Snapdragon Summit 2023 - Qualcomm Announces a Slew of New Processors - Updated with Day 2

[Dracarris]

7 minutes ago, LAwLz said:

yes we have to assume that the y-axis is linear.

Not only that, but we have to assume that bottom left is zero for both y and x axis. So without even a zero marking (could be anywhere "below the table"), this pretty much remains useless.

[WolframaticAlpha]

The SDM 8g3 also seems very interesting. While still lagging behind the A17 Pro on ST, it is 4-6% faster in MT on geekbench in some Xiaomi phone. It also has quite high clocks(3.1-3.3 Ghz). 

https://beebom.com/snapdragon-8-gen-3-vs-apple-a17-pro/amp/

Considering that the chip has been estimated to be 10-12% faster, Qcomm might finally be closing the gap(although they are held back by their dogshit update policy)

 

MTL, M3, Oryon and maybe zen5... 2024 is shaping up to be a massively interesting year for laptop processors

[NavyCobra1417]

These new chips sound promising, hopefully they don't disappoint when we can get them in some products.

[LAwLz]

3 hours ago, Dracarris said:

Not only that, but we have to assume that bottom left is zero for both y and x axis. So without even a zero marking (could be anywhere "below the table"), this pretty much remains useless.

Except it's not useless as I demonstrated in the next few sentences in my post. 

Can we please stop with the overreacting? 

 

 

Anyway, day 2 of Snapdragon summit included some more details which I'll add to this thread later. 

 

Some of the announcements are:

There are three groups of 4 cores in the Elite X. 

 

The NPU will show up in Task Manager, alongside the CPU and GPU usage graphs. 

 

Lots of devices and partners shown. 

 

Some more benchmarks (although a lot seem to be CPU vs NPU which I guess is nice for real world performance, but not to compare CPU vs CPU). 

 

The Snapdragon 8 gen 3 will deliver about 10% better efficiency overall. 

 

The camera processing is getting a major upgrade. 

They also confirmed that they are working with Samsung to make it support a 200MP sensor. Galaxy S24 Ultra maybe? 

[Dracarris]

1 hour ago, LAwLz said:

Except it's not useless as I demonstrated in the next few sentences in my post. 

No. As long as we can't be sure that the datapoints/shape of the curves outside of the labeled point is accurate, your method of drawing a horizontal line and inferring "same performance at x Watt" is flawed and invalid. And we have no clue how they arrived at said shape.

[LAwLz]

11 hours ago, Dracarris said:

No. As long as we can't be sure that the datapoints/shape of the curves outside of the labeled point is accurate, your method of drawing a horizontal line and inferring "same performance at x Watt" is flawed and invalid. And we have no clue how they arrived at said shape.

Please explain how my method is flawed.

[Dracarris]

Just now, LAwLz said:

Please explain how my method is flawed.

I already did. Unless we can be sure that the curves outside of the few labeled points are accurate and not just a nice looking shape that fits said points, you will end up with arbitrarily wrong data when drawing horizontal lines in the way you explained.

[LAwLz]

1 minute ago, Dracarris said:

I already did. Unless we can be sure that the curves outside of the few labeled points are accurate and not just a nice looking shape that fits said points, you will end up with arbitrarily wrong data when drawing horizontal lines in the way you explained.

The same can be said for literally every single graph in existence, even if the axes are labeled. I dare you to show me any graph where this isn't the case.

 

This is a very silly argument to make.

[Dracarris]

3 minutes ago, LAwLz said:

The same can be said for literally every single graph in existence, even if the axes are labeled. I dare you to show me any graph where this isn't the case.

 

This is a very silly argument to make.

Yeah, no. Like not at all.

 

With a graph with labeled y-axis, data can be expected to be accurate, or otherwise someone plainly lies to you. They are literally saying at point x, data is y.

Without a labeled y-axis, data can literally be anything. Aka marketing. There is absolutely no commitment to what y is at a given x, outside of the very few labeled points.

[LAwLz]

7 minutes ago, Dracarris said:

Yeah, no. Like not at all.

 

With a graph with labeled y-axis, data can be expected to be accurate, or otherwise someone plainly lies to you. They are literally saying at point x, data is y.

Without a labeled y-axis, data can literally be anything. Aka marketing. There is absolutely no commitment to what y is at a given x, outside of the very few labeled points.

