How 3DMark works: Simplified

[teethri]

This discussion is focused on how the overall 3DMark score is calculated from the two component scores. Other discussions have talked about what the component scores for each part are.

The simplest explanation for how your total 3DMark score is calculated from the two component scores is that the graphics score has 85% weight while the CPU score has 15% weight. (The component scores are graphics, which is based on your graphics card AKA GPU, and CPU, which is based mostly on your CPU but is also affected by how many memory sticks you have and the RAM MHz.) You might think this means that the total score is just 85% of the graphics score plus 15% of the CPU score, but it's slightly more complicated than that.

The actual total score you get is:

T = GC / (.15G + .85C)

Here T is the total score, G is the graphics score, and C is the CPU score. (I originally used the letter T because this is the Target overall score we are trying to reach, but it makes more sense to think of T as the Total score.)

The fancy name for this way of calculating the total score is to call it a "weighted harmonic mean", as opposed to the simpler arithmetic mean you probably studied in school.

What does this mean as a practical matter? Well, for starters, as you would expect, if your CPU score and GPU score are equal, your total score will equal that as well. But let's dig a little deeper.

Another very important implication is that the ratio of the total score to the graphics score cannot exceed 1/.85 = 1.17647. This, more than anything else, helped drive home for me how crucially your total score is tied to the graphics score - if your graphics score is 1000, your total score cannot go higher than 1176, no matter how overpowered your CPU is. The ratio of overall score to CPU score cannot exceed 1/.15 = 6.66667.


If you already know the CPU score of the CPU you are going to use, to reach a certain target, you need a GPU with a 3DMark component score of at least
G = .85CT / (C - .15T)
Where C is the CPU score and T is the total score. So just multiply the target score and the CPU score together, and divide that result by an amount equal to the CPU score minus 15% of the target score.
Symmetrically, if you know your GPU score for the graphics card you plan to use, your CPU score needs to be at least
C = .15GT / (G-.85T)

[In this post, I will put certain things in italics. If you don't understand what is written in italics, just skip the part that is in italics and the post will make just as much sense without it.]

The partial derivative of the log of the total score with regard to the log of the GPU score - in other words, the percentage by which the total score increases when the GPU score increases 1% - is
1 - [.15G / (.15G + .85C)] , which equals 1 - (.85T / C).
Symmetrically, the partial derivative of log total score over log CPU score is 1 - (.85T / G).
What this means is that, in the special case where our graphics score and CPU score are equal (meaning the total score is also equal), a 1% increase in the graphics score will increase our total score by .85%, while a 1% increase in the CPU score will only increase our total score by .15%.
Using a larger, 10% increase may be easier to think about. If our CPU and GPU scores are both 1000, our total score is 1000. If we increase the GPU score to 1100, the total score goes up to 1085. But if we leave the graphics score back at 1000 and increase the CPU score to 1100 instead, the total score only goes to 1015. (The actual total scores are 1083.7 and 1013.8, not 1085 and 1015, because the rule that a 1% increase in GPU score increases the total score by .85% only applies when the scores are equal, and as we went up by 10% the scores were no longer equal. In other words, the local derivative when the scores were equal was .85, but the average derivative over the interval was smaller than that.)

The partial derivative of log total score with respect to log GPU score is equal to the partial derivative of log total score with respect to log CPU score when G/C is equal to .85/.15 = 5.6667. If our build is heavily stacked toward GPU so that the GPU score is 5.66667 times larger than the CPU score, a 1% increase in GPU score will have the same effect on the total score as a 1% increase in CPU score. Specifically, either change will increase the total score by 0.5%. So, if we pretend for a moment that 5.667 equals 5.7, imagine a computer with a GPU score of 5700 and a CPU score of 1000. (The total score would basically be 3333.) In this case, increasing the GPU score by 1% to 5757 would have the same effect on the total score as increasing the CPU score by 1% to 1010. Since increasing CPU score from 1000 to 1010 would be much cheaper than increasing GPU score from 5700 to 5757, we should do that if we need a higher score, or just not use a build that is this heavily stacked toward GPU.

