Core i9 CORRECTION - Performance Per Dollar graph

[Enderman]

Just now, GabenJr said:

That's actually not a terrible idea, and one that people seem to still fail to understand by buying either stupidly high end boards or buying K-series CPUs on a non-Z series chipset.

Thanks.

[lightningterror]

You should add annotations , comment on the video , add it in the description and other things if you're not going to re upload the video.

It's a bit misleading to people that only watch on youtube.

[patthehat]

5 minutes ago, GabenJr said:

It's worth mentioning that we had to crank the voltage to 1.25V per core to get it stable at 4.5 (with a negative AVX offset of 5). I undervolted everything else I could in order to get it to that temperature; I've been told other publications have managed 4.7 and higher, but I'm not so sure how they did outside of silicon lottery. Maybe different boards? Gigabyte AORUS boards were also seeded to others.

 

That's actually not a terrible idea, and one that people seem to still fail to understand by buying either stupidly high end boards or buying K-series CPUs on a non-Z series chipset.

It's even more than just buying the wrong mb for their CPU. This feature is inherent to performance/cost ratios. If your whole setup costs more, higher performance parts will have better impact on overall cost/performance ratio even though they themselves have a worse cost/performance ratio than weaker parts.

So, if you want to optimize your $1,200 build for best performance/cost ratio, you don't want a G4560 (which btw has one of the best performance/cost ratios), but something that has higher performance overall to get the highest performance/cost ratio.

 

This is because the more expensive your system is, the less of an impact has the difference in cost between two parts, therefore the more of an impact has the difference in performance of these parts.

[Knowbody]

7 minutes ago, patthehat said:

It's even more than just buying the wrong mb for their CPU. This feature is inherent to performance/cost ratios. If your whole setup costs more, higher performance parts will have better impact on overall cost/performance ratio even though they themselves have a worse cost/performance ratio than weaker parts.

So, if you want to optimize your $1,200 build for best performance/cost ratio, you don't want a G4560 (which btw has one of the best performance/cost ratios), but something that has higher performance overall to get the highest performance/cost ratio.

 

This is because the more expensive your system is, the less of an impact has the difference in cost between two parts, therefore the more of an impact has the difference in performance of these parts.

So, rather than having just making a chart based on a flat division of total performance divided by total cost, it would probably be better to make a chart with cost as one axis and performance as the other, and each CPU is a dot point on the graph.

[patthehat]

1 minute ago, Knowbody said:

So, rather than having just making a chart based on a flat division of total performance divided by total cost, it would probably be better to make a chart with cost as one axis and performance as the other, and each CPU is a dot point on the graph.

This is a very good and easy to implement suggestion!

[Knowbody]

It would probably be most intuitive to people to have performance as the Y axis, and cost as the X axis. Then, the faster CPU's are at the top, and the cheapest are on the left.

[Space Reptile]

23 minutes ago, GabenJr said:

It's worth mentioning that we had to crank the voltage to 1.25V per core to get it stable at 4.5 (with a negative AVX offset of 5). I undervolted everything else I could in order to get it to that temperature; I've been told other publications have managed 4.7 and higher, but I'm not so sure how they did outside of silicon lottery. Maybe different boards? Gigabyte AORUS boards were also seeded to others.

1.25 is not alot tho 
according to the "leaked" one they where at 1.22 and at ~100c on a dualrad ... wich is wayyyy to high 

 

how are you supposed to keep this thing cool in a workstation doing rendering? 

[nicklmg]

19 minutes ago, lightningterror said:

You should add annotations , comment on the video , add it in the description and other things if you're not going to re upload the video.

It's a bit misleading to people that only watch on youtube.

Annotations are no longer available on YouTube. We've added a card linking to this thread, which is actually better than an annotation would have been since it is available on all platforms including mobile.

[Enderman]

Here we go:

@ 15:20

[leadeater]

5 hours ago, Space Reptile said:

how are you supposed to keep this thing cool in a workstation doing rendering? 

In the professional work space CPUs are not over clocked, I'm referring to much larger outfits than LMG or ones that are not actual PC technology reviewers.

[Space Reptile]

10 minutes ago, leadeater said:

In the professional work space CPUs are not over clocked, I'm referring to much larger outfits than LMG or ones that are not actual PC technology reviewers.

it still runs unreasonably hot at stock boost , and takes unreasonable amounts of power

[spycrash]

Doesn't matter now. The bulk of the audience already saw the video. 

[croppy]

@Enderman I've tried to make sense of what you're saying but I'm just not sure if I am.

 

Are you saying the $/perf comparisons, should be made with the cost of the entire rest of the system + each individual parts costs?

[King_of_Oz]

What software do you use to make the graphs?

[Drak3]

11 hours ago, GabenJr said:

Recorded under AIDA64 load

Ah, so meaningless.

[Emily Young]

4 hours ago, Drak3 said:

Ah, so meaningless.

If you say so - The load was similar under Blender, too.

[Drak3]

1 hour ago, GabenJr said:

If you say so - The load was similar under Blender, too.

Call me skeptical of what settings you used in blender. Before reworking half of my house's wiring, an i5-4460+GTX1070 on a 450W PSU would trip a breaker if the AC was on if I just loaded up word, but it ran Aida64 without a single hitch.

[Enderman]

8 hours ago, croppy said:

@Enderman I've tried to make sense of what you're saying but I'm just not sure if I am.

 

Are you saying the $/perf comparisons, should be made with the cost of the entire rest of the system + each individual parts costs?

I'm saying that in order to avoid mistakes like pairing a $35 GPU with a $1000 PC, or a $1500 GPU with a $300 PC, price performance needs to be calculated by adding the cost of the part you're calculating (such as GPU or CPU) to the cost of the rest of the system.

