[SOLVED] Does lowering memory latencies have positive effects on overall performance of the system if GPU is bottlenecking it?

nmvx

Prominent
Sep 16, 2019
7
0
510
I tried to find answer to the question in the title on various forums, but couldn't find anything except gaming benchmarks that showed that if GPU is bottlenecking the system then decreasing memory latency doesn't give any noticable FPS increase. But I'm not interested in gaming and FPS.

My question is, does decreasing memory latency and increasing RAM bandwidth have positive effects on the system performance when GPU is very weak and seriously bottlenecks it? What kind of benchmarks will show benefits of decreasing memory latency in this case?

As an example I have the following build:

Ryzen 5 1600 (stock),
gt 1030,
2x8 GB DDR4 clocked at 2800 CL18

I know the GPU is weak and bottlenecks the system, but I really don't need better GPU, because I'm not a gamer and I don't do video/photo editing.

I overclocked the RAM from 2666 CL20 to 2888 CL18. But there weren't any benchmark result improvment for PCMark 10. But the latency and bandwidth benchmarks in Aida64 showed around 5% improvent. But I don't know whether this improvevent is any good for my system?

I want to know, will any further ovcerclocking this RAM give any benefits to the system performance and if the answer is yes then what kind of tasks will benefit from further overclocking RAM? and how do I meaningfully measue these improvements in benchmarks?
I'm especially interested in these kind of performances: responsiveness of the system, faster app open/load times, faster windows load times, faster calculation in CAS programs like Mathematica/Maple etc.
I'm not interested in gaming benchmarks and FPS.
 
Last edited:
Solution
ryzen cpu performance is sensitive to ram performance.
Unfortunately, increasing ram speed usually comes with an increase in latency negating much of the boost.
Synthetic tests might look good, but actual app performance is not that much greater.
And... if you are overclocking, there will be an interaction such that backing off ram performance might be the price you pay to get a higher multiplier.

For faster file open times and overall quickness, a good SSD for windows will be a marvelous boost compared to a HDD.
I tried to find answer to the question in the title on various forums, but couldn't find anything except gaming benchmarks that showed that if GPU is bottlenecking the system then decreasing memory latency doesn't give any noticable FPS increase. But I'm not interested in gaming and FPS.

My question is, does decreasing memory latency and increasing RAM bandwidth have positive effects on the system performance when GPU is very weak and seriously bottlenecks it? What kind of benchmarks will show benefits of decreasing memory latency in this case?

As an example I have the following build:

Ryzen 5 1600 (stock),
gt 1030,
2x8 GB DDR4 clocked at 2800 CL18

I know the GPU is weak and bottlenecks the system, but I really don't need better GPU, because I'm not a gamer and I don't do video/photo editing.

I overclocked the RAM from 2666 CL20 to 2888 CL18. But there weren't any benchmark result improvment for PCMark 10. But the latency and bandwidth benchmarks in Aida64 showed around 5% improvent. But I don't know whether this improvevent is any good for my system?

I want to know, will any further ovcerclocking this RAM give any benefits to the system performance and if the answer is yes then what kind of tasks will benefit from further overclocking RAM? and how do I meaningfully measue these improvements in benchmarks?
I'm especially interested in these kind of performances: responsiveness of the system, faster app open/load times, faster windows load times, faster calculation in CAS programs like Mathematica/Maple etc.
I'm not interested in gaming benchmarks and FPS.
Aida64 benchmarks memory latency but the free version is time limited. You also get a latency benchmark in 1usmus' EXCELLENT tool, Ryzen DRAMCalculator:


The only problem is that since it's less well known (for this feature) it's less quoted and therefore a bit harder to compare to others. But the latency figures are still very close to the figures quoted by Aida and still very useful for seeing if you are improving things on your system as you tweak.

It makes sense that in an application, e.g., games, if your GPU is the bottleneck then anything you do to CPU or main memory won't help since, well, that's what being the bottleneck means. By the same token, in memory-intensive applications that are latency sensitive a weak GPU won't be much of a problem since it's not the bottleneck.

From my experience, and the comments made by other discerning tinkerers: even significant latency changes rarely have measurable effect in real world performance characteristics. But there are those who will swear otherwise, so as they say: YMMV.
 
  • Like
Reactions: nmvx
ryzen cpu performance is sensitive to ram performance.
Unfortunately, increasing ram speed usually comes with an increase in latency negating much of the boost.
Synthetic tests might look good, but actual app performance is not that much greater.
And... if you are overclocking, there will be an interaction such that backing off ram performance might be the price you pay to get a higher multiplier.

For faster file open times and overall quickness, a good SSD for windows will be a marvelous boost compared to a HDD.
 
  • Like
Reactions: nmvx
Solution