD's integrated graphics of the time. In that context the survey is relevant in order to understand market penetration.

Ah, so the age does not matter. Do you not think it's not an accurate representation with dual cores are from 2006 may be counted in along with new dual cores, or quad cores from 2007 along with modern quad cores? Or AMD's old 6 core systems that are more than twice as slower performing than modern 6 core processors by AMD? All you get is a very generic idea of what's being used. The minute you see more data than what's provided you can accurately paint a picture.

Old or new what difference does it make? There are people who are playing with their older systems as well as those who play with newer systems. Again, it's about what Steam's users play with. As far as CPUs are concerned, there is an additional set of data on instruction set support. That tells you the extent of new vs old hardware. For example 93% of users have a CPU with AVX support. That's Sandy Bridge and Bulldozer or newer. So most of the system requirements of newer games mention Sandy Bridge and Bulldozer CPUs. Steam HW survey data is useful to developers in that regard.

Mindfactory is a small German etailer. That's sales data, not active use data. We're looking for lone systems with the intended hardware. Not 4-5 components purchased and used in various computers. You'd still want an accurate idea of what people are using while accessing your gaming network.

The only viable data Steam offers up are gaming resolution and video cards in use. CPU speeds and CPUs (cores) are unhelpful.

See point above.

 

[A///]

This survey doesn't ask a set of complicated questions that requires the respondent to answer truthfully and sincerely. It just collects data that's already available and just requires consent to send it over.

Are you just repeating yourself here? We know how the survey works. It's why we're telling you it's bunk. That's not how a quality survey works. There is zero quality control involved.

A fraction of the users not opting to send the data doesn't change anything - it just adds another layer of randomness to the data.

A fraction? Where did you get that figure from? Not only does Valve not release the total number of respondents, that already invalidates the survey's quality, but also do not state how many abstain from the survey. There is no control for the survey either if you wanted to make it highly technical and use it in a study to examine hardware trends over a certain period of time. If users abstain, they don't add another layer of randomness to the data; they're not counted in that data.

The OP mentioned the Steam HW survey in the context of the widespread adoption of Intel HD 4000 graphics compared to AMD's integrated graphics of the time. In that context the survey is relevant in order to understand market penetration.

The OP also makes questionable posts and threads to antagonize users. It doesn't mean the OP is using common sense when making a thread.

Old or new what difference does it make? There are people who are playing with their older systems as well as those who play with newer systems. Again, it's about what Steam's users play with. As far as CPUs are concerned, there is an additional set of data on instruction set support. That tells you the extent of new vs old hardware. For example 93% of users have a CPU with AVX support. That's Sandy Bridge and Bulldozer or newer. So most of the system requirements of newer games mention Sandy Bridge and Bulldozer CPUs. Steam HW survey data is useful to developers in that regard.

To quantify accurate date you need a lot of data points. A six core processor from 2010 performs a lot differently than one in 2020, unless you overclock the former to its limits and are practically playing with fire. You can't point to the surveys over time and state, "Yes, this is where users began shifting away from Core 2 Duo processors to newer quad core processors in 2011, however, it seems we had these users who are fringe and were using hyperthreaded quad cores in 2009. These were an expensive platform at the time. The majority of users in 2011 to 2014 were quad core users without hyperthreading. This changed later on as usescases increased in complexity. In 2017 we saw a double increase to eight core processors due to AMD's Ryzen arriving on the scene."

But, no, all you get is lame static figures that mean nothing. The frequencies mean nothing at all either. A slower Ryzen outperforms a faster Intel in a lot of workloads except gaming, where Intel is king, and mostly due to core to core latencies.

See point above.

Learn to comprehend the above.

 

[tamz_msc]

Are you just repeating yourself here? We know how the survey works. It's why we're telling you it's bunk. That's not how a quality survey works. There is zero quality control involved.

Nope, you provide absolutely no rationale to justify why the survey is bunk. The survey collects anonymized data and there is no scope for 'quality control' in such data.