Yes, the data on the y-axis can be anything, but using my methodology that doesn't matter. The only thing that matters is that the x-axis is labeled because we only use the y-axis as a reference to find a common point of comparison. Whatever that point happens to be on the y-axis is irrelevant.

 

Do you not understand how graphs work?

 

 

Here is the graph:

 

 

Here is an example of what I was talking about:

 

By drawing a line straight out from the y-axis and then straight down, we can see that:

"The Snapdragon at ~9 watts gets the same relative performance as the 7940HS gets at around 17 watts".

 

Whatever the y-axis is in absolute numbers is irrelevant. Since the line goes straight out we know that the number is the same for both the Ryzen and Snapdragon chip. The only number that changes between the two chips is the power consumption number, and that we do actually have labeled.

It does not matter at all what the Y-axis means, how it scales, which number it starts at etc, because we use the same number and it is only used as a reference in order to make a comparison between the values on the X-axis.

 

We are comparing performance:watt and we set the "performance" number to be the same on both.

[Dracarris]

5 minutes ago, LAwLz said:

Yes, the data on the y-axis can be anything, but using my methodology that doesn't matter. The only thing that matters is that the x-axis is labeled because we only use the y-axis as a reference to find a common point of comparison. Whatever that point happens to be on the y-axis is irrelevant.

 

Do you not understand how graphs work?

You don't understand how graphs work. Obviously.

 

Heck no, data on y CANNOT be anything, do you even think for a split second about what you just typed there? It's nonsense. If y-data is "anything", the curves end up in different places on the graph, and your genius mehtod fails utterly because the shifted curve means you'll end up at random points on the x-axis! A curve establishes a relationship between x and y, so hell no, y matters. A lot.

 

The only data change your method is immune to is shifting all curves (both in this case) by a fixed amount in the y direction. That's not exactly "y can be anything".

 

Stop lecturing me about things you clearly don't fully understand and stop believing marketing crap from companies.

[LAwLz]

Just now, Dracarris said:

You don't understand how graphs work. Obviously.

 

Heck no, data on y CANNOT be anything, do you even think for a split second about what you just typed there? It's nonsense. If y-data is "anything", the curves end up in different places on the graph, and your genius mehtod fails utterly because the shifted curve means you'll end up at random points on the x-axis! A curve establishes a relationship between x and y, so hell no, y matters. A lot.

 

Stop lecturing me about things you clearly don't fully understand and stop believing marketing crap from companies.

Holy shit this is unbelievable...

 

Check my edited post, please. I added pictures to make it easier for you to understand.

[Dracarris]

2 minutes ago, LAwLz said:

Holy shit this is unbelievable...

 

Check my edited post, please. I added pictures to make it easier for you to understand.

Yeah, it's unbelievable that you still think you are in the right here.

 

If any of the curves is slightly bent to the left or right, or somehow else different, aka, not showing accurate y-data, you end up at all different points on the x-axis! For gods sake, how do you fail to see this?

 

I fully understand your method. It works well and is well established if we can trust the y-data. This is not the case here.

 

Believe whatever you want. You are wrong here.

[LAwLz]

Just now, Dracarris said:

Yeah, it's unbelievable that you still think you are in the right here.

 

If any of the curves is slightly bent to the left or right, or somehow else different, aka, not showing accurate y-data, you end up at all different points on the x-axis! For gods sake, how do you fail to see this?

 

I fully understand your method. It works well and is well established if we can trust the y-data. This is not the case here.

 

Believe whatever you want. You are wrong here.

The only way for your argument to make any sense is if they used different y-axis numbers for the Snapdragon and AMD chip.

For example "0,10 on Snapdragon means 100 performance, and 0,10 on Ryzen means 200 performance". But if that is the argument you are making then that does in fact literally apply to every single graph. Even if they had labeled the y-axis we can't actually be sure that they didn't use different y-axis values for the two data sets and decided to only show one of them. 

 

We humans have to make assumptions about certain things, and I find it absolutely asinine to assume that the values on the y-axis have different scales for the various vendors. Again, this is as silly as saying "we can't be sure W stands for watt because it could stand for whatever-qualcomm-wants-it-to-mean".

 

Also, labeling something as "useless" because we can't with 100% certainty read it without making some very normal assumptions (like W standing for watt, or one coordinate meaning the same value for both the data sets) is moronic.

