Jump to content

Core i9 CORRECTION - Performance Per Dollar graph

Hey guys, so sorry about this. As pointed out by @patthehat over in the official thread, the Performance Per Dollar graph as shown in the video is incorrect - It was showing the weighted averages of the benchmarks.

 

It SHOULD show the following:

 

BjlZO5U.jpg 

Hopefully that makes a bit more sense! :) 

Emily @ LINUS MEDIA GROUP                                  

congratulations on breaking absolutely zero stereotypes - @cs_deathmatch

Link to comment
Share on other sites

Link to post
Share on other sites

still does not improve X299's value , RIP 

RyzenAir : AMD R5 3600 | AsRock AB350M Pro4 | 32gb Aegis DDR4 3000 | GTX 1070 FE | Fractal Design Node 804
RyzenITX : Ryzen 7 1700 | GA-AB350N-Gaming WIFI | 16gb DDR4 2666 | GTX 1060 | Cougar QBX 

 

PSU Tier list

 

Link to comment
Share on other sites

Link to post
Share on other sites

Thanks for the update

 

I think you should have on each slide/graph, the same spot for each cpu. For example the 1800X should be always at the bottom. The newest video I did watch all the way through, but the intro was lame.

 

Linus did say they are back to leap frogging so thats great for the consumer.

Today, its AMD that won!

Link to comment
Share on other sites

Link to post
Share on other sites

Thanks LTT's Gaben Newell!

Personal Desktop":

CPU: Intel Core i7 10700K @5ghz |~| Cooling: bq! Dark Rock Pro 4 |~| MOBO: Gigabyte Z490UD ATX|~| RAM: 16gb DDR4 3333mhzCL16 G.Skill Trident Z |~| GPU: RX 6900XT Sapphire Nitro+ |~| PSU: Corsair TX650M 80Plus Gold |~| Boot:  SSD WD Green M.2 2280 240GB |~| Storage: 1x3TB HDD 7200rpm Seagate Barracuda + SanDisk Ultra 3D 1TB |~| Case: Fractal Design Meshify C Mini |~| Display: Toshiba UL7A 4K/60hz |~| OS: Windows 10 Pro.

Luna, the temporary Desktop:

CPU: AMD R9 7950XT  |~| Cooling: bq! Dark Rock 4 Pro |~| MOBO: Gigabyte Aorus Master |~| RAM: 32G Kingston HyperX |~| GPU: AMD Radeon RX 7900XTX (Reference) |~| PSU: Corsair HX1000 80+ Platinum |~| Windows Boot Drive: 2x 512GB (1TB total) Plextor SATA SSD (RAID0 volume) |~| Linux Boot Drive: 500GB Kingston A2000 |~| Storage: 4TB WD Black HDD |~| Case: Cooler Master Silencio S600 |~| Display 1 (leftmost): Eizo (unknown model) 1920x1080 IPS @ 60Hz|~| Display 2 (center): BenQ ZOWIE XL2540 1920x1080 TN @ 240Hz |~| Display 3 (rightmost): Wacom Cintiq Pro 24 3840x2160 IPS @ 60Hz 10-bit |~| OS: Windows 10 Pro (games / art) + Linux (distro: NixOS; programming and daily driver)
Link to comment
Share on other sites

Link to post
Share on other sites

19 minutes ago, Space Reptile said:

still does not improve X299's value , RIP 

It's still a useless graph, the proper way of calculating performance per dollar is taking into account the cost of the WHOLE system, not just the CPU or GPU.

 

NEW PC build: Blank Heaven   minimalist white and black PC     Old S340 build log "White Heaven"        The "LIGHTCANON" flashlight build log        Project AntiRoll (prototype)        Custom speaker project

Spoiler

Ryzen 3950X | AMD Vega Frontier Edition | ASUS X570 Pro WS | Corsair Vengeance LPX 64GB | NZXT H500 | Seasonic Prime Fanless TX-700 | Custom loop | Coolermaster SK630 White | Logitech MX Master 2S | Samsung 980 Pro 1TB + 970 Pro 512GB | Samsung 58" 4k TV | Scarlett 2i4 | 2x AT2020

 

Link to comment
Share on other sites

Link to post
Share on other sites

11 minutes ago, Enderman said:

It's still a useless graph, the proper way of calculating performance per dollar is taking into account the cost of the WHOLE system, not just the CPU or GPU.

