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

[Dracarris]

1 hour ago, LAwLz said:

If we start at 0,0 on the graph

You don't know where (0,0) is. You only know where x=0 is.

 

In other news: Source in German. Only Geekbench, not sure how trustable, but things stay interesting for sure.

[Kisai]

2 hours ago, Dracarris said:

You don't know where (0,0) is. You only know where x=0 is.

 

In other news: Source in German. Only Geekbench, not sure how trustable, but things stay interesting for sure.

That's entirely misleading without the GPU and Compute charts. Sure, the CPU might be comparable, but this is a SoC, so you need to compare the GPU, because it could quite literately be running laps around one part, or being lapped by another, with the same TDP.

 

Which, surprise, surprise, we don't have one for the Snapdragon.

 

And when you look at multi-core charts with real software, the Qualcomm chips still looks like it's behind.

 

Like if you frame things correctly, sure, maybe it looks good if you are only comparing power budgets, but those i9 desktop intel parts and AMD Ryzen 9 parts non-K are 65w TDP, and not on the chart. Intel and AMD's "U" parts are 25w parts. The HX parts are 55w parts.

 

It starts looking a bit dishonest when Apple and Qualcomm's charts are basically exaggerations without comparisons to products testers wouldn't test them against (most benchmark tests always test against flagship parts, not the TDP locked parts which will clearly have a lower performance.)

 

M3

https://browser.geekbench.com/search?q=mac15%2C3

 

Snapdragon X Elite (assuming it's not fake)

https://browser.geekbench.com/v6/cpu/3326512

No compute chart

 

 

What we can infer from the Qualcomm chart

it's 10% ahead of the M2 (not the M3), maybe that puts it in spitting distance of the M3, or maybe it doesn't, since we only see comparison to Intel mobile parts, we don't know if the benchmark was massaged.

 

It's not like GPU vendors haven't a history of cheating benchmarks.

[Dracarris]

4 hours ago, Kisai said:

That's entirely misleading without the GPU and Compute charts. Sure, the CPU might be comparable, but this is a SoC, so you need to compare the GPU, because it could quite literately be running laps around one part, or being lapped by another, with the same TDP.

Geez, relax. Those are simply early, unverified results with no demand to paint a full picture, whatsoever. More will follow. And the way I see it, CPU perf in itself even with complete disregard for the GPU is very interesting. I'm probably not the only one who is mainly interested in CPU perf for their laptop.

 

As for TDP I don't fully get your point. Comparisons were made against both QC SoCs which have heavily different TDPs, and against a desktop CPU with god-knows-how-high TDP.

 

Here are some comparisons against high-end laptop chips from both AMD and Intel, and again the M3 looks really appealing.

[LAwLz]

11 hours ago, leadeater said:

No you in fact cannot. Tell me exactly what the value is then.

I am literally facepalming right now.

It's like talking to a brick wall. Which part of:  

12 hours ago, LAwLz said:

We might not know what that value is, but we know it is the same.

Do you not understand?

 

 

11 hours ago, leadeater said:

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.

What are you talking about?

1) There is no ISO power specified on the graph I posted and made my example with.

2) Neither the x-axis nor the y-axis specifies percentages. It's either "relative performance" or power measured in watts. There are no percentages.

 

I am starting to suspect that you are looking at some other graph and that's why you are confused.

The y-axis on the graph I posted does not represent percentages at all. It represents relative performance. You're talking about something completely different.

 

 

11 hours ago, leadeater said:

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.

You keep trying to compare various points on the y-axis despite me saying time and time again that you should not do that.

Again, try and be a bit more open-minded because you have tunnel vision right now.

 

Read very carefully what I wright now, and do not try and think.

Take the graph I posted a few posts ago.

Draw a line straight out from the y-axis and never ever change it to any other position. It really doesn't matter which position you start at on the y-axis because it is irrelevant. It's meaningless. The only thing that means something is the number we will arrive at on the x-axis. In case you don't know, the x-axis is the one labeled "power consumption" at the bottom.

Once you have drawn the line, you can ignore that the y-axis even exists. Do not try and infer any number from it. Do not try and figure out what scale it is because it is irrelevant.

 

Once the line from the y-axis intersects the Snapdragon line, draw a line straight down and look at what the wattage number says. Write that number down.

Then keep drawing on the same line all the way until it intersects with the Ryzen line. Once it does that, draw a line straight down and note down which number it has. Your picture should now look like this:

 

 

What this picture says is that at the same performance target, the Snapdragon's GPU uses about 9 watts of power while the Ryzen one uses about 18 watts of power.

That's what we can gather.

 

We can erase the lines and draw new lines using the same method to change the "performance target". We will never be able to infer what the "performance target" number is or even what it means, but we will always be able to make a direct comparison with how much wattage the two chips use for an unknown performance target. 

How valuable is that information? Not that valuable, but it is more than nothing.

 

What we can't do, which is what you seem to keep trying to do, is infer what the performance is. We can't do that. We will never be able to compare the numbers at two different points on the y-axis. We will always have to pick a single point on the y-axis and then use that to make comparisons along the x-axis. 