So, in the example right above this, raising the GPU score by a certain percentage raised the total score by the same amount as raising the CPU score by that same percentage (as opposed to when the scores were equal, where increasing GPU was 5.667 times as valuable). But that was a 57 point increase versus a 10 point increase. Is there a ratio where raising the GPU score by a certain amount raises the total score by just as much as raising the CPU score by the same amount? There is! Instead of setting the ratio of GPU to CPU to .85/.15, set the ratio of GPU score squared to the CPU score squared to .85/.15. In other words, instead of using .85/.15 = 5.667 as the ratio, use the square root of that, which is 2.38047. So, if our GPU score is 2380 and our CPU score is 1000, increasing the GPU score by a few points will have the same effect on the total score as increasing the CPU score by that many points. In fact, the effect will be half - increasing either score by 8 points would increase the total by 4 points; increasing either score by 10 points would increase the total by 5 points, etc. (The overall score when GPU is 2380 and CPU score is 1000 would by 1971.8.)

This means that if the cost of increasing CPU score by a certain number of points and the cost of increasing GPU score by that many points were equal (it's not), we might want to keep our GPU score equal to 2.38 times the CPU score.


As a practical matter, the cost of increasing our GPU score is actually higher than the cost of increasing our CPU score. Assuming that the lowest-cost CPU and the lowest-cost GPU are not good enough, what matters for keeping costs down is how much it costs to increase our GPU score versus how much it costs to increase our CPU score. For simplicity, let's assume that there is a fixed ratio between how much it costs to increase GPU score by a certain number of points and how much it costs to increase CPU score by a certain number of points. If you look at it, you can get about 17.2 additional points on your CPU per dollar versus about 10 additional points on your GPU per dollar (this is based on a level around 11-12; you can get slightly better bang for buck when you level up in the game and unlock more parts). Suppose the ratio of how much it costs to increase our GPU score to how much it costs to increase our CPU score is called r. Then, to maximize 3DMark score at a low cost, we should set our GPU score to CPU score ratio to the square root of .85/(.15r). I chose to use 1.735 as the value of r. So, this means my GPU to CPU ratio should be 1.807. If my CPU score was somehow 1 and my GPU score was 1.807, my overall score would be r / (.15*r + .85), so my total score would be 1.612. So, if my GPU score is 1.807 times the CPU score, the total score would be 1.612 times my CPU score. This gives me a very practical result: My target CPU score is 62.04% of the overall 3DMark score target, while my target GPU score is 112.11% of the overall 3DMark score target. If I exceed those targets, I will get an affordable build that exceeds the total target 3DMark score.

As a practical matter, if your target overall score is below 3000, and especially if it's below 2200, you can probably ignore ratios and just use the cheapest CPU, a Celeron 3930, if your goal is just to get a good 3DMark score cheaply. A Celeron 3930 with one stick of 2133MHz RAM has a CPU score of 1392, by the way. Then use the G=.85CT/(C-.15T) formula to find out how good your graphics card needs to be.




Note that everything in this post is based on the actual 3DMark scores of your CPU and GPU, which have been posted in other discussions like this one, and not the numbers listed in the parts ranking table in-game. I tend to multiply the number in the parts ranking table for GPU by 1.46, and the number in the CPU table by 2.62, to get an estimate. The actual 3DMark component scores are usually at least as high as the number I get, and often a fair bit higher.


This post is focused on how the overall 3DMark score is calculated from the component scores, not how the individual component scores are calculated. However, people may be very interested to know that the worst cards for crossfire are the vegas, which sometimes have a score for two cards that is about 1.79 times the score for a single card. For SLI, the score for two cards is always at least 1.83 times the score for a single card, and for crossfire other than vega the score for two radeons is always at least 1.86 times the score for a single card. For the "best" cards for use in multi-card setups, which are the MSI 980 Ti Gaming 6G LE, followed by the followed by the DFL 280 GR8 3G, followed by the Asus Strix 560 OC Ed., the ratio is higher than 1.98, meaning that having two cards really is basically double the power of having one.

