Formula that calculates CPU score with ram impact

[Snuke]

I want to know whether there is some kind of formula with which you can precalculate a 3d mark score more precisely by calculating the "ram influence" on the cpu score. So not only just 1 / ( 0.85/S_graphics + 0.15/S_cpu). What I found out is that the ram perf increase is linear (e.g. every 400 MHz ram increase in dual channel results in a 0.197% total score increase).

[🦊Λℚ𝓤ΛƑΛᗯҜᔕ🦊]

Time Spy Score = Wgraphics + Wcpu / (Wgraphics / Sgraphics + Wcpu / Scpu)

Where:
Wgraphics = The Graphics score weight, equal to 0.85
Wcpu = The CPU score weight, equal to 0.15
Sgraphics = Graphics test score
Scpu = CPU test score

(Simplified) Time Spy Score = 1 / (0.85/Sgraphics + 0.15/Scpu)

Quoted from this thread: https://steamcommunity.com/app/621060/discussions/0/1696043806564899052/

[Snuke]

Originally posted by Aquafawks:

Time Spy Score = Wgraphics + Wcpu / (Wgraphics / Sgraphics + Wcpu / Scpu)

Where:
Wgraphics = The Graphics score weight, equal to 0.85
Wcpu = The CPU score weight, equal to 0.15
Sgraphics = Graphics test score
Scpu = CPU test score

(Simplified) Time Spy Score = 1 / (0.85/Sgraphics + 0.15/Scpu)

Quoted from this thread: https://steamcommunity.com/app/621060/discussions/0/1696043806564899052/

That was not actually the point of my question. I wanted to know how you can exactly calculate the score with RAM. I already mentioned the basic calculation but I (or maybe we) don't know how you can calculate the exact score. For example, a system with i7-8700 and GTX 1060 FE with dual channel 2133 MHz RAM will result in a cpu score of 7556 instead of the calculated ≈7541 since ram influences the cpu score. (4025= 1/(0.85/3719+0.15/x))

[🦊Λℚ𝓤ΛƑΛᗯҜᔕ🦊]

Originally posted by Snuke:

That was not actually the point of my question. I wanted to know how you can exactly calculate the score with RAM. I already mentioned the basic calculation but I (or maybe we) don't know how you can calculate the exact score. For example, a system with i7-8700 and GTX 1060 FE with dual channel 2133 MHz RAM will result in a cpu score of 7556 instead of the calculated ≈7541 since ram influences the cpu score. (4025= 1/(0.85/3719+0.15/x))

Sorry, I misunderstood. The ram is wrapped in to the CPU score. You adjust/overclock/change the ram speed, then run 3dmark with the modified ram speed and you will generate a new cpu score. Then plug in that in to the equation. No one knows exactly what formula ram impacts CPU score (Or if they do know, they haven't shared with us in the steam forums yet). But I don't think it really matters. Just do like I said and it works.

[BonPadre]

You should probably try to reach to BiuBiu in that post
https://steamcommunity.com/app/621060/discussions/0/3216031607491943889/
As he has a pretty damn good calculator that factors the RAM channels, and speeds, it has to be know at some point or an other.

Hope he/ she will be of help for your question.
In all the time I roamed this forum, I admit, like Aquafawks said, I have never seen anyone mentionning what was the formulae for RAM in that calculation, and add to that I'm not a math guy, so I personally never dig that much into numbers.

[Snuke]

Originally posted by BonPadre:

You should probably try to reach to BiuBiu in that post
https://steamcommunity.com/app/621060/discussions/0/3216031607491943889/
As he has a pretty damn good calculator that factors the RAM channels, and speeds, it has to be know at some point or an other.

Hope he/ she will be of help for your question.
In all the time I roamed this forum, I admit, like Aquafawks said, I have never seen anyone mentionning what was the formulae for RAM in that calculation, and add to that I'm not a math guy, so I personally never dig that much into numbers.