[Dracarris]

6 minutes ago, LAwLz said:

The only way for your argument to make any sense is if they used different y-axis numbers for the Snapdragon and AMD chip.

For example "0,10 on Snapdragon means 100 performance, and 0,10 on Ryzen means 200 performance". But if that is the argument you are making then that does in fact literally apply to every single graph. Even if they had labeled the y-axis we can't actually be sure that they didn't use different y-axis values for the two data sets and decided to only show one of them. 

Yeah, no. As soon as any of the curves is not accurate, you end up in different intersection points with the x-axis. In your pictures above, simply bend any of the curves a bit, make it a bit steeper or less steep, then draw your horizontal line in an affected/changed area. You'll end up at different point(s) on the x-axis and draw wrong conclusions from there.

 

You are right if only the y-labels are missing but otherwise accurate data is given. And we don't know that here.

 

This has nothing to do with assuming 'W' stands for Watt.

 

Maybe others want to jump in and explain it more clearly to you, I've wasted enough time on this nonsense already.

[LAwLz]

4 minutes ago, Dracarris said:

As soon as any of the curves is not accurate, you end up in different intersection points with the x-axis.

Yes, but now you're making the argument that they faked the numbers (or used different scales for Snapdragon and AMD) and therefore we can't trust the chart. We have to make the assumption that the numbers are accurate because we have to make that assumption for every graph.

You can't just say "yeah but if we redraw one of the lines then the numbers will be different". Of course the numbers will be different if we change one of the lines. The same goes for all graphs.

[Dracarris]

Just now, LAwLz said:

You can't just say "yeah but if we redraw one of the lines then the numbers will be different". Of course the numbers will be different if we change one of the lines. The same goes for all graphs.

We're turning in circles, now:

28 minutes ago, Dracarris said:

With a graph with labeled y-axis, data can be expected to be accurate, or otherwise someone plainly lies to you. They are literally saying at point x, data is y.

Without a labeled y-axis, data can literally be anything. Aka marketing. There is absolutely no commitment to what y is at a given x, outside of the very few labeled points.

Everything has been said.

[LAwLz]

Just now, Dracarris said:

We're turning in circles, now:

Everything has been said.

We can't expect the data to be accurate if the y-axis is labeled. You're just making an assumption that it is. The numbers could be entirely fabricated, or the y-axis might only represent the numbers for one of the lines, while the other line uses a different y-axis scale that is not shown on the graph.

If you look at the graph with the mindset that you have to find ways in which it could be inaccurate then you will always be able to find them, no matter how many labels there are on the graph. If you look at a graph with the mindset that it is trustworthy then you always have to make some assumptions about it.

 

Just because we have to make some assumptions does not mean it is useless, because that will always, every single time, be the case.

[Dracarris]

1 minute ago, LAwLz said:

We can't expect the data to be accurate if the y-axis is labeled.

Yes we can. Showing wrong data with a labeled y-axis is deceptive, fraught, lying, and making an author susceptible to legal consequences.

 

Showing "wrong" data without a labeled y-axis is marketing. "Wrong" because without a label there isn't really a way to even establish whether data is wrong or not. Curves can in this case simply be for illustrative purposes with zero commitment to the actual data points, outside of what is manually labeled.

 

You do you think companies keep going for unlabeled axis/graphs?

[LAwLz]

4 minutes ago, Dracarris said:

Yes we can. Showing wrong data with a labeled y-axis is deceptive, fraught, lying, and making an author susceptible to legal consequences.

 

Showing "wrong" data without a labeled y-axis is marketing. "Wrong" because without a label there isn't really a way to even establish whether data is wrong or not. Curves can in this case simply be for illustrative purposes with zero commitment to the actual data points, outside of what is manually labeled.

 

You do you think companies keep going for unlabeled axis/graphs?

How do you know the y-axis on a labeled chart actually applies to both products being compared? "Because it would be deceptive" is not exactly a great answer because the same can be said about using different scales for two products on an unlabeled y-axis.

 

It is a very simple, normal and reasonable assumption to make that the scale on one axis is the same for both products being compared.

 

As for why companies use unlabeled axis and graphs, it's because they want to shift the focus from absolute numbers (probably because they might not be that great) and make direct comparisons harder. They want us to only be able to make the comparisons they want. By not giving us the absolute performance numbers, we are stuck with the handful of comparisons they gave us. If they had given us the absolute performance numbers we could have made our own comparisons.

