Post your FPS & CPU's scores - instructions how to measure included - 5950x + 12900k + now 5800x3D!!

[lmike]

Hi. My spec is
CPU: 13900K - 100mv
GPU: RTX 4090
RAM: 32 GB DDR5-6200 CL40
OS - Windows 11
All DLCs enabled, v 5.10

This is my results:

Resolution: Windowed 1280x720
Settings: Low preset, no MSAA or upscaling

Dense Empire Empty: 126 - 130 (pressed TAB to escape outside another ship)
Dense Empire: 30-33. After 5 min 22-25
Young Gun Empty: 156

E-cores 0% utilization. They are useless. CPU power consumption is about 60-80W

[LandogarX4]

APPROXIMATE performance chart

[Berni]

13900K@5,8GHz
RTX 4090@70%PL
64GB DDR5 5600CL40
3440x1440 with AA enabled

All DLCs

young gun: 157
dense empire: 35-40
dense empire empty: 135-141

My results are more or less the same like my old 12900K (4,9GHz) + RTX 3090.

I think the X4-Engine has reached it's limits because it can't profit any more from much higher (single core) clock speeds.

[Buzz2005]

it should on long games and that dense empire, im jealous of those fps of yours and should have got intel cpu when I did upgrade, seems overall way better for x4, and some nice intel cpus are even cheaper then amd

[Eyeklops]

64GB DDR5 5600CL40

For such a high spec PC (13900k/4090) this ram is mid-to-low end from a performance standpoint. 5600CL40 has 14.29ns of latency which is pretty high and 5600MT is also mid-to-low end for DDR5 bandwidth. With X4 being very memory transaction intensive I suspect this is your bottleneck. I can't think of any games needing 64GB so I'd suggest dropping total system memory down to 32GB and pushing the performance level to at least 6000CL30 (10.00ns).

If your motherboard and CPU can handle it 6600CL32 (9.70ns) is even better but would be a much smaller gain over 6000CL30. If you're trying to be a total madlad, $$ is no object, and your CPU memory controller is a golden sample, 7800CL36 sports a truly impressive 9.23ns latency. Be prepared to pay $400 for 32GB tho.

If you really need 64GB that can be had in 6000CL30 but it's not cheap ($450).

EDIT: If you do go with something very high end like 7800CL36 make sure you have good airflow over the ram as it runs at 1.45v and can get quite toasty. If you think heat will be a problem consider the 6600CL32 (1.40v, a little cooler) or the 6000CL30 (1.35v, a lot cooler).

EDIT2: Thinking more about it I'd only recommend 7800CL36 if you've have experience tweaking memory timings. That speed is so high end there is a much higher probability of CPU memory controller or motherboard issues preventing a "turn-key" fully stable system.

[Maebius]

I think the X4-Engine has reached it's limits because it can't profit any more from much higher (single core) clock speeds.

Indeed, that's quite obvious judging from 13900K+RTX4090 results.

The engine needs some serious redesign if they plan on moving forward.

The rumored AMD 7X00 with 3D V-Cache might edge the max performance a bit, but doubtful. I'm holding out on upgrading for one of those.

[Berni]

64GB DDR5 5600CL40

For such a high spec PC (13900k/4090) this ram is mid-to-low end from a performance standpoint. 5600CL40 has 14.29ns of latency which is pretty high and 5600MT is also mid-to-low end for DDR5 bandwidth.

I wanted 2x 32GB and not 4x 16GB!
I looked at a lot of comparison benchmarks between 5600 and 6400 ram. I don't care if I have 150 fps with 5600 or 155 fps with 6400 ram.

Believe it or not, but 64GB DOES make a difference for my other favorite game Star Citizen.

EDIT:
And besides, my hardcore overclocking days are over!
...now I prefer an efficient system paired with undervolting.

Making the ram glow with high voltage is no longer my thing

[Eyeklops]

I wanted 2x 32GB and not 4x 16GB!
I looked at a lot of comparison benchmarks between 5600 and 6400 ram. I don't care if I have 150 fps with 5600 or 155 fps with 6400 ram.

Believe it or not, but 64GB DOES make a difference for my other favorite game Star Citizen.