A fraction? Where did you get that figure from? Not only does Valve not release the total number of respondents, that already invalidates the survey's quality, but also do not state how many abstain from the survey. There is no control for the survey either if you wanted to make it highly technical and use it in a study to examine hardware trends over a certain period of time. If users abstain, they don't add another layer of randomness to the data; they're not counted in that data.

Steam collects hundreds of thousands of data points, given that there are nearly 20 million peak daily active users. Even if 90% of people opt out(which I figure is on the higher side), that's tens of thousands of data points which is enough for a representative sample.

To quantify accurate date you need a lot of data points. A six core processor from 2010 performs a lot differently than one in 2020, unless you overclock the former to its limits and are practically playing with fire. You can't point to the surveys over time and state, "Yes, this is where users began shifting away from Core 2 Duo processors to newer quad core processors in 2011, however, it seems we had these users who are fringe and were using hyperthreaded quad cores in 2009. These were an expensive platform at the time. The majority of users in 2011 to 2014 were quad core users without hyperthreading. This changed later on as usescases increased in complexity. In 2017 we saw a double increase to eight core processors due to AMD's Ryzen arriving on the scene."

But, no, all you get is lame static figures that mean nothing. The frequencies mean nothing at all either. A slower Ryzen outperforms a faster Intel in a lot of workloads except gaming, where Intel is king, and mostly due to core to core latencies.

It means nothing because you say so? Sure, believe whatever you want to believe.

 

[A///]

Nope, you provide absolutely no rationale to justify why the survey is bunk. The survey collects anonymized data and there is no scope for 'quality control' in such data.

I did provide rationale. You've repeatedly chosen to ignore it and are now trolling. Anonymized date is imperfect and doesn't provide a clear picture of what is actually happening. If there is no scope for quality control then the data cannot be validly applied to any argument. If the data isn't clear cut then your argument falls face first. At best, the data can suggest resolution and GPU in use. Beyond that, it's useless. And even then, it doesn't state how that resolution is gathered. Game level, system level or hardware level.

Game: In game resolution
System: Setting in Windows/Linux/macOS
Hardware: What the hardware is capable of

How can you state that the survey data represents a large degree of the gaming population that uses Steam, then state that the data is imperfect? You torpedoed your original argument.

Steam collects hundreds of thousands of data points, given that there are nearly 20 million peak daily active users. Even if 90% of people opt out(which I figure is on the higher side), that's tens of thousands of data points which is enough for a representative sample.

Hundreds of thousands of data points per individual? Can you support that claim with any official documentation stating as such?

Are you referring to this site: https://steamdb.info/app/753/graphs/

?

You've only shown peak user data. You have yet to explain that data into how it's being counted. In other words, an internet cafe would be included in that chart. Chinese, emphasis on Chinese, internet cafes were excluded from Steam surveys by Valve. Valve never disclosed how they manage to find this in their data to exclude it. Even then, you're still willingly ignoring how flawed the data is in its most raw form.


You can click each area on the survey to see the breakdown.

OS: VMs can and do count.
RAM: Correct... to a point. If it broke down into being able to collect DDR generation you can narrow done the age of equipment.
Intel CPU speeds: Means zilch. IPC increases and mild clock increases deliver more performance for less. Also plays with the core count issue.
Physical CPUs: Core counts. Doesn't break down into whether CPUs are HT/SMT or not. Additionally, Core counts remain the same over the years. Lynnfield was the first quad core of the Core i era. 4/4 processors on i5. The i7s were Bloomfield 4/8s. Neither range of processors would be identified as such in the survey. Neither perform anything like modern equivalents. They must be clocked sky high to have parity.

It means nothing because you say so? Sure, believe whatever you want to believe.

No, because I've been where you were and then informed myself. At the time I hadn't looked into how flawed the survey data was.