People may also be interested to know that increasing from 1 ram stick to 2 on, say, a Ryzen 1600X, can increase the CPU score by over 11%. Increasing from 2133MHz RAM to 2400 MHz RAM can increase it by over 3%. It's usually best to just get one stick of 2GB 2133 RAM if there's no RAM requirement and your goal is to keep costs down, since that money is better spent on GPU. You can get two sticks and go up to 2666 MHz or maybe even 2800MHz if you want to, but it's usually not worth it to go higher than that for 3DMark if your budget or your client's tight. The total amount of RAM does not affect 3DMark, just the number of sticks and the MHz.


By the way, the official rules for how the real Time Spy 3DMark score is calculated are contained in a document called 3DMARK Technical Guide. (I have omitted the link since it is a pdf download and I think that hotlinking a download on a forum could eat up the host's bandwidth.) They do in fact use the formula I described above, although they refer to .85 as Wgraphics, .15 as Wcpu, the graphics score as Sgraphics, and the CPU score as Scpu. They use the equation that the total score equals
[Wgraphics + Wcpu] / [ (Wgraphics / Sgraphics) + (Wcpu / Scpu) ]
However, when you note that the scores add up to one, so the numerator is 1, that means that the whole thing is equal to 1 / (.85/SGraphics + .15/Scpu), and then rearrange the fraction, you get that the total score is equal to (Sgrpahics*Scpu) / (.15 * Sgraphics + .85 * Scpu). This is the same as the formula I gave above if the labels C and G are used in place of Scpu and Sgraphics. A shortened version of the Time Spy animation is actually shown in game. The exact same formula is used for the real 3DMark Night Raid score (which is explicitly stated to be rounded down).

Fire Strike and Sky Diver 3DMark scores, which are not used in PC Building Simulator, include a physics score, and give 75% weight to graphics, 15% weight to physics, and 10% weight to a "combined" score based on a test that uses both CPU and GPU.

[🦊Λℚ𝓤ΛƑΛᗯҜᔕ🦊]

Good write up with lots of useful information. I just wanted to add in a little information so people know: On most Desktop systems in-game, the cpu score does not increase when adding a 3rd or fourth memory stick to these systems because they are dual channel systems and only scale to two ram sticks, or two channels. It's the "HEDT" class systems (AMD ThreadRipper and Intel X299) that scale up with a 3rd and forth ram sticks because those are Quad-Channel (4 channel) systems. In the HEDT systems, just like the desktop systems though, ram performance does not increase beyond 4 ram sticks installed in these systems, 4 channel ram means 4 max. Even if you install a 5'th or 6'th or 7'th or 8'th stick it won't be any faster than having 4 in there.

What I would be really curious about (and maybe teethri can educate us there.. I don't have the time to get the data myself): What does the percentage increase in cpu score look like for quad channel systems in-game?

[teethri]

Based on Wow Actual's numbers, threadripper seems to be very sensitive to RAM frequency. The 1900x gains slightly more than 1 point of CPU performance per additional MHz, the 1920x gains more than 1.43 points of CPU performance per additional MHz, and the 1950x gains almost 1.6 points per MHz! On the other hand, adding a stick doesn't seem to make an enormous difference; each additional stick adds about the same performance as adding 266 MHz.

Still, in percentage terms, at least when you're at low MHz, you're getting about 6.8 to 8.7% additional performance per extra stick with the 1920x and 1950x (smaller benefit with the 1900x). The benefit of adding sticks for threadripper seems to be a consistent amount of points that each CPU gains rather than a consistent percentage benefit, so the percent benefit for adding more sticks gets smaller as you go up in MHz and your total score gets higher. Again, for threadripper, adding a stick is roughly equivalent to just going up one big step in MHz (even though for most other processors the benefit of adding a stick is considerably larger than a MHz step; for many, you would have to go from a single 2133 stick to a single 3000 or 3200 stick for it to be better than two 2133 sticks), and the number of extra points of performance each CPU gets from extra MHz is basically a CPU-specific multiple of the number of extra MHz.