I very much doubt it is because they want to use different y-axis scales for the various products because, again, they can do that while still labeling the graphs. It is not that uncommon to do just that in fact. 

 

But if you want to use different scales for the y-axis depending on which thing you are measuring, you put the labels on both sides, like this:

 

In this case, we actually do have a label on the left side. As a result, we can be 99% sure that it is the same scale for both products. They could still be lying, but we are in fact in the same situation as if we look at a graph with more proper labeling.

What is missing are the numbers for the y-axis, not what the axes measure (W and absolute performance).

[leadeater]

1 hour ago, LAwLz said:

By drawing a line straight out from the y-axis and then straight down, we can see that:

"The Snapdragon at ~9 watts gets the same relative performance as the 7940HS gets at around 17 watts".

 

Whatever the y-axis is in absolute numbers is irrelevant.

Without knowing the y axis scale and that it's actually linear then we can't actually reliably do that. And that's the problem, the only thing I can trust about the graph is the text next to it stating what they have, the rest of it could be anything at all.

 

And the reason it does actually matter is the graphed line is trying to present a picture, an idea, an indication of the difference between the products but if I don't know the scale then I don't know if the graph is proportionately depicting the difference or not. Having it not labelled makes it really easy to "embellish for effect" rather than it be an accurate depiction. That's the problem I have with it.

 

Also without fixed data points along the line I don't know if the line is even accurate at all. Typically you have points along a curve with known data points and the line between them is inferred values. The whole damn line is inferred as far as we know.

 

And I know it's not really that important but an awful graph is an awful graph and if a company isn't going to do it properly then don't do it at all. Use something else.

 

1 hour ago, LAwLz said:

it's because they want to shift the focus from absolute numbers (probably because they might not be that great) and make direct comparisons harder. They want us to only be able to make the comparisons they want.

 

1 hour ago, LAwLz said:

is not exactly a great answer

Liberties were taken with these quotes, just to point out your answer is also not a great answer either. 

 

Let me put it this way, if we take your 17W data point and make the Y value 75% and the upper data point is 80% making everything in-between 5% what does that portray? Now change that Y value to 50%, how about now? So does actually knowing what the Y axis is matter? I say yes. 

 

Line graphs axis don't have to start at zero and don't have to be linear scales to be valid graphs, frequency response graphs are a very good example of this.

 

I would like to thank Intel for actually doing it

This is the standard I expect if such a company wants to use a line graph to compare market options.

[LAwLz]

The Snapdragon 8 gen 3 is the first flagship SoC from Qualcomm that completely lacks Aarch32 support. In other words, it's a 64-bit chip only. 

 

However, the Xiaomi 14 is still able to run 32-bit apps because they are licensing an Aarch32 to Aarch64 translation layer called Tango from the company Amanieu Systems. 

 

 

 

[LAwLz]

On 10/27/2023 at 11:22 AM, leadeater said:

Without knowing the y axis scale and that it's actually linear then we can't actually reliably do that.

Yes you can.

If the y-axis is in the same place for both products being compared, it will be the same value. The only time this is not the case is if the graph is either incorrectly made, or if there is a different scale written on the other side of the graph.

 

If we start at 0,0 on the graph and then move 300 pixels straight up, we know that the value of the y-axis will be the same for both products being compared. We might not know what that value is, but we know it is the same. As a result, we can use that as our base value when making comparisons along the x-axis, which is labeled.

 

 

On 10/27/2023 at 11:22 AM, leadeater said:

And that's the problem, the only thing I can trust about the graph is the text next to it stating what they have, the rest of it could be anything at all.

No, because the x-axis is labeled so it can't be "anything".

The y-axis can be anything, but since we can use that as the reference point we can gauge some information from it using the x-axis.

 

 

On 10/27/2023 at 11:22 AM, leadeater said:

And the reason it does actually matter is the graphed line is trying to present a picture, an idea, an indication of the difference between the products but if I don't know the scale then I don't know if the graph is proportionately depicting the difference or not. Having it not labelled makes it really easy to "embellish for effect" rather than it be an accurate depiction. That's the problem I have with it.

That is one way of looking at it, but it is not the only way.