 

All other hardware being equal, the motherboard would be more expensive for X299 compared to X370 or Z270, but compared to X99, that would be more or less identical. The delta would be about $100-150 either way, which would knock 0.012 off the stock 7900X's score, 0.013 off the OC'd 7900X, and 0.006 off the 6950X.

Emily @ LINUS MEDIA GROUP                                  

congratulations on breaking absolutely zero stereotypes - @cs_deathmatch

Link to comment
Share on other sites

Link to post
Share on other sites

@GabenJr

 

I'm also of the opinion that it would be better for products ( cpus , video cards , etc ) to stay at the same place vertically and not be sorted automatically as highest at the top.

Each test is only shown for a short time on the screen and it's really hard to process each screen and by the time you figure out you lose the graph, so everyone's basically forced to pause video multiple times just to see how items compare.

 

The ideal way would be probably to animate the charts ... show a chart for a few seconds, then change the titles and subtitles and make them flicker or shine for a second to attract eyes to the change and while that happens, animate the bars to decrease or increase to their new values.  5-10 seconds per chart with about 2-3s animation (so the bars would be fixed for about 5s) should be enough for everyone.

 

Also speaking of just the image above, it's kinda low contrast, you're forcing viewers to squint when reading the charts and the font looks ugly, pixelated

.

Maybe it would help to make the texts on the left (the names of the cpus)  in fixed width rectangles that are black with maybe 50-60% transparency (so it's white text on blacker stuff, instead of that cross-stitch texture)

 

Link to comment
Share on other sites

Link to post
Share on other sites

27 minutes ago, GabenJr said:

All other hardware being equal, the motherboard would be more expensive for X299 compared to X370 or Z270, but compared to X99, that would be more or less identical. The delta would be about $100-150 either way, which would knock 0.012 off the stock 7900X's score, 0.013 off the OC'd 7900X, and 0.006 off the 6950X.

Assuming the prices you used to calculate that are $350 for the 7700k and $1000 for the 7900X, then the stock performance numbers you used are 77.35 and 89 respectively.

 

Then, if you assume the rest of the system costs $1000, the price for the 7700k build is $1350 and the 7900X build is $2000. The performance per dollar is then .057 and .045, far closer than your original graph shows.

 

If we take a $2000 PC for example, the 7700k build would be $2350 and the 7900X build $3000. The price performance is then .033 and .03

 

As you can see, the rest of the cost of the system certainly does matter for price performance, because it is not as significant a difference as your graph shows it to be.

Even linus said so himself years ago that calculating price performance without taking into account the system cost is completely wrong.

 

 

 

Now you don't see it in this example because the price difference is large (300%) and performance difference is small (only 12 points) but there is a point where if the system cost is high enough, the price performance of the more expensive CPU/GPU is actually BETTER than that of the cheaper CPU/GPU.

 

For example, let's take a $35 GPU that gets 10fps in a game and a $700 GPU that gets 100fps in a game.

If you only take into account the price of the card, you will see the price performance of the $35 GPU is .285 fps/$ while the $700 GPU is .143 fps/$

So then you buy the card with better price performance right?

Well, if your PSU+case+motherboard+ram end up costing $1000, the cheap GPU build will be $1035 while the expensive build will be $1700.

10/1035 = .0097

100/1700 = .0588

 

You can see that it switched around! The more expensive GPU now has the better price performance, even though the GPU and the PC both cost more.

If you only take the GPU's or CPU's price into account, people can end up making the mistake of thinking it is a much bigger difference than it actually is (such as in your graph above) or even making the completely incorrect choice, such as pairing a $35 GPU with a $1000 PC thinking it has better price performance.