I wouldn't consider myself a math guy either but still I tried to find the formula on my own. My current approach is that there's some kind of "ram constant" to be found where the calculation and the actual score are equal to each other. The closest you can get to that is 2400 MHz whose deviation of the real score vs. calculation is only 4 points (using the test setup mentioned in my last answer).
So maybe the formula could look like this: cpu score = database cpu score + (constant * linear percentage increase (e.g. if constant is 2350 MHz and the ram you're using is 3 GHz then it's 650 MHz difference so 1.625 * 0.197% so 0.32%. So in the end it's =database cpu score + 0.32% cpu score.
Dumb thing from my side was that I took the percentage increase from the total score instead of the cpu score. But it doesn't theoretically change anything as the gpu score stays the same. The real percentage increase would still be linear albeit higher increases.
I suggest you could draw something like a graph for the calculations (first finding a formula) and a graph for the real score and the point in which they both meet is the constant that's needed...? But yeah that's where I'm currently stuck at.
Furthermore, these tests were all on dual channel ram. Single and quad channel may differ from that and I'd have to test that out as well.

[teethri]

I am assuming the numbers from "Wow! Actual" are accurate. Here is a detailed look at the importance of RAM frequency.

For Kaby Lake Pentiums like the 4620, increasing the RAM frequency makes almost no difference in total score. It's like a 0.4% increase in the CPU score for the first step up!

For Ryzen, increasing from 2133 MHz to 2400 MHz will always increase the 3DMark component score for the CPU by at least 1.896%. That's how much it helps the 2600X. The biggest Ryzen percentage change for 2133 to 2400 is 3.291% for the 1800X.

If you include all the processors, the best fit line for predicting the CPU component score at 2400 MHz based on the score at 2133MHz is 1.0324x - 34.305. In other words, you take the old CPU score, multiply by 1.0324, then subtract 34 or 35. This is pretty flawed, since it actually predicts a score for the Celeron 3900 that is LOWER than the score at 2133MHz, but it's called a best fit line, not a perfect fit line. Similarly, the best fit line for increasing from 2400 MHz to 2666MHz is 1.0316x - 34.449. If you force the fit lines to use a zero intercept, it's a 2.47% increase in the CPU component score for 2133MHz to 2400 MHz and another 2.41% for 2400 MHz to 2666MHz.

The increases from 4600 MHz to 4700 MHz tend to produce small percentage increases. For each step besides that one, the smallest one-step percentage increase with Ryzen is 0.821% increase in CPU component score for increasing the 2600X from 4000 MHz to 4133 MHz.

The bigger RAM MHz steps, like 3733 to 4000, tend to produce bigger increases.

So basically, each processor has its own sensitivity to RAM frequency.

For similar CPUs like the i5-6400, the i5-6500, and the i5-6600, the changes are IDENTICAL. Not the percentage changes; the actual number of points by which the CPU score changes with each RAM frequency upgrade are EXACTLY the same. In fact, it is nearly a fixed exact multiple of the number of MHz by which the RAM increased: for every 2.778 extra MHz, these three processors will each get 1 extra point of CPU score. Although it can vary from 2.83 to 2.735, the number of extra points of CPU score each of these three processors will get at each step is always exactly the same as each other.

Aside from the Celeron 3900, which is extremely sensitive to RAM frequency but isn't worth using, the CPUs that are fairly sensitive to RAM frequency, in percentage terms, are
Ryzen 1200, 1400, 1600x through 1800x (including 1700), and 2700x
i5-6400 through i5-6600K
i5-7400 through 7600K
i7-8700
All of the above processors have CPU score increases from the first step up from 2133 to 2400 that round to 3%, meaning they are at least 2.5%. The most is 3.32% for the i5-7400.

In absolute terms (the number of points gained when you upgrade the RAM), the most sensitive CPUs are the Ryzen 2700 and 2700X (224 bonus points from 2133 to 2400 MHz), followed closely by the 1800X, followed by the 1700, followed by the 1700X. The Ryzen 1600X just barely beats the 8th-gen i7 processors in sensitivity (they get 156 points for going from 2133 to 2400). Next is the Ryzen 1600. The 8th-gen i5s just barely beat the 7th gen i5s and the Ryzen 2600 and 2600X, which gain 104 points. The low-end Ryzens, the 6th-gen i5s, and the 7th-gen i7s all gain between 50 and 100, while the i3s gain less and the Pentiums and most Celerons gain almost nothing; in particular, anything that is Kaby Lake and is below i5 gains very little.

All of the above is based on a single RAM stick.