Core i9 CORRECTION - Performance Per Dollar graph

[croppy]

@Enderman

if I can try to summarise your point it's that...

 

the individual components of any build should all compliment each other towards your individual goal

 

Which I can totally get on board with, I just think the perf/$ metric is still useful given this. I'd agree that specing out all the individual parts of a machine is a totally different beast and striking the right balance would be a totally interesting video to make.

 

this is how I think of it, Building a PC is like a plate spinning routine.

You want to keep the plates all spinning, to do this you might have a very stable and strong stick with a heavy plate here, and a skinny stick and smaller plate over here, you might add an extra plate on top of one, you might swap out one plate for a different size plate so you can leave it spin longer to keep more plates spinning elsewhere but either way, it's all a delicate balance that is all working towards an end goal.. and ocourse the revolutions per minute of one type of plate might differ slightly from another and you are constantly trying to find the best one for each individual stick, but in no circumstance are you going to put some off center, half cracked terrible plate up there and try to keep the whole thing going.

 

That might be a bit of a weird metaphor, but it's just how it manifests when I try to visually articulate my position.

[Enderman]

6 minutes ago, croppy said:

Which I can totally get on board with, I just think the perf/$ metric is still useful given this.

The problem is that perf/$ is wrong if you only count for the price of the GPU or CPU.

Look at how linus explained it.

In that scrapwyard wars, they had a set budget, and had to get the most performance possible in that budget.

 

Spending $30 instead of $20 on a GPU did not mean he had to get 150% performance, because what matters is the total cost of the system not just the GPU.

If the GPU is only 10% of the system cost, he only needs to get 10% more performance to have the same price/performance as with the $20 GPU.

 

Go watch the video again if you don't understand.

[croppy]

@Enderman Thought experiment, Lets say someone gave you a system with everything except a CPU and that's all you had to buy, would you still do the calculations your way in that situation?

[LabonSky]

Khem. How about just making 2 graphs? One for perf/dollar taking into consideration only CPU. Other where u take let's say:

 

• CPU
• an average competing mobo from ASUS Prime/Aorus series (so X370 for AMD and X299 for Intel)
• Best perf/dollar GPU (like RX 570) or most popular GPU (GTX 1070)
• 16GB of RAM (in whatever configuration is best)
• Samsung 850 EVO 250 GB
• Corsair RM750x

Because the thing is that most people will not buy the 999$ CPU. The price is sorta silly. That's why I'm confused when it's compared to 350$ i7 or 450$ Ryzen 1800x. You can get a whole PC for a 1000 USD (the asking price for the lowest i9) which will probably perform like 15-25% slower in most common enthusiast usage scenarios (games? BTW what do enthusiasts do with their PCs? I just OC and play LOL but my pc is 10 yo so...). Honestly 1800x is not even a good perf/$ cpu. 1600x Takes that crown I think and you can have a whole PC based on that for about 1000$. 

 

 

 

PS. Why is 1800x not OC'ed? It can go up to 4.1 GHz on air from what I've read.

Edited June 21, 2017 by LabonSky
OC the 1800x?

[Enderman]

4 hours ago, croppy said:

@Enderman Thought experiment, Lets say someone gave you a system with everything except a CPU and that's all you had to buy, would you still do the calculations your way in that situation?

Yes, because otherwise you might buy a CPU that doesn't match the build, such as a crappy pentium or something that costs $10 but has really good "price performance" (since you only divide by $10) and then pair it with the rest of the parts that maybe cost thousands of dollars.

 

But of course for the total amount that YOU pay, which is just for the CPU, technically the $10 CPU would still be better price performance, but it would be a mistake to pair that with a high end system for the exact reason above.

[Knowbody]

17 hours ago, Enderman said:

???

What do you not understand about an "example" ?

Nothing, I know what point you're making. You need to explain its relevance to this particular article.

The value of a faster CPU verses a faster GPU is outside the scope of the article, to the point where the same GPU is being used in all of the tests to eliminate it being a factor affecting the results. It makes no sense to include the cost of components that are not relevant and are not being compared.

 

7 hours ago, Enderman said:

Yes, because otherwise you might buy a CPU that doesn't match the build, such as a crappy pentium or something that costs $10 but has really good "price performance" (since you only divide by $10) and then pair it with the rest of the parts that maybe cost thousands of dollars.

 

But of course for the total amount that YOU pay, which is just for the CPU, technically the $10 CPU would still be better price performance, but it would be a mistake to pair that with a high end system for the exact reason above.

This is why having a price:performance graph where price and performance are separate axes on the graph is better; you don't lose information. You can see if a CPU has a "good price:performance ratio" but has low performance.

And to make it easier to tell the price:performance ratio, you can add a line showing the mean price:performance. CPU's above the line have a better than average ratio, and the further above it they are, the better the ratio.