NEW PC build: Blank Heaven   minimalist white and black PC     Old S340 build log "White Heaven"        The "LIGHTCANON" flashlight build log        Project AntiRoll (prototype)        Custom speaker project

Spoiler

Ryzen 3950X | AMD Vega Frontier Edition | ASUS X570 Pro WS | Corsair Vengeance LPX 64GB | NZXT H500 | Seasonic Prime Fanless TX-700 | Custom loop | Coolermaster SK630 White | Logitech MX Master 2S | Samsung 980 Pro 1TB + 970 Pro 512GB | Samsung 58" 4k TV | Scarlett 2i4 | 2x AT2020

 

Link to comment
Share on other sites

Link to post
Share on other sites

@Enderman If all other hardware is the same, then only the cost of the CPU and the cost delta of the motherboard is a factor as far as price to performance goes. In the case of our test benches, that was the case. We're talking apples to apples against the CPUs with otherwise identical builds. Anything else is muddying the waters unnecessarily.

Emily @ LINUS MEDIA GROUP                                  

congratulations on breaking absolutely zero stereotypes - @cs_deathmatch

Link to comment
Share on other sites

Link to post
Share on other sites

6 minutes ago, GabenJr said:

@Enderman If all other hardware is the same, then only the cost of the CPU and the cost delta of the motherboard is a factor as far as price to performance goes. In the case of our test benches, that was the case. We're talking apples to apples against the CPUs with otherwise identical builds. Anything else is muddying the waters unnecessarily.

Read the post again, when calculating price performance, you assume that the rest of the PC costs the same, just like you said.

 

The point is that it is NOT $350 vs $1000, it is $1350 vs $2000 or $2350 vs $3000, it depends on how expensive the rest of the system is.

The calculations I did are with IDENTICAL PCs, that's why I assumed (for my two separate examples) that it would be $1000 for the rest of the parts in the first example, and $2000 for the rest of the parts in the second example.

 

How expensive the system is completely changes what the price performance is, as you can see with my GPU example, if you only take into account the GPU price you will think that the $35 GPU has better price performance, but when you put it into a $1k PC it will have worse price performance than a $700 GPU.

You compare both the $35 GPU and $700 GPU with the SAME SYSTEM that costs $1000.

 

6 minutes ago, GabenJr said:

If all other hardware is the same, then only the cost of the CPU and the cost delta of the motherboard is a factor

This is what I just proved 100% wrong in my above post.

NEW PC build: Blank Heaven   minimalist white and black PC     Old S340 build log "White Heaven"        The "LIGHTCANON" flashlight build log        Project AntiRoll (prototype)        Custom speaker project

Spoiler

Ryzen 3950X | AMD Vega Frontier Edition | ASUS X570 Pro WS | Corsair Vengeance LPX 64GB | NZXT H500 | Seasonic Prime Fanless TX-700 | Custom loop | Coolermaster SK630 White | Logitech MX Master 2S | Samsung 980 Pro 1TB + 970 Pro 512GB | Samsung 58" 4k TV | Scarlett 2i4 | 2x AT2020

 

Link to comment
Share on other sites

Link to post
Share on other sites

This is the last time I am going to explain myself, the price performance needs to be calculated taking into account the rest of the system.

 

My example again:

Quote

For example, let's take a $35 GPU that gets 10fps in a game and a $700 GPU that gets 100fps in a game.

 

If you only take into account the price of the card, you will see the price performance is:

$35 GPU  = 10/35 = .285 fps/$

$700 GPU = 100/700 = .143 fps/$

 

Cheap GPU has better price performance? Nope. See:

 

Well, if your build's PSU+case+motherboard+ram+storage end up costing $1000, the cheap GPU build will be $1035 while the expensive build will be $1700.

10/1035 = .0097 fps/$

100/1700 = .0588 fps/$

 

Expensive GPU has better price performance. SIX TIMES better. It is a better match for the rest of your parts.