I willl confess to not having played around with threadripper benchmarking myself; I'm playing with a spreadsheet of Wow Actual's numbers.


That's how much it increases your CPU component score; as described above, that's only about 15% of your total 3DMark score when the CPU and GPU scores are equal. If you slant the build toward GPU, as you should if 3DMark is your focus, the importance ratio changes from .15 to .85 to something more like 36.56% versus 63.44% when measuring how big the effect of a certain number of extra points to each component score is (meaning that GPU is still more important, but less than twice as important). Or, the CPU to GPU importance ratio is 24.18% to 75.82% in terms of the relative importance of raising a component score by a certain percentage of what that score already is (meaning that if the GPU score is already larger we're looking at a proportionate increase, so the GPU increase is 3 times as important in the percent effect on the total score but it's a bigger increase). This is due to the way the harmonic mean reflects CPU bottlenecking, and is based on the 62.04% / 112.11% numbers in the first post.


I posted in Snuke's thread about "Formula that calculates CPU score with ram impact" about the role of RAM frequency as well. I focused on frequency there, even though for most CPUs (threadripper not included) adding a second stick usually makes a bigger difference than one or even three incremental MHz boosts.

[🦊Λℚ𝓤ΛƑΛᗯҜᔕ🦊]

Originally posted by teethri:

Based on Wow Actual's numbers

Just to be clear here.. you are talking about PCBS and not real life in all of your posts in this thread, right? Because there is no "Wow Numbers" that I'm aware of relating to performance in PCBS. And you can't compare real life 3dmark performance to 3dmark performance in PCBS. It's not 1:1 with real life, the performance in 3dmark in-game is vastly different than what the same hardware would perform in real life.

[teethri]

Wow! Actual's PC Building Simulator Guide - Benchmarking Lookup Tables and Caclulators [sic]

[BonPadre]

Originally posted by teethri:

Wow! Actual's PC Building Simulator Guide - Benchmarking Lookup Tables and Caclulators [sic]

just a head up... this calculator is outdated (maybe even obsolete ?)
It was made for version v0.8.6.1 (like the title implies)
And we are now in v1.3.1

So I don't know how much of your post is related to variations in numbers between the v0.8.6.1 & v1.3.1, but you should really, if you want to look for things, use BiuBiu's calculator that is up to date
https://steamcommunity.com/sharedfiles/filedetails/?id=1596577871
Way faster in spitting results than the sheet you're using (that I used in the past too btw, when it was up to date)

And the link to BiuBiu's thread in case you want to communicate mistakes and such
https://steamcommunity.com/app/621060/discussions/0/3216031607491943889/

[teethri]

Originally posted by BonPadre:

So I don't know how much of your post is related to variations in numbers between the v0.8.6.1 & v1.3.1, but you should really, if you want to look for things, use BiuBiu's calculator that is up to date

Well, the first comment was about the equation for determining total 3DMark score from the component scores, which is based on the actual Time Spy scoring system, so the difference between game versions should have almost no bearing on what I said in the first comment. The only things it would affect are the part where I briefly mentioned the 3DMark score for the Celeron 3930, and whether or not my estimate that it would cost around 1.72 or 1.735 times as much to raise GPU score as CPU score is in the ballpark. For my later comments that got more specific about how RAM affects CPU score, the difference in game versions could be more important.

[BonPadre]

Also, an other head up, because honnestly I'm not a math guy, and most of what you say here is just a lost cause with me lol... (please remember that, it's important for what follows, to avoid some painful facepalm )

But with your thing about value of upgrading CPU vs GPU... just remember that, later game, when you unlock GPU tuner and can OC GPU, as you can push them way more than CPU OC, and due to their 85% "weight" of the score, all that value thing just goes to trash nope ???