[Enderman]

17 minutes ago, Knowbody said:

Nothing, I know what point you're making. You need to explain its relevance to this particular article.

The value of a faster CPU verses a faster GPU is outside the scope of the article, to the point where the same GPU is being used in all of the tests to eliminate it being a factor affecting the results. It makes no sense to include the cost of components that are not relevant and are not being compared.

It does matter though, that's what you don't understand.

What matters is the price performance of the PC you're trying to build, you can't use a CPU or GPU without all the other components a PC is made of.

 

In order to know what is the best for your money, you can't just say "this GPU costs $1 but gets 5fps therefore I should totally get it over a 1060 that costs $200 and gets 200fps because it has 5x better price performance than the 1060"

That's ridiculous. That's what you're saying by only counting the price of the GPU (or CPU).

 

When you spend $1k on the rest of the PC, the 1060 has better price performance.

Why is this so hard to understand?

 

 

 

The same thing applies to CPUs. Calculate price performance based on the price of the whole system.

Obviously you don't try to calculate them both at the same time, you first pick one component then pick the other.

[Knowbody]

30 minutes ago, Enderman said:

It does matter though, that's what you don't understand.

What matters is the price performance of the PC you're trying to build

No, GPU performance may or may not matter to the people reading the article. And if it does, they can read other articles that review GPU's.

CPU review articles should be about comparing the CPU's in question to other CPU's.

 

If you start introducing price:performance of GPU's alongside CPU's, you have to take into account all sorts of different use cases, because the price:performance ratio changes with different types of usage, and that magnifies the amount of testing you need to do.

And it also becomes a slippery slope, too. I mean, if you include GPU's, why not include RAM? Or SSD's? Or all sorts of other hardware? And then to take into account all the different use cases again, you test test all of the different possible combinations of CPU/GPU/RAM/SSD, and suddenly there are 500 times more tests you have to run.

[Enderman]

11 minutes ago, Knowbody said:

No, GPU performance may or may not matter to the people reading the article. And if it does, they can read other articles that review GPU's.

CPU review articles should be about comparing the CPU's being reviewed to other CPU's.

 

If you start introducing price:performance of GPU's alongside CPU's, you have to take into account all sorts of different use cases, and that magnifies the amount of testing you need to do.

And it also becomes a slippery slope, too. I mean, if you include GPU's, why not include RAM? Or SSD's? And then to take into account all the different use cases again, you test test all of the different possible combinations of RAM and SSD's, and suddenly there are 500 times more tests you have to run.

Ram and SSDs will not affect your fps by more than a few percent, so it's far from as important as a CPU or GPU.

 

As I said 20 times before, what I'm explaining applies both to CPUs and GPUs.

 

For example, you have a $10 xeon from ebay which gets you idk say maybe 100 cinebench points, or a $300 7700k which gets you 200 points.

According to you, nobody should buy the 7700k because the cheap xeon has better price performance.

 

However, when you consider that the rest of your PC parts will cost maybe $1500, then the 7700k has better value for your money.

100/1510 = .066

200/1800 = .111

7700k build has better price performance.

[Knowbody]

16 minutes ago, Enderman said:

Ram and SSDs will not affect your fps by more than a few percent, so it's far from as important as a CPU or GPU.

RAM and SSD's could provide enormous performance benefits in particular use cases. Gaming FPS isn't the only kind of performance people might care about.

GPU's can provide enormous benefits in particular use cases, or absolutely none in others, such as many CPU-bound use cases.

 

And maybe people reading a review about CPU's don't care about GPU performance, or hard drive read/write speeds. Or maybe they care much more about hard drive read/write speeds than GPU performance. They can look at multiple different articles that address their own particular use cases.

 

It shouldn't be up to the article to include all sorts of other hardware that may or may not be relevant to people. At best, you could argue that perhaps the article should outline all of the particular use cases that are relevant to the products being reviewed, and the person reading the article should be able to decide if those use cases are relevant to them.

[Enderman]

1 minute ago, Knowbody said:

RAM and SSD's could provide enormous performance benefits in particular use cases. Gaming FPS isn't the only kind of performance people might care about.

Try reading the rest of the post.

[Knowbody]

1 minute ago, Enderman said:

Try reading the rest of the post.

I read it. It's not relevant. That's the point.

You're not telling me anything I don't know. The issue is that the information is not relevant to this particular case.

[Rodskagg]

Wow, really interesting discussion here guys. I just had to create a account just to take part. (not joking this is really interesting)

 

One thing I think you miss is that when it comes to GPU and CPU performance a person is not looking for best performance per dollar but rather a performance that is satisfying the needs one have. For example for my grandma a satisfying CPU performance would be that of being able to surf the webb and sending emails and a GPU performance so that she can look at youtube. However for a gamer a satisfying CPU and GPU performance would be so that one can play the games one want to play at a graphical level and frame rate one think is good enough.