EDIT:
And besides, my hardcore overclocking days are over!
...now I prefer an efficient system paired with undervolting.

Making the ram glow with high voltage is no longer my thing

Understood (& certainly agree on the undervolting/efficiency standpoint). I should have been more straightforward in my prior comment. I question the assessment "the X4-Engine has reached it's limits because it can't profit any more from much higher (single core) clock speeds" due to RAM performance holding back the system (specifically for X4).

[Berni]

Yes, you're right that faster RAM yields better results with tighter timings.

I didn't go into the RAM now, because I was simply of the opinion that a nextgen CPU with 900MHz higher clock should more than compensate for the RAM difference.

Old system: 4.9GHz with 6000CL36

New system: 5.8GHz with 5600CL40

I really expected higher numbers, so i was a bit surprised they weren't.

[Imperial Good]

The worse memory latency likely counteracted the higher performance CPU core. The CPU would either not run near maximum boost or it would have a high pipeline stall percentage (lower effective frequency) if running near maximum boost.

[spookywatcher]

I slapped in a 5800x3d. It definitely made a difference. Performance was better than the 5950x. I didn't do any real testing to see by how much I just did a quick peek to see what it looked like. But the main takeaway for x4 is four sticks (sheesh that's a lot of 4's) of ram at any speed above 2666 mhz was better than 2 sticks at any speed that I could get to which was 4000mhz. This is on AM4 I have no idea about AM5 or alderlake.

And I agree with Berni...SC is my new favorite game. Even with all of it's bugs. And the 5800x3d w/ 4 sticks (32 gb is a MINIMUM) performs better than 2 sticks again just like x4.

[Berni]

The CPU would either not run near maximum boost or it would have a high pipeline stall percentage (lower effective frequency) if running near maximum boost.

I'm monitoring my clock speeds, my 13900K is running at 5,8GHz in X4

A Cinebench R23 Singlecore score of 2270 confirms that too.

[spookywatcher]

The CPU would either not run near maximum boost or it would have a high pipeline stall percentage (lower effective frequency) if running near maximum boost.

I'm monitoring my clock speeds, my 13900K is running at 5,8GHz in X4

A Cinebench R23 Singlecore score of 2270 confirms that too.

Are you certain that your "effective clock" is 5.8ghz? You can find that in hwinfo. You'll see clocks and effective clocks. Effective clocks account for clock stretching.

[Berni]

The 13900k is on average 10-15% faster in games like the 12900K!
But in no case deliver 400MHz faster ram clock about 15% more FPS!

[Imperial Good]

But in no case deliver 400MHz faster ram clock about 15% more FPS!

6000CL36 -> latency of 12 ns
5600CL40 -> latency of 14 ns

5600CL40 has a latency 14/12 or ~16.7% longer than 6000CL36.

That latency is the minimum possible length of time the CPU thread stalls during a cache miss. It is stalled even longer in practice due to other latencies in the CPU.

That 2 ns difference wastes 12 additional cycles at 5.8 GHz. If this happens frequently, which is does due to the cache bound nature of X4 at times, that is a lot of wasted cycles that the CPU core cannot execute instructions.

For reference DDR4 3200 with latency of 16 cycles, the sort that was used in some of the older benchmarks in this thread, has a latency of 10ns. The only reason DDR5 is even remotely competitive in this case is due to other feature improvements it offers over DDR4 that help compensate for the currently worse latency.

[Berni]

But in no case deliver 400MHz faster ram clock about 15% more FPS!

6000CL36 -> latency of 12 ns
5600CL40 -> latency of 14 ns

5600CL40 has a latency 14/12 or ~16.7% longer than 6000CL36.

So you really insist that a 6000CL36 Ram delivers 16.7% more FPS than a 5600CL40 Ram??


Here are a few examples 5600cl36 vs. 6000cl36

Cyberpunk 2077: 5600: 184fps - 6000: 188fps - ~ 2,1% slower

Horizon Zero Dawn: 5600: 216fps - 6000: 219fps - ~ 1,4% slower

God of War: 5600: 244fps - 6000: 246fps - ~ 0,9% slower

[Eyeklops]

But in no case deliver 400MHz faster ram clock about 15% more FPS!