I mean, you can be happy if you oc a CPU with +400MHz, so the impact on the score is still "limited", while on the other hand, you can go so much higer with GPU's and thus their cost vs price when OC is totally crushing CPU's nope ? I mean, I always looked at CPU's to just not penalize too much the score when you need to reach a given benchmark score.

Again, it's just a question I offer to you, and that's just to add "fuel" to your thinking, but I don't need an answer for me, as :
- I'm not a math guy
- Budgets are now so ridiculously high, I just don't see the point of saving on costs for the few hundreds you can save early game, and that anyway don't go into your bank account.

So again, for my "limited" undertsanding in your math, while I applaud the effort of your calculations and typing, I just don't see a use for me in game, and also nor do I see a use for the higest benchmarck scores, as it's more about binning known parts, due to silicon lottery, than understanding how much this or that "lower" tier parts could impact this or that score for a few hundreds of points.

**************************************************

Also, you can clerely see here in my comments, I have a slight "issue" understanding mybe 1/5 of your initial post, while your title is...
How 3DMark works: Simplified....
simplified... really ?

To me simplified means :
GPU score = 85% of the benchmark score
CPU + RAM impact = 15% of the benchmark score

The higher RAM speed I can use will give my base CPU score a few more points (and as it's exponantial with the CPU score, with low scores CPU, no need to go crazy on RAM speed, as it will still be limited, nor do I need an expansive Mobo that can give me better RAM max speed)

So the quickest and simpliest way to increase benchmark score to a SIGNIFICANTLY higher score = a GPU upgrade.

END

That's simplified to me.

**************************************************

Also I'm still at loss with you saying that :

Originally posted by teethri:

As a practical matter, the cost of increasing our GPU score is actually higher than the cost of increasing our CPU score.

Because to me, you are probably very right if you build everything from scratch, but nowhere in your post you mention if all your theory is about overall gameplay, or only "new build" theory build.

So due to the fact that, when you have a given CPU "architecture", let's say a Kaby Lake, you are limited to a CPU that will have a max base score of 4886 with a single RAM stick of 2133MHz.
Then if you want to have a better CPU score (stil with 1x 2133MHz RAM stick reference), you need to swap the Mobo as well... Adding to the cost.

So example of a possible mid game upgrade job

I have a build, with a :
Mortoni H170 Mtech (limited to 2666MHz RAM speed)
A single RAM stick of 2133MHz
An Intel Celeron G3930
A MSI GeForce GTX 960 GAMING 4G
Score of the build
CPU score : 1392
GPU score : 2225
Total score : 2041
Est. Wattage : 368

Now let's say the game asks me to upgrade that PC to a 2422

Now to reach that score you need an Intel Core i7-7700K = $270
But to reach said score a GPU of $265 does the job, like a MSI GeForce GTX 970 GAMING 4G for a total score of 2674.

So in this example the upgrade for a given target is $5 cheaper the the CPU upgrade. Specially since the GPU becomes the bottleneck if not upgraded.

Then with that same example you come to a bottleneck point, where if you're not upgrading the mobo, you will be still very limited to a 5456 score for the CPU, no matter what, due to the limitation of the RAM speed of Mobo, and the best Kaby Lake CPU possible.

When you have upgrade jobs, it's so situational that I porobably don't need so many calculations as yours to understand that :

If it's the original GPU that will bottleneck the overall score (like a stupid 1050), due to it's "weight" in calculation, I have no better choice than to upgrade GPU, or my "$ value" from CPU upgrade will be completely negated by the GPU. (just like in my previous practical example)

If the CPU is a VERY low end CPU, and I need to raise TOTAL score by about 20 points, indeed I can probably get my score with a second RAM stick and/or quicker RAM.

If I need to raise total score by +100pts to +300pts, probably a better CPU will make it.

If I need a score increase that goes from +500 and higher, I probably will get there with the GPU upgrade.