 

To get back to your example @Enderman. If a 35$ dollar GPU is satisfying ones needs (even for a 1000$ PC build) then yes that person should and will buy the 35$ GPU.

 

And in general I agree with @Knowbody more. What should be compared between buying the 35$ and 700$ GPU is the price difference, that is 665$ in price difference and how this compares to the performance gain. The cost of the rest of the system is irrelevant.

[Enderman]

9 minutes ago, Rodskagg said:

To get back to your example @Enderman. If a 35$ dollar GPU is satisfying ones needs (even for a 1000$ PC build) then yes that person should and will buy the 35$ GPU.

No, because in that case they would not be spending $100 on the rest of the parts, they would be spending a few hundred dollars.

Then when you calculate performance per dollar using a few hundred instead of $1000, you see that the $35 GPU has better price performance.

[Knowbody]

11 minutes ago, Enderman said:

No, because in that case they would not be spending $100 on the rest of the parts, they would be spending a few hundred dollars.

Or, maybe they would still only spend $100 on them, because for their particular use case, it's more than enough.

The real questions should be:

• Do you understand what the products you're potentially purchasing actually do?
• Do you understand your own particular use cases?

Because if you do, you should then know which types of products you should read reviews on, and work out for yourself some sort of appropriate balance based on the benefits each particular product has.

[Enderman]

7 minutes ago, Knowbody said:

Or, maybe they would still only spend $100 on them, because for their particular use case, it's more than enough.

The real questions should be:

• Do you understand what the products you're potentially purchasing actually do?
• Do you understand your own particular use cases?

Because if you do, you should then know which types of products you should read reviews on, and work out for yourself some sort of appropriate balance based on the benefits each particular product has.

That was a typo, mean to say $1000.

 

The entire point of price performance is to balance a build according to how much you have to spend.

So yes, by taking the entire price of the system into account, you can find out if one CPU or another (or one GPU or another) is better for the price you're willing to spend.

 

Sometimes people want to spend more, and then a more expensive CPU or GPU will make more sense for them for how much money they spend on their build vs how much performance they get from their build.

 

Again, only using the price of the single component to determine price/performance gives you completely incorrect results.

[Rodskagg]

I can agree with you @Enderman that the price/performance metric is problematic. But I think we disagree why it is problematic.

 

The problem with only showing the price/performance diagram is that you loose information. The earlier presented idea that one should show the performance on one axis, price on another and let each CPU be represented by a dot on this graph I think is a very good idea. Then you can not only see the performance and the price but the price/performance will also be visualized. More important the price difference and performance difference between different CPU/GPU will also be visualized.

[Knowbody]

16 minutes ago, Enderman said:

That was a typo, mean to say $1000.

 

The entire point of price performance is to balance a build according to how much you have to spend.

So yes, by taking the entire price of the system into account, you can find out if one CPU or another (or one GPU or another) is better for the price you're willing to spend.

 

Sometimes people want to spend more, and then a more expensive CPU or GPU will make more sense for them for how much money they spend on their build vs how much performance they get from their build.

And the price:performance is different depending on your particular use case.

 

Like I said, maybe someone doesn't care about GPU performance at all, because it doesn't benefit them. Maybe they benefit from much higher hard drive speeds, or RAM (if they're building a database or something).

 

If you understand your particular use case and what each product actually does, then you should know which reviews to read.

 

It shouldn't be up to the review to include information on other types of products that may or may not be relevant to people reading the reviews. Because like I said, then it just becomes a slippery slope, with an enormous amount more work, and potentially annoying to people reading the review who don't actually care about those other products anyway.

 

If you care about CPU performance, you read reviews on CPU's. If you care about GPU performance, you read reviews on GPU's. If you care about SSD speed, you read reviews about SSD's.

You don't need to include GPU information in a review on CPU's; people who care about both CPU's and GPU's can still find information on both by reading multiple different articles.

[Enderman]

15 minutes ago, Knowbody said:

And the price:performance is different depending on your particular use case.

 

Like I said, maybe someone doesn't care about GPU performance at all, because it doesn't benefit them. Maybe they benefit from much higher hard drive speeds, or RAM (if they're building a database or something).

In this case obviously you would not use a GPU benchmark to compare components...

Maybe you would use a memory benchmark and compare the prices of two different sets of ram.

Maybe you would use a SSD benchmark and compare the speeds (although this isn't a good idea, speed is not what's most important, reliability and consistency is)

 

The point is, for someone taking a look at a specific benchmark for their use case, you divide by the cost of your whole system, not by only the price of the component you're buying.

 

A video editor might compare CPUs only by render benchmarks rather than gaming benchmarks. I'm not saying what kind of benchmark people should use.

I'm saying that you do not divide by only the cost of the part.