 

The price of the rest of the components DOES matter, even when they are the exact same components and price for both systems.

 

If this super simplified example is too difficult to understand, please talk to @LinusTech and get him to explain this concept to you, because I can't make it any more simple than this.

NEW PC build: Blank Heaven   minimalist white and black PC     Old S340 build log "White Heaven"        The "LIGHTCANON" flashlight build log        Project AntiRoll (prototype)        Custom speaker project

Spoiler

Ryzen 3950X | AMD Vega Frontier Edition | ASUS X570 Pro WS | Corsair Vengeance LPX 64GB | NZXT H500 | Seasonic Prime Fanless TX-700 | Custom loop | Coolermaster SK630 White | Logitech MX Master 2S | Samsung 980 Pro 1TB + 970 Pro 512GB | Samsung 58" 4k TV | Scarlett 2i4 | 2x AT2020

 

Link to comment
Share on other sites

Link to post
Share on other sites

@GabenJr

 

@Enderman is right on this one, and I can show it with some math. First lets set up a function that calculates relative price-performance:

 

factor * (cost1+fixcost)/(perf1) = (cost2+fixcost)/(perf2)        | solved to factor

 

factor = (perf1/perf2) * (cost2+fixcost)/(cost1+fixcost)           | translate to general variables with fixcost being x

 

f(x) = (perf1/perf2) * (cost2+x)/(cost1+x)

 

This function gives us relative cost-performance of element 1 vs element 2, dependent on the fixcost of the whole system.

 

I drew two graphs at http://www.mathe-fa.de/en:

 

blue: i7 (perf/cost2) vs i9 (perf/cost1), performance from the non-updated chart (which only showed raw weighted performance): f(x)= 0.957/0.804* (450+x)/(1300+x)

 

red: Enderman's example of the two GPUs: g(x) = 100/10*(35+x)/(750+x)

 

As you can see in the graphs depending on the x-value (the cost of the whole system - CPU+mb cost) the overall cost-performance ratio can be lower or higher than 1. 1 is where cost-performance is equal on both setups. For the GPU example, the more expensive GPU actually hits breakevenpoint relatively quickly. For the actual i7 vs i9 cost/perf the breakeven point is at about $4,100, but the first $1000 of extra system cost raises relative value of the i9 compared to the i7 from 40% to 75%.

 

This actually might be really useful to decide which CPU+mobo to buy.

 

plot_price_performance.png

Link to comment
Share on other sites

Link to post
Share on other sites

If somebody is looking at a performance per dollar graph of a CPU, it's probably because they're interested in figuring out which CPU to buy. There's no reason to dilute things by adding the cost of graphics cards, power supplies, or things which aren't relevant to the CPU you're buying.

I mean, unless one CPU requires a beefier power supply, or a different type of RAM or something, then you might be justified in including them in the cost. But otherwise, no.

 

Also, including the motherboard cost into the equation could also get somewhat messy, because there are different motherboards with different costs for each platform.

 

As far as readability of the graphs go, it would probably be better to use different colours for the different CPU's, or at least colour them by brand.

Link to comment
Share on other sites

Link to post
Share on other sites

1 minute ago, Knowbody said:

If somebody is looking at a performance per dollar graph of a CPU, it's probably because they're interested in figuring out which CPU to buy. There's no reason to dilute things by adding the cost of graphics cards, power supplies, or things which aren't relevant to the CPU you're buying.

I mean, unless one CPU requires a beefier power supply, or a different type of RAM or something, then you might be justified in including them in the cost. But otherwise, no.

You don't understand.

The price performance depends on the cost of the WHOLE system.

If you calculate price performance just using the cost of the CPU/GPU you could be buying WORSE price performance.

 

Look at my GPU example, it shows exactly why putting a $35 GPU in a $1000 PC is a bad choice.

A $700 GPU in a $1000 PC ($1700 total) would be better price performance, even though when you take only the cost of the GPU it seems like worse price performance.