SINGLE GPU scores are never impacted by the Mobo, while on the opposite, CPU scores are very dependent of Mobo (OC able for CPU IF CPU is unlocked for Intel, so yeah, there is also that... AMD or Intel build ? Why ? Then all AMD mobo are limited to a max 3600MHz Ram speed in game), the mobo will cap the max RAM speed too.

So again, if upgrade jobs, it's so situational of original customer's PC build, that so often you are probably better of simply don't care much and upgrade GPU anyway, specially now that budgets are a NON limiting factor, you have NO incentive to spare a few hundreds on builds and upgrades (as your only income is anyway only labour)

So as the game is as much about upgrading as building new from scratch, I feel (again, FEEL, because honnestly I don't understand 1/5 of your original post I think ?) your text is more a "theory in a vaccum" that works probably 100% if building from scratch, but has it's limitations when it comes to upgrading.

Only thing I know, I have never "overthought" the game, even when budgets were WAY tighter than they are now, I had no issues building and upgrading within budgets, and once more, as your ONLY INCOME is from labour, I hardly see a use of "penny pinching" for this game, and moreover now that budgets have been stupidly raised to an unchallenging level of any sorts.

When Budgets were tight, you always had 2 requests possible for upgrades, aka, install AIO or more RAM and then a given better score.
The trick was to always buy the CHEAPEST request (AIO or RAM) and then deduct that from parts budget, and buy the best available GPU with remaining budget. That never ever failed me to reach the target requested score.

Sorry for wall of text, and again, it's just fuel for your thoughts, no need to try to explain me more, because I probably will not understand half of what you say XD

[teethri]

When I said that increasing GPU score is more expensive than increasing CPU score, I meant that if I need to buy both a CPU and a GPU, and I choose to spend extra money to switch from the GPU I'm considering buying to a better GPU, that costs a lot of money for a few extra points on the GPU subscore (about 10 extra points per extra dollar on the price of the GPU) compared to how picking a more expensive CPU can get me (about 17 extra points per extra dollar on the price of the CPU). You should still get a better GPU anyway, though, since it's such a huge part of the score. If it's an upgrade job and it only makes sense to upgrade one of them (for example, if you use the 62.1% / 112.2% numbers and either the CPU or GPU is already better than your target subscore), then just use this formula to plug in for what the other subscore needs to be:

G = .85CT / (C - .15T)
Or
C = .15GT / (G-.85T)

So if you know that you're keeping a Ryzen 1600 and just need a GPU upgrade, and you know the CPU component score for that Ryzen, you can just multiply .85 times the target total score times that CPU score, divided by the CPU score minus .15 times the target total score, and that's what your GPU score needs to be.

I agree that having a guide on how to penny pinch builds that target benchmarks in this game may not be necessary, since consumer budgets tend to be way bigger than necessary.

As I mentioned in Reply #2, although the GPU is 85% of the score when the two scores are equal, if your build is already slanted toward GPU, increasing the GPU score further may only account for 63% of the amount of potential increase you can get by increasing the subscores, not 85%.

[🦊Λℚ𝓤ΛƑΛᗯҜᔕ🦊]

Originally posted by teethri:

When I said that increasing GPU score is more expensive than increasing CPU score, I meant that if I need to buy both a CPU and a GPU, and I choose to spend extra money to switch from the GPU I'm considering buying to a better GPU, that costs a lot of money for a few extra points on the GPU subscore (about 10 extra points per extra dollar on the price of the GPU) compared to how picking a more expensive CPU can get me (about 17 extra points per extra dollar on the price of the CPU). You should still get a better GPU anyway, though, since it's such a huge part of the score. If it's an upgrade job and it only makes sense to upgrade one of them (for example, if you use the 62.1% / 112.2% numbers and either the CPU or GPU is already better than your target subscore), then just use this formula to plug in for what the other subscore needs to be:

G = .85CT / (C - .15T)
Or
C = .15GT / (G-.85T)

So if you know that you're keeping a Ryzen 1600 and just need a GPU upgrade, and you know the CPU component score for that Ryzen, you can just multiply .85 times the target total score times that CPU score, divided by the CPU score minus .15 times the target total score, and that's what your GPU score needs to be.