[Knowbody]

8 minutes ago, Enderman said:

In this case obviously you would not use a GPU benchmark to compare components...

Maybe you would use a memory benchmark and compare the prices of two different sets of ram.

Maybe you would use a SSD benchmark and compare the speeds (although this isn't a good idea, speed is not what's most important, reliability and consistency is)

 

The point is, for someone taking a look at a specific benchmark for their use case, you divide by the cost of your whole system, not by only the price of the component you're buying.

No, you divide by the cost of the individual component, because that's the component in question.

 

But also, when a reviewer reviews CPU's, they're going to make the rest of the system the same in each test wherever possible to make sure they aren't a factor affecting the results.

So, the total cost of the system is only going to change based on the cost of the CPU anyway, and maybe to a degree, the motherboard. And maybe in specific cases, the RAM, if the CPU's aren't compatible with the same type of RAM. But if they are compatible with the same type of RAM, the same type of RAM is used.

 

But things like power supplies, graphics cards, hard drives, etc aren't relevant at all to the price:performance of the CPU.

[Enderman]

3 minutes ago, Knowbody said:

No, you divide by the cost of the individual component, because that's the component in question.

It appears as though you're another one of those people that doesn't understand that it is wrong to divide by the cost of only the individual component.....

 

3 minutes ago, Knowbody said:

So, the total cost of the system is only going to change based on the cost of the CPU anyway, and maybe to a degree, the motherboard. And maybe in specific cases, the RAM, if the CPU's aren't compatible with the same type of RAM. But if they are compatible with the same type of RAM, the same type of RAM is used.

Again, it is NOT only the difference that matters.

Go look at literally all the examples and calculations in the other 2 pages of this topic.

Yes the cost of the system only changes by the cost of the CPU, no the price performance of the system does not stay the same.

 

I will copy paste my example until you actually read it and try to understand this simple elementary school math.

 

 

Quote

For example, you have a $10 xeon from ebay which gets you idk say maybe 100 cinebench points, or a $300 7700k which gets you 200 points.

According to you, nobody should buy the 7700k because the cheap xeon has better price performance.

 

However, when you consider that the rest of your PC parts will cost maybe $1500, then the 7700k has better value for your money.

100/(1500 + 10) = .066

200/(1500 + 300) = .111

7700k build has better price performance.

 

[Knowbody]

3 minutes ago, Enderman said:

It appears as though you're another one of those people that doesn't understand that it is wrong to divide by the cost of only the individual component.....

 

Again, it is NOT only the difference that matters.

Go look at literally all the examples and calculations in the other 2 pages of this topic.

Yes the cost of the system only changes by the cost of the CPU, no the price performance of the system does not stay the same.

 

I will copy paste my example until you actually read it and try to understand this simple elementary school math.

I didn't say it stayed the same, the magnitude just shrinks, but the relative positions stay the same.

In other words, all you're doing is watering down the difference.

[Enderman]

3 minutes ago, Knowbody said:

I didn't say it stayed the same, the magnitude just shrinks, but the relative positions stay the same.

In other words, all you're doing is watering down the difference.

This is 100% incorrect, as you can see in the example.

 

Before:

Xeon = 10 points per dollar

7700k = 0.666 points per dollar

 

After (taking into account $1500 system price):

Xeon = 0.066

7700k = 0.111

 

LOOK AT THAT! THEY SWITCHED!

 

As you can see, the price performance may completely change depending on how much the system cost is.

Do the calculation again with a $200 system cost, you will see the Xeon has better price performance in that case.

[Knowbody]

Just now, Enderman said:

This is 100% incorrect, as you can see in the example.

 

Before:

Xeon = 10 points per dollar

7700k = 0.666 points per dollar

 

After (taking into account $1500 system price):

Xeon = 0.066

7700k = 0.111

 

LOOK AT THAT! THEY SWITCHED!

How many times is a $10 CPU going to be put in a system that otherwise costs $1500? That's just idiotic.

[Enderman]

3 minutes ago, Knowbody said:

How many times is a $10 CPU going to be put in a system that otherwise costs $1500? That's just idiotic.

A lot of people did, ever since linus made that video about a cheap xeon gaming rig.

 

Anyway, let's do a different example.

$200 i5 with 700 points vs $350 i7 with 1000 points.

 

Before:

i5 = 700/200 = 3.5

i7 = 1000/350 = 2.86

The i5 is 22% better price performance (or that's what you think).

 

With $500 system:

i5 = 700/(200+500) = 1

i7 = 1000/(350+500) = 1.17

The i7 is 17% better price performance.

 

With $1500 system:

i5 = 700/(200+1500) = .411

i7 = 1000/(350+1500) = .541

The i7 is 30% better price performance.

 

As you can see, the price performance of the build will change depending on how much you spend on your build, not just the CPU.

 

Do you get it yet?