NEW PC build: Blank Heaven   minimalist white and black PC     Old S340 build log "White Heaven"        The "LIGHTCANON" flashlight build log        Project AntiRoll (prototype)        Custom speaker project

Spoiler

Ryzen 3950X | AMD Vega Frontier Edition | ASUS X570 Pro WS | Corsair Vengeance LPX 64GB | NZXT H500 | Seasonic Prime Fanless TX-700 | Custom loop | Coolermaster SK630 White | Logitech MX Master 2S | Samsung 980 Pro 1TB + 970 Pro 512GB | Samsung 58" 4k TV | Scarlett 2i4 | 2x AT2020

 

Link to comment
Share on other sites

Link to post
Share on other sites

Plotting all CPUs relative to the i9 ($1000 + $300) with weighted performance according to the pre-fixed chart of Linus' video with the function described above:

 

f(x) = relative performance cost of i9 7900X compared to other CPUs

 

blue: i7 7700k OC ($330 + $100) green: i7 6950x ($1400 + $300) red: R7 1800X ($500 + $100) gray: R7 1700 OC ($300 + $100)

 

 

plot_price_performance_all.png

Link to comment
Share on other sites

Link to post
Share on other sites

Okay, so I didn't read too closely into @Enderman's post since it was my lunch break, but insults aside, that IS a useful metric, but only when paired with a graph like @patthehat's. Otherwise, we could slide the scale any way we pleased, which is why we haven't done something like this previously. For example, adding a big SSD, a monitor, keyboard, mouse, what have you, x keeps rising; Any single point on the curve outside of the y=1 is useless information, which is why I dismissed the original post.

 

I'm not sure if this is something we can properly convey for the purposes of a video, however, since I'm not sure how easy it would be to parse to the average viewer. The intent of the original performance per dollar graph was mainly to show whether or not that single component's level of performance across testing was worth what they're asking for it, while finding the "price to performance sweet spot for a given budget" is kind of a different thing, albeit very valuable information. It'll take a bit of consideration and trial-and-error to make it work.

Emily @ LINUS MEDIA GROUP                                  

congratulations on breaking absolutely zero stereotypes - @cs_deathmatch

Link to comment
Share on other sites

Link to post
Share on other sites

also @GabenJr can we talk about the powerdraw and temp figures for a second? they are stupendously high 

RyzenAir : AMD R5 3600 | AsRock AB350M Pro4 | 32gb Aegis DDR4 3000 | GTX 1070 FE | Fractal Design Node 804
RyzenITX : Ryzen 7 1700 | GA-AB350N-Gaming WIFI | 16gb DDR4 2666 | GTX 1060 | Cougar QBX 

 

PSU Tier list

 

Link to comment
Share on other sites

Link to post
Share on other sites

Just now, Space Reptile said:

also @GabenJr can we talk about the powerdraw and temp figures for a second? they are stupendously high 

They are. Recorded under AIDA64 load and with a Kill-A-Watt at the wall.

Emily @ LINUS MEDIA GROUP                                  

congratulations on breaking absolutely zero stereotypes - @cs_deathmatch

Link to comment
Share on other sites

Link to post
Share on other sites

Just now, GabenJr said:

They are. Recorded under AIDA64 load and with a Kill-A-Watt at the wall.

jesus , that might be something worthy to point out in reviews , thats impossible to keep cool 

RyzenAir : AMD R5 3600 | AsRock AB350M Pro4 | 32gb Aegis DDR4 3000 | GTX 1070 FE | Fractal Design Node 804
RyzenITX : Ryzen 7 1700 | GA-AB350N-Gaming WIFI | 16gb DDR4 2666 | GTX 1060 | Cougar QBX 

 

PSU Tier list

 

Link to comment
Share on other sites

Link to post
Share on other sites

5 minutes ago, Enderman said:

You don't understand.The price performance depends on the cost of the WHOLE system.

The graphics card is not relevant to the price:performance of the CPU. It just dilutes things to include it.