 

11 hours ago, leadeater said:

We MUST know the value of Y to draw any and all conclusions, comparisons and information.

No we don't.

As I said above, assuming the data in the graph is correct and not just made up, we can say for certain that at some performance target, the X Elite will be able to provide the same performance at ~9 watts as the Ryzen chip needs ~18 watts to do. 

We can't know how much performance that is using only the information provided in the graph, but the statement I made above is factually correct and was created using only the information the graph provides us with. As a result, we can't say it is useless nor that it "doesn't tell us anything".

 

 

If you want to keep this conversation going then I want you to refuse this statement:

"Any some unknown GPU performance target the Snapdragon X Elite will use roughly 9 watts of power to do the same amount of work as the Ryzen 9 7940HS needs about 18 watts of power to do".

 

 

 

  

11 hours ago, leadeater said:

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.

The Y value could be anything because it does not matter in the comparison I am making.

All that matters to make the conclusion I am making is that it is the same for both chips being compared.

Let me repeat that, what the value of the Y-axis is is irrelevant when interpreting the graph the way I do. The y-axis is only used to create a common ground between the two chips so that we can point to different points along the x-axis.

Some point on the y-axis might mean "10 fps in BG3" or it might mean "100 fps in BG3". It is irrelevant because the comparison I am making isn't about absolute performance numbers. It's about the absolute wattage numbers at some performance target. Which performance target doesn't matter.

[LAwLz]

10 hours ago, Dracarris said:

You don't know where (0,0) is. You only know where x=0 is.

When I said 0,0 (sorry I forgot the paraphrasis) I didn't mean the values the graph represents. (0,0) are coordinates for the origin. It's the value of the origin, where the x-axis and y-axis intersect. The coordinates and values are two different things. 

 

The absolute value given to the coordinates (0,100) does not have to be 100. 

(0,0) in this case just meant "where the x-axis and y-axis intersect", not "where the absolute value is 0 on the x-axis and 0 on the y-axis".

[leadeater]

3 hours ago, LAwLz said:

1) There is no ISO power specified on the graph I posted and made my example with.

You didn't, Qualcomm did, read the bottom text note on the graph.

 

P.S. ISO means equal so any point you choose actually is ISO, ISO performance heh.

Quote

Iso is a Greek-derived prefix meaning "equal"

 

3 hours ago, LAwLz said:

2) Neither the x-axis nor the y-axis specifies percentages. It's either "relative performance" or power measured in watts. There are no percentages.

Yes it does, this graph is literally the same as the Intel one I have posted missing the Y-Axis label but it is in percentage.

 

I do not care how many times you want to draw a line out from any point along Y if I don't know the value and I don't know it is not a total lie and I don't know the rate of change along Y aka the scale. You might want to say you can ignore the actual relative Y value and just look at the power but the entire purpose of the graph is to plot the relative performance change against power so it CANNOT be ignored. This actual value is important to analyzing the graph even at isolated chosen points because you must be able to actually compare between different chosen points which you cannot accurately do without knowing the values of Y.

 

WE MUST KNOW THE VALUE OF Y.

 

And you want to know why it matters?

 

3DMark TimeSpy iGPU:

13800H: 1857 (I couldn't find the power figure for this)
7940HS: 2911 ~45W

 

So you want to tell me that two different GPUs with ~60% the performance difference will graph like (preface context, your own words and understanding state that the same point along the line with be the same unknown value and both graphs use the same Y point along the Elite X line)

 

This:

and this:

 

And not immediately know there is a giant ass problem and someone is pulling some EPIC BS. You want to tell me that the performance of the Elite X can possibly graph out in near identical relative performance to two different product with 60%+ performance differences at identical power scale? How? Absolutely no way.

 

How is the Elite X at ~30W 2x faster than the 13800H iGPU and also at ~30W 1.8x faster than the 7940HS iGPU when the difference between the 13800H and 7940HS also at ~30W is roughly 1.6x?

 

Unless evidenced otherwise the Elite X line on the graph is made up and is for effect only and doesn't represent true values along it other than at true known points so picking arbitrary points along it is dangerous.

 

https://www.cpu-monkey.com/en/cpu_benchmark-3dmark_time_spy_graphics_score

 

And yes I am assuming the performance differences is the same between 3DMark Wildlife and TimeSpy i.e. whatever the scoring used the difference in scores for the products relative to each other is fairly consistent, unless the benchmark itself is flawed in actually showing performance differences in products.

 

3 hours ago, LAwLz said:

How valuable is that information? Not that valuable, but it is more than nothing.

I'm literally saying it has zero value without the relative performance value. Your opinion on the matter is different, fine. But I'm holding some actual standards here. This is a graph that is graphing relative performance and I will absolutely not accept anything less than labelled axis.

 

I will therefore not engage or draw any conclusions, compare at any points for power or relative performance along a graph I truly believe are erroneous and misleading.

 

I believe there is a very real reason Qualcomm did not make a graph like this with both the 7940HS and 13800H on it, because it's not possible for the shape of the Elite X line to be the same on both graphs if the scale on both is the same, when you start playing funny business with graphs and scales I stop trusting you.