 

This 'total cost' is what you divide the performance by, not just the cost of the CPU or GPU.

 

In case you missed it, here is my example of why just taking performance and dividing it by the cost of the part alone can give you a completely wrong idea of what is good:

 

[croppy]

@Enderman

yeah I read that, and it just didn't make any sense to me to use that method.

I WANT to know the price performance of the individual component, why care about the rest of the system?

 

If I use the first method from your quote, then the $35 GPU is the better deal... full stop.

However, The only reason I would be comparing them, is if 10fps was within my acceptable parameters. Which is why I think it's silly.

 

Adding in your extra system costs is just completely arbitrary

 

if 1 video card costs $100 more than another video card... I want to know the performance difference of those two items and their absolute costs.

That $100 dollars is still $100 regardless of how much else the system costs, that $100 doesn't change in value when other money is spent.

 

Where I can see your method making sense is when making comparisons like, getting 1xGPU vs 4xGPU (obviously not for gaming)

where a 4xGPU system, also has other attached costs, like beefier motherbaord and powersupply, that aren't needed if you only get 1. but again, you would only add the delta of the required gear, not the whole system... as that's a base cost that you have to spend REGARDLESS, so it can be just forgotten about.

 

again, happy to hear why I'm mistaken, I'm just able to still logic myself into my original position, rather than yours at the point.

[Enderman]

9 minutes ago, croppy said:

That $100 dollars is still $100 regardless of how much else the system costs, that $100 doesn't change in value when other money is spent.

The point is that when calculating performance per dollar you want to get the most out of your money, right?

 

Just look at the video I posted and linus talks about it there.

 

Calculating performance/dollar by only using the cost of the one component can give you the WRONG idea of what product to buy.

You might buy an expensive GPU and pair it with a cheap $300 system, or get a really cheap GPU and pair it with a system that costs thousands of dollars, all because you thought that it had better price performance when you divided it by its cost.

Depending on the cost of the system, one GPU or another may have better price performance.

 

A low end GPU should not be paired with a high end system, and a high end GPU should not be paired with a low end system.

By taking the system cost into account, you get the REAL performance/dollar value of what that entire PC will be capable of.

 

[Emily Young]

43 minutes ago, croppy said:

Adding in your extra system costs is just completely arbitrary

That's my position, too (and without spending much time to think about it, the reason for the initial kneejerk reaction), but with a line graph showing the price/performance curve for a range of system dollar values, it can be useful information because you're not taking just one arbitrary point on the line as in the original example, and necessarily such a formula would also include the data point we've been using (at x = 0).

 

While price doesn't quite make for a super accurate indicator for system performance, it does give you an idea for how much value something will add for a given budget if, say, you were buying a new PC. Given that information, you could plug in your build budget and figure out the sweet spot for each component you'll be buying based on a review that uses that formula. Its price/performance "sweet spot" (at y=1) would also fairly accurately convey the target audience for a given component (or what the target audience should be).

 

Where it falls short is indicating value when upgrading an older machine, because the monetary value of the parts versus performance is no longer in sync for a past or future video. However, it can still give a broad strokes idea for overall value that you can then interpret for your own usage (by, say, adding or subtracting a certain value per generation of age as a simplistic example). It's something I'm going to try to implement, but I'm not entirely certain how well it will translate on a video. As I said before, it's going to take some trial and error to present the information in a legible way. It's a lot more information to process visually than a simple bar graph.

[Knowbody]

1 hour ago, Enderman said:

The point is that when calculating performance per dollar you want to get the most out of your money, right?

 

Just look at the video I posted and linus talks about it there.

 

Calculating performance/dollar by only using the cost of the one component can give you the WRONG idea of what product to buy.

You might buy an expensive GPU and pair it with a cheap $300 system, or get a really cheap GPU and pair it with a system that costs thousands of dollars, all because you thought that it had better price performance when you divided it by its cost.

Depending on the cost of the system, one GPU or another may have better price performance.

 

A low end GPU should not be paired with a high end system, and a high end GPU should not be paired with a low end system.

By taking the system cost into account, you get the REAL performance/dollar value of what that entire PC will be capable of.

 

The question of how much to spend on your CPU vs how much to spend on your GPU is a separate question to the performance per dollar value of an individual component, and should be treated separately, such as in its own article that deals with that question specifically.

 

The only thing you'll actually do when you include the cost of the other components is to reduce the scale of the difference. Because when a reviewer is making a performance per dollar comparison of CPU's, they're going to use the same components such as graphics cards, power supply, etc anyway. There is zero point to including them.

 

The only time it would really make sense to include other components is if, for example, they're not compatible with the same type of RAM, or if there is a real difference in the cost of equivalent motherboards.

Graphics cards, power supplies, hard drives, etc should not be included in any case.

[Enderman]

56 minutes ago, Knowbody said:

The only thing you'll actually do when you include the cost of the other components is to reduce the scale of the difference.

This is not true, as you can see in my GPU example.

[Knowbody]

4 minutes ago, Enderman said:

This is not true, as you can see in my GPU example.

Your GPU example is not addressing the same question that's being addressed in the article. It's a separate question that should really be its own article.

[Enderman]

3 minutes ago, Knowbody said:

Your GPU example is not addressing the same question that's being addressed in the article. It's a separate question that should really be its own article.

???

What do you not understand about an "example" ?

The same thing applies to CPUs.

CPU A costs some amount of money and gets some amount of performance.

Then CPU B costs a different amount of money and gets a different amount of performance.

Literally the same calculation.