 

If you want to know whether you should spend more on a CPU, or more on a GPU, that's a separate question and that would warrant looking at the total cost of the CPU and GPU, as well as performance in the particular scenarios you're interested in.

Link to comment
Share on other sites

Link to post
Share on other sites

2 minutes ago, GabenJr said:

Okay, so I didn't read too closely into @Enderman's post since it was my lunch break, but insults aside, that IS a useful metric, but only when paired with a graph like @patthehat's. Otherwise, we could slide the scale any way we pleased, which is why we haven't done something like this previously. For example, adding a big SSD, a monitor, keyboard, mouse, what have you, x keeps rising; Any single point on the curve outside of the y=1 is useless information, which is why I dismissed the original post.

 

I'm not sure if this is something we can properly convey for the purposes of a video, however, since I'm not sure how easy it would be to parse to the average viewer. The intent of the original performance per dollar graph was mainly to show whether or not that single component's level of performance across testing was worth what they're asking for it, while finding the "price to performance sweet spot for a given budget" is kind of a different thing, albeit very valuable information. It'll take a bit of consideration and trial-and-error to make it work.

Talk to linus and see if he can make a techquickie on it so that normal people can understand.

Just because a $35 GPU has great "price performance" (when calculated using only the GPU cost) does not mean it is good to pair it with a $1000 system.

 

The more expensive the system is, the better price performance you get by buying a more expensive GPU/CPU.

NEW PC build: Blank Heaven   minimalist white and black PC     Old S340 build log "White Heaven"        The "LIGHTCANON" flashlight build log        Project AntiRoll (prototype)        Custom speaker project

Spoiler

Ryzen 3950X | AMD Vega Frontier Edition | ASUS X570 Pro WS | Corsair Vengeance LPX 64GB | NZXT H500 | Seasonic Prime Fanless TX-700 | Custom loop | Coolermaster SK630 White | Logitech MX Master 2S | Samsung 980 Pro 1TB + 970 Pro 512GB | Samsung 58" 4k TV | Scarlett 2i4 | 2x AT2020

 

Link to comment
Share on other sites

Link to post
Share on other sites

4 minutes ago, Knowbody said:

The graphics card is not relevant to the price:performance of the CPU. It just dilutes things to include it.

The point of knowing the price performance is to decide which CPU (or GPU) is more worth getting.

Not including the total system price in this calculation will give you incorrect results, such as thinking a $35 GPU has better price performance than a $700 one.

 

This "price performance" graph in this video is for CPUs, all I'm saying is that this applies to both CPUs and GPUs. You don't do both at the same time.

 

The TOTAL SYSTEM PRICE is relevant to the price:performance of the CPU.

It does not "dilute" anything, it gives correct information on which is a better purchasing decision.

NEW PC build: Blank Heaven   minimalist white and black PC     Old S340 build log "White Heaven"        The "LIGHTCANON" flashlight build log        Project AntiRoll (prototype)        Custom speaker project

Spoiler

Ryzen 3950X | AMD Vega Frontier Edition | ASUS X570 Pro WS | Corsair Vengeance LPX 64GB | NZXT H500 | Seasonic Prime Fanless TX-700 | Custom loop | Coolermaster SK630 White | Logitech MX Master 2S | Samsung 980 Pro 1TB + 970 Pro 512GB | Samsung 58" 4k TV | Scarlett 2i4 | 2x AT2020

 

Link to comment
Share on other sites

Link to post
Share on other sites

3 minutes ago, Space Reptile said:

jesus , that might be something worthy to point out in reviews , thats impossible to keep cool 

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.

 

Just now, Enderman said:

Talk to linus and see if he can make a techquickie on it so that normal people can understand.

Just because a $35 GPU has great "price performance" (when calculated using only the GPU cost) does not mean it is good to pair it with a $1000 system.

 

The more expensive the system is, the better price performance you get by buying a more expensive GPU/CPU.

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.

Emily @ LINUS MEDIA GROUP                                  

congratulations on breaking absolutely zero stereotypes - @cs_deathmatch

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×