Take the computer know-how, the love of games, and the interest in components, and mix them all together. It's a perfect recipe for diving into benchmarking. In this article, we'll explain how you can use games to benchmark your PC and what you can do to analyze the results.
Exactly
Now I only wish Techspot use higher performing RAM config when benchmarking their GPU, something like 4x8GB 3600mhz cas 14 DDR4 kit, I know these kits cost like 400usd but it's one way to alleviate the CPU bottleneck that the highest end GPU will run into at 1080p. The 3200mhz 2x8GB bog standard kit that Techspot is using doesn't really even represent the the mainstream PC gamer nowadays.
Test setup: Core i9 9900K + RTX 2080
This is with a RTX 2080 and the difference is already 10fps+ with the higher performing kit, imagine the 2080 Ti or next gen 3080 Ti
Now I only wish Techspot use higher performing RAM config when benchmarking their GPU, something like 4x8GB 3600mhz cas 14 DDR4 kit, I know these kits cost like 400usd but it's one way to alleviate the CPU bottleneck that the highest end GPU will run into at 1080p. The 3200mhz 2x8GB bog standard kit that Techspot is using doesn't really even represent the the mainstream PC gamer nowadays.
Test setup: Core i9 9900K + RTX 2080
This is with a RTX 2080 and the difference is already 10fps+ with the higher performing kit, imagine the 2080 Ti or next gen 3080 Ti
Sorry but I think you are living on another planet if you think 16GB DDR4 3200Mhz doesn't represent the mainstream PC today. Even if you look at the Top 10 best selling memory on Amazon, only 2/10 kits are 3600Mhz and only 3/10 kits are more then 16GB. So if current buyers don't even buy 32GB 3600Mhz, what do you think that all the buyers in the past years did? Compared to a mainstream gaming PC, 16GB 3200Mhz is the sweetspot. And the majority will still have slower RAM then that.
And even if 3600Mhz is the standard these days, the bottleneck in most games is very small. And from all sites to compare with, you pick graphics from TPU, wich doesn't even use 1% lows. TPU reviews aren't even usable for GPU/CPU comparisons because they only use AVG FPS... That might have been fine in 2010, but in 2020 we know that 1% lows give a FAR better view of the actual game experience you can expect, especially when you have hardware slower then the hardware used in the review.
Techpowerup reader will buy System A, Techspot reader will buy System B and will be much happier
This is something I can do. And have been putting off...As you can see in Techspot article, if you populate all 4 DIMM slots it should give you higher performance than 2x8GB 3000mhz even when you are stuck at 2666mhz, not to mention you can reduce the timings at 2666mhz too.
www.3dmark.com
In both scores, was the system set to use the XMP profile? If so, then the measurable difference between the use of 2 vs 4 DIMMs slots has been mixed in with the changes between the RAM speed and timings. If you can spare the time, try running the tests again, with the new memory kit, but between each run alter the number of DIMMs used - I.e. start with 1, then go to 2, etc.
Unless you only alter one variable in a test, you're never going to be able to tell for certain that the experienced change in performance is down to the aspect you're hoping to analyse.
Edit: It's worth being a little cautious when using 3DMark to compare changes that result in just a few % difference. Here are 3 runs I did this morning:
Result
There's a 6% difference between the lowest and highest CPU results, and yet the only thing between each run was rebooting the PC.
Yes, including blocking network traffic.
There is one other variable: amount of memory. But let's assume that there is no difference at all between the system parameters between the two tests (e.g. temperatures for all components is exactly the same, so there's no difference in clock speeds) and that the results are harmonic means of multiple runs (so the results represent the worse case scenario of the central tendency of the frame rate distribution), then the use of 4 DIMMs has increased the more graphically intense test of the two by 0.1%, so for GPU testing it's far too small to be significant enough.
www.techspot.com
Using a RAM configuration to raise performance just creates the same situation for the high end GPU - I.e. its cost-per-frame ratio at 1080p is artificially lower than it should be, because the test is CPU-limited (or more rather, RAM-limited). It's a bit of a no-win situation, though, as comparing a raft of GPUs at a given resolution will invariably result in the performance dependency being associated with the wrong component: at 1080p, a 2080 Ti won't look at much better than, say, a 1660 Ti but at 4K, it will be vastly better.
While I disagree with your initial assertion that 4 DIMMs of 8 GB DDR4-3600 is the current standard (Amazon sales suggest 2 x 8GB 3000 or 3200), there always comes a time when a test system needs to be updated. Whether or not the team wishes to do so is down to them.
Would users with high end CPUs and GPUs be gaming at 1080p though? I appreciate the points you're raising, as it's valid for CPU testing/reviewing, but it is down the review team as to what system configuration they want to use.
Also note that 3000 MHz, with manual timings at the FLCK at 1500 MHz, produced higher performance than 3800 MHz, CAS 16, FLCK 1900 MHz. If one just used those two pieces of data alone, it would seem that there's little need to go with faster RAM, as long as one can adjust the timings appropriately.
That's a good point about low resolution, super high refresh rate gaming needing surprisingly powerful components. It does beg the question that should reviews containing performance benchmarks of components be separated or flagged up for specific users/scenarios? For example, the latest CPU review would be initially a general one, using the usual test platform, whereas a similar review could be targeted at e-sport professionals by only looking at those particular game benchmarks, but with additional test paraments (such as memory overclocking). The only problem there is that it's less of a CPU review, and more of a RAM/platform one.