I agree that having a guide on how to penny pinch builds that target benchmarks in this game may not be necessary, since consumer budgets tend to be way bigger than necessary.

As I mentioned in Reply #2, although the GPU is 85% of the score when the two scores are equal, if your build is already slanted toward GPU, increasing the GPU score further may only account for 63% of the amount of potential increase you can get by increasing the subscores, not 85%.

One thing I think you aren't factoring in for upgrade jobs is keeping used parts. Sometimes you could potentially for example: Swap out a customer's i7-7700K for a dual core pentium chip, invest in a big video card, overclock said video card, still meet their objective and then you would get a 7700K chip to keep as a used part that you could in turn later put in some other customer's upgrade job for free and make a huge profit there. Or use the 7700K in some other customer's "Build me a computer" job and also put that part in there free.

[teethri]

So, I found out something remarkable:

Once you reach level 18 and unlock the GIGABYTE B450 AORUS M
, which is a low-cost AM4 motherboard that allows multi-GPU support, literally EVERY target score between 3606 and 16239 is achieved most cheaply by using that motherboard, a Ryzen (which is what that motherboard is for), and a dual graphics card solution. Now, there are some caveats here. First, this is only considering CPU, GPU, motherboard, and RAM cost, not power supply or case fan cost, so that's obviously going to bias the results toward multi-GPU solutions. Second, at the extreme high end, you might need some graphics cards that are not available at level 18, like the AORUS 2080 you get at level 25 or an 11 gigabyte version of the 1080 that you get at level 21. But even with the pro multi GPU bias of not considering PSU costs, I was stunned that, as far as I can tell, every target score in that range is achieved most cheaply with Ryzen and a dual graphics card solution. In many cases, you can still get the exact same score with the exact same CPU, GPU pair, and RAM without that motherboard at a higher price before level 18 with a different motherboard that supports multi-GPU, but I haven't checked whether it's still the cheapest option. I haven't actually checked every score in that range, but I did check every budget from 565 to 1435 (as the budget for the parts included in this comparison), and it always comes up Ryzen and double graphics card. Jumping up to higher budgets like 2085 and 2705, it's still Ryzen and double graphics card.

For example:
AMD Ryzen 3 Quad Core 1200 1 RAM 2133 DFL Radeon R9 280 GR8 3G 2 GPU GIGABYTE B450 AORUS M Mortoni Value Supreme 2GB 2133 Mhz 565 Cost 465 Watts Score 3724 Level 18


You don't have to worry about your client's electricity cost, but I'm pretty sure that once you factor in PSU and case fans, using two graphics cards versus one is basically break even.

Most of the target scores of 3605 or lower can be achieved at the same cost at any level that is 5 or higher (as in, the cheapest build and its cost does not change as you level up once you've reached level 5) and involve using a Celeron or Pentium.

[scrane5757]

how do you explain that 3dmark can't get 2060 but the game pushes card to 2190?

[scrane5757]

rtx 3060 12 gig btw

[scrane5757]

game as in plural.

[🦊Λℚ𝓤ΛƑΛᗯҜᔕ🦊]

Originally posted by scrane5757:

how do you explain that 3dmark can't get 2060 but the game pushes card to 2190?

Originally posted by scrane5757:

rtx 3060 12 gig btw

Originally posted by scrane5757:

game as in plural.

This thread is old and dead as of 4 years ago and contains old information many updates ago. Nothing in this thread is relevant to today's version of the game because they changed how the game works in all the updates after this thread was written. In the future you might want to check the dates before you reply to something.

[Tanyon]

Man I really wish I could understand half the things you guys are saying.. haha.. I got to the point in the game where I can fix a lot of stuff and have no issues with customer support but wanted to start buying/making my own PCs to sell on the sale site to make profit and I was hoping for a guide to maximize profits but this is like Japanese to me lol.