 

[tamz_msc]

I did provide rationale. You've repeatedly chosen to ignore it and are now trolling. Anonymized date is imperfect and doesn't provide a clear picture of what is actually happening. If there is no scope for quality control then the data cannot be validly applied to any argument. If the data isn't clear cut then your argument falls face first. At best, the data can suggest resolution and GPU in use. Beyond that, it's useless. And even then, it doesn't state how that resolution is gathered. Game level, system level or hardware level.

Where's your rationale? You and a bunch of others here just believe the data is imperfect because its random, anonymized and opt-in. None of those characteristics invalidate the data.

Game: In game resolution
System: Setting in Windows/Linux/macOS
Hardware: What the hardware is capable of

How can you state that the survey data represents a large degree of the gaming population that uses Steam, then state that the data is imperfect? You torpedoed your original argument.

Now this tells me you've either never used Steam or are pretending to not know how the data is gathered. And where did I state that the data is imperfect? It's you and a bunch of others here who are making that claim.

Hundreds of thousands of data points per individual? Can you support that claim with any official documentation stating as such?

Lol, why would it be hundreds of thousands times per user? It's hundreds of thousands of individual users who are surveyed each month, and each user is surveyed once per year.

https://www.reddit.com/r/hardware/comments/bbvnb3

There was another thread where someone had emailed Valve on the number of users surveyed, and that number runs into hundreds of thousands, but I can't seem to find that thread now.

 

[A///]

Where's your rationale? You and a bunch of others here just believe the data is imperfect because its random, anonymized and opt-in. None of those characteristics invalidate the data.

I suggest you take your sweet time to do some research before arguing. I made the same comment to Mark a year ago and was told to look into it and figure out why that data is flawed.

Now this tells me you've either never used Steam or are pretending to not know how the data is gathered. And where did I state that the data is imperfect? It's you and a bunch of others here who are making that claim.

I've been a Steam member since 20010 or 2011. I have roughly 80 games purchased through it. I've posted about Steam before, and even recommended the survey once before I was asked to look into it and never mention it again for my own sanity. Page 3 on Zen 4 thread and the next few pages.

Lol, why would it be hundreds of thousands times per user? It's hundreds of thousands of individual users who are surveyed each month, and each user is surveyed once per year.

I never said it was. I never implied. You made a statement that was unclear. As to your image:

1. CPU brand is Intel or AMD
2. CPUID instruction tells the OS what is being run and it leaks this data to software that can look it up.
3. He's bringing up "Skylake" as an example, but none of this is exposed on the public survey.

 

[tamz_msc]

I suggest you take your sweet time to do some research before arguing. I made the same comment to Mark a year ago and was told to look into it and figure out why that data is flawed.

I've been a Steam member since 20010 or 2011. I have roughly 80 games purchased through it. I've posted about Steam before, and even recommended the survey once before I was asked to look into it and never mention it again for my own sanity. Page 3 on Zen 4 thread and the next few pages.



I never said it was. I never implied. You made a statement that was unclear. As to your image:

1. CPU brand is Intel or AMD
2. CPUID instruction tells the OS what is being run and it leaks this data to software that can look it up.
3. He's bringing up "Skylake" as an example, but none of this is exposed on the public survey.

Yup, just as before you fail to come up with any plausible argument as to why the survey is flawed.

 

[A///]

Yup, just as before you fail to come up with any plausible argument as to why the survey is flawed.

The way I'm interpreting this post is you've given up because you have no way of winning the debate. Have a nice day, sport.

 

[tamz_msc]

The way I'm interpreting this post is you've given up because you have no way of winning the debate. Have a nice day, sport.

Lol you pointed me to a thread that in turn points to another thread that is nearly ten years old with posters who are no longer active arguing about the survey because it doesn't suit their beliefs. That's quite some 'research' you've done.

 

[A///]

Lol you pointed me to a thread that in turn points to another thread that is nearly ten years old with posters who are no longer active arguing about the survey because it doesn't suit their beliefs. That's quite some 'research' you've done.