I would have agreed with you if you had said the graph was poor and should have had the y-axis labeled more accurately. What I don't agree with is calling the entire graph "utterly worthless" and that we can't see the information I have pointed out from it. We absolutely can get quite a bit of information from the graph using the methodology I pointed out. That is how graphs work. It might not be the information you care about. It might not be as much information as if the y-axis was labeled, but this whole discussion I am having with you and Dracarris is quite frankly ridiculous. I am not sure if you both pretend to not understand how graphs work, or if you genuinely do not understand them.

I totally get that the graph might not show what you two want it to show, but I feel like you don't understand the point I am trying to make because you absolutely can get the information I have pointed out. 

 

"At the same performance target, product A uses X watts of power and product B uses Y watts of power" is something we can see from the graph.

What we can't do is accurately control what we want the performance target to be, because that number is a mystery to us. We could do that if the y-axis was something more tangible than "relative performance" like "average FPS in 5 games". I totally understand wanting to be able to get that info out of the graph, but that is not the type of graph this is. That does not mean it is a worthless graph. Graphs can show different things which are more or less relevant depending on who watches it. Just because this graph doesn't show what you want it to show does not make it useless.

It doesn't show what I want to it to show either for that matter, but it's better than nothing and that's my point.

 

 

 

On 10/27/2023 at 11:22 AM, leadeater said:

Let me put it this way, if we take your 17W data point and make the Y value 75% and the upper data point is 80% making everything in-between 5% what does that portray? Now change that Y value to 50%, how about now? So does actually knowing what the Y axis is matter? I say yes. 

You clearly don't understand the methodology I am trying to explain.

As soon as you say "change Y" you should stop, because we can not do that, ever. We always have to pick one specific point along the y-axis and then ever, ever change it. That point on the y-axis is only to be used for a specific measurement and can not be changed (because we can not assign absolute values to anything on the y-axis). 

 

I understand your objections to the graph, but you do not understand my point. It feels like you are stuck with thinking about this graph in a very specific way and my attempts to make you look at it in a different way is failing. I think you need to open your mind a little bit.

 

On 10/27/2023 at 11:22 AM, leadeater said:

This is the standard I expect if such a company wants to use a line graph to compare market options.

I agree and I think that graph is way better. That doesn't mean this graph is completely useless and that we can't get any information out of it however.

[leadeater]

1 hour ago, LAwLz said:

Yes you can.

No you in fact cannot. Tell me exactly what the value is then. We know the maximum is 80% right. So tell me any other value along the Y axis, I'll wait.

 

Edit: Sorry no my mistake, I was figuring that ISO power would be 1 or 100% but it's not for that graph. So the upper Y value point represented is some value close to but not +80%. Doesn't really change the point or issue but I think you'd want me to correct that.

 

1 hour ago, LAwLz said:

I would have agreed with you if you had said the graph was poor and should have had the y-axis labeled more accurately. What I don't agree with is calling the entire graph "utterly worthless"

Oh but it is, I contend to you that the Y axis starts at 79% and the upper value is 80%. Now dispute that with factual evidence 

 

A graph with an unknown axis scale is worthless, completely and utterly.

 

1 hour ago, LAwLz said:

I understand your objections to the graph, but you do not understand my point.

Oh but I do, it's flawed and wrong. No matter what you want to say. If we don't know the value of Y then it's worthless, literally always. We MUST know the value of Y to draw any and all conclusions, comparisons and information. We indeed only know one value of Y, the point at 80% and nothing else. So the only point of comparison possible is at that graph point and nothing else.

 

I don't need to open my mind. It's very clear what is wrong and I have told you. There isn't any countering to that, graphs with no scales can't be used for anything. It's a painting, an art work, it's anything other than graph.

[leadeater]

2 hours ago, LAwLz said:

As soon as you say "change Y" you should stop, because we can not do that, ever. We always have to pick one specific point along the y-axis and then ever, ever change it.

You obviously and clearly do not understand at all what I said. You are saying pick a point right? Yes. So that point has a Y value. What is it? I'm saying that Y value could be anything. That is the point. I'm saying it could be value of a or value of b. I'm not changing anything about the coordinates (the point along the line) on the graph being chosen. I am literally pointing out the Y value could be anything and what the value is drastically changes the interpretation and conclusion you draw from the difference in the products.

 

This is not difficult. Not once ever in what I told you did I say move the point anywhere along the graph. It's entirely static, read what I said again.

Edited November 3, 2023 by leadeater