6000CL36 -> latency of 12 ns
5600CL40 -> latency of 14 ns

5600CL40 has a latency 14/12 or ~16.7% longer than 6000CL36.

So you really insist that a 6000CL36 Ram delivers 16.7% more FPS than a 5600CL40 Ram??


Here are a few examples 5600cl36 vs. 6000cl36

Cyberpunk 2077: 5600: 184fps - 6000: 188fps - ~ 2,1% slower

Horizon Zero Dawn: 5600: 216fps - 6000: 219fps - ~ 1,4% slower

God of War: 5600: 244fps - 6000: 246fps - ~ 0,9% slower

Imperial Good wasn't implying there would be a 16.7% direct frames-per-second performance uplift, but simply that 16.7% is the latency difference (which can be some fraction of the performance delta).

The way any particular game utilizes a specific systems hardware to produce a bottleneck is wildly variable. My personal opinion (and possibly Imperial Good agrees) is that for your specific hardware, specifically for X4, the reason you didn't see any uplift in performance is because you're dancing a CPU and memory latency/bandwidth bottleneck and the notable decrease in memory performance offset the gains in raw CPU processing power.

Also, CP 2077, HZD, & GoW are not the best games to evaluate hardware performance deltas with regards to X4. Factorio is a much better fit (see THIS reddit thread).

[Imperial Good]

So you really insist that a 6000CL36 Ram delivers 16.7% more FPS than a 5600CL40 Ram??

I said that the access takes 16.7% longer, not that it will produce 16.7% more frames per second. If the CPU has a lot of cache misses, then such performance difference could occur.

Most games, that do not try to simulate an entire universe, are designed to mostly fit inside the cache of CPUs at the time. This gives optimum performance but does limit their complexity. In such cases the performance difference between memory of different speeds, latency and even channel count will be small (6% or less is usual). However games or applications that have an active state set much larger than the CPU cache so have a high cache miss rate will become highly reliant on both memory access latency and bandwidth (as both are factors in total delay) for performance. In such situations performance differences of 15% are common.

This is why the 5800X3D can out perform the i9 13900k in some applications. Despite the CPU cores being a lot slower and it having fewer of them, the huge amount of extra cache can allow it to have significantly fewer cache misses in appropriate games and applications allowing it to stall less waiting for memory reads/writes and so averaging a faster execution rate.

With memory there are also other timings at play than CAS latency. As such the logic used to explain the issue above is quite simplified and there are likely other aspects which help contribute to slowdown. For example, memory writes are significantly slower for a processor than memory reads, meaning that an application that changes a lot of state will perform worse than an application that just reads a lot of state.

Here are a few examples 5600cl36 vs. 6000cl36

All of them are slower by some amount, which supports that memory latency, and memory timings in general, play a part in performance. They are likely less impacted than X4 due to having a smaller frequently used state set. For example God of War, the least affected title, is structured as a scripted action adventure game for Sony consoles and so has as good as no simulation outside the immediate player area and was highly optimised for consoles with very weak CPUs. Cyberpunk 2077 was the most affected because being an open world game set in a dense city it likely has the most complex, and so largest, state to track of those 3 titles. Cyberpunk 2077 is less affected than X4 likely because much of the background simulation is being faked, Elder Scrolls Oblivion/Skyrim style, rather than in X4 where the state of every single entity is tracked at all times.

Factorio mega factories should show similar scaling to X4 for the same reason. The only limiting factor to how big, and as a result how complex, a person makes their factory is how bearable the frame rate is. As such mega factories in factorio end up having tens of thousands of entities processing thousands of items per second, each being individually tracked to some extent with no trickery occurring to skip simulating parts of it.

[nemo1887]

CPU : 13900k 5.6 mhz
MEM : 32 Go 6000 mhz cl 40 DDR5
GPU: GTX 960 soon RTX 2080 Ti
Dense Emp : 24 fps

[Repli]

Hey guys, for those of you that are using laptops:

I7-12700H
Geforce 3070 TI (8GB, 150W TGP)
16GB DDR5 RAM (4800Mhz)

Empty: ~130 fps
with empire: ~60 fps
dense empire: ~20 fps