Well, argue with other people. I used to have the same views as you until I informed myself. Valve has made valiant efforts ever since they began collecting data to make it as accurate as possible without encroaching on user privacy and privacy laws around the globe. Fact is, you can't generate remotely accurate data. There is also a bug or maybe was a bug where it counted individual user logins (not logging in and out, but a set point when the data gets collected) where the same computer may send data multiple times as the data isn't tied to the individual machine. This bug affected Intel CPUs according to Scott Herkelman at AMD.

Not trying to be rude here. Google out "steam hardware survey flawed" and you'll see reports spanning years up until recently of how bad the data is. Which really sucks because Valve is an incredible company that changed how we buy and consume games, and probably the most trusted platform out there. Epic is sketchy and Microsoft's store is... run by orangutans.

 

[VirtualLarry]

Just a note: I believe that threads revolving around "Steam User Survey" are verboten in the Graphics sub-forum, so IMHO, for the same reasons, it should be verboten in the CPUs & Overclocking forum as well. Just IMHO.

 

[tamz_msc]

Google out "steam hardware survey flawed" and you'll see reports spanning years up until recently of how bad the data is.

That's not research. That's you succumbing to confirmation bias based on what a couple of posters told you over here. I've seen countless search results but none of them have soundly argued why the data is flawed. This is excluding those instances where Steam counted Chinese iCafe users multiple times.

 

[Kenmitch]

That's not research. That's you succumbing to confirmation bias based on what a couple of posters told you over here. I've seen countless search results but none of them have soundly argued why the data is flawed. This is excluding those instances where Steam counted Chinese iCafe users multiple times.

You can argue all you want, but in the end nobody really cares about the steam survey. No need to keep babbling on about it.

Do you have the last word syndrome?

 

[A///]

That's not research. That's you succumbing to confirmation bias based on what a couple of posters told you over here. I've seen countless search results but none of them have soundly argued why the data is flawed. This is excluding those instances where Steam counted Chinese iCafe users multiple times.

This is you succumbing to your own dunning kruger effect.

Do you have the last word syndrome?

Quite positively, along with word diarrhea. He's repeated himself eight times, I think, and keeps ignoring any explanation. It's okay, he's a genius. Literally like a certain European expat we all know and love.

 

[Markfw]

This is you succumbing to your own dunning kruger effect.


Quite positively, along with word diarrhea. He's repeated himself eight times, I think, and keeps ignoring any explanation. It's okay, he's a genius. Literally like a certain European expat we all know and love.

And a technical explanation from a statistics person (me) explained it, and he ignored that.

 

[tamz_msc]

And a technical explanation from a statistics person (me) explained it, and he ignored that.

You call your explanation technical? Thanks for the laugh.

This is you succumbing to your own dunning kruger effect.


Quite positively, along with word diarrhea. He's repeated himself eight times, I think, and keeps ignoring any explanation. It's okay, he's a genius. Literally like a certain European expat we all know and love.

Keep believing your own delusions.

You can argue all you want, but in the end nobody really cares about the steam survey. No need to keep babbling on about it.

Do you have the last word syndrome?

You mean nobody on this forum? Outside of this bubble which pretends to know everything about the survey, there's plenty who care. Game Devs for example.

 

[A///]

You keep insulting people. I found this interesting. Coincidentally found this through Google in a reddit page regarding the flaws of the Steam survey. You've yet to provide an in depth explanation as to why the date provided in the survey can be used to examine market trends when your original claim was 20M users use Steam, except the below image shows that not only do they not target 20M users a month, but there's no confirmation of random testing.

 

[Shmee]

Guys, we should really drop the steam hardware survey arguments, or take it to another thread. It could be a legitimate discussion I feel, but this is getting ridiculous and derailing the thread badly.

 

[PlanetJosh]

I have to agree. Over the years, I have known quite a few gamers. And when I hosted lan parties for my son ? (and they had to bring their computers) one out of 15 had a laptop.

I was worried I was missing out on laptop gaming but I see from this and other posts it's not nearly as popular as I thought.