Fast Ram vs Gaming

[Jeff G]

This will be obvious to veterans of ram, but hopefully it helps some people debating if fast ram is worth the extra cost. Lots of debate on whether or not 3600mhz ram is best for Zen2, so here's my findings.

I decided to test the difference on my Ryzen 3600/980 ti combo. I used 3600mhz/c15 ram.
For testing, I used the built in benchmarks on Division 2, Ashes of the Singularity, Shadow of War, Metro: Exodus, and Farcry 5 because those are the games I currently have with built in benchmarks.

I first set it to 2133mhz/c19 and ran all the built in game benchmarks on low graphics settings, then medium settings, then high settings.
I then set my ram to 3600mhz/c15 and repeated the same tests.
My findings:
On low graphics settings, I gained between 20-30fps just from increasing the ram speed.
On medium graphics settings, I gained between 10-15fps from the ram increase.
On High graphics settings, I gained 0fps across all titles. Zero gain at all.

TL;DR - If you're bottleneck is your gpu, which most peoples probably will be, faster ram saw zero gaming benefit. It was only when I removed the gpu bottleneck that I was able to see gains. If you play on low settings to get max fps, it should help. If you game on high because you want the best quality, you're better off saving your money on the faster ram.

 

[Alaric]

For those of us who expect to get several more years from our cards, potentially having to turn down settings to do so, this is some very useful info to have.

I have to ask, though, would it be the same with an Intel CPU? I ask because the Ryzen 2 CPU is faster with faster RAM (to 3600 MHz) due to the Infinity Fabric I/O architecture.

 

[UltraTaco]

I saw increase in FPS when playing fsx. It's a flight simulator for those who don't know. I saw better minimum fps and average fps when increasing ram speed. Effect was similar to increasing 2 bins on cpu speed. I ran lga1366 platform. Nehalem.

 

[yoadknux]

Thanks for this topic, it's consistent with other results. RAM speed seems to behave similar to CPU bottleneck, when you try to achieve very high FPS (144+) it can become important, but becomes irrelevant if you're squeezing every little bit out of your GPU (for example 4k gaming).

 

[mackerel]

I have to ask, though, would it be the same with an Intel CPU? I ask because the Ryzen 2 CPU is faster with faster RAM (to 3600 MHz) due to the Infinity Fabric I/O architecture.

Short answer is yes. If you need more ram bandwidth, you need more ram bandwidth. Before Zen 2, Ryzen hasn't really been fast enough to feel the ram limitations as quickly as Intel CPUs.

The complication I find with Ryzen is decoupling the influence of IF and actual ram bandwidth. In the past I had tried to explore this by using single channel high speed ram, vs dual channel low speed, to vary the balance between actual bandwidth and IF speed. Results were inconclusive - too small to call. I guess with Zen 2, we can now manually set IF separately from ram speed so this could be explored in more detail, but it isn't high enough of my list of things to do. I still haven't done my own IPC comparison of Zen 2 vs Intel for example.

There was a Linus Tech Tips video that I think has hit the 'tube now (I saw it on early access) where they suggested they had a good result with slow speed, low latency ram with IF speed really turned up. I believe the connection between cores and L3 cache is also run at IF speed, so that could influence performance. Not looked at ram pricing recently, but if 3600 is out of reach, a lower latency 3000 kit might do as well with an IF speed bump.

 

[EarthDog]

This will be obvious to veterans of ram, but hopefully it helps some people debating if fast ram is worth the extra cost. Lots of debate on whether or not 3600mhz ram is best for Zen2, so here's my findings.

I decided to test the difference on my Ryzen 3600/980 ti combo. I used 3600mhz/c15 ram.
For testing, I used the built in benchmarks on Division 2, Ashes of the Singularity, Shadow of War, Metro: Exodus, and Farcry 5 because those are the games I currently have with built in benchmarks.

I first set it to 2133mhz/c19 and ran all the built in game benchmarks on low graphics settings, then medium settings, then high settings.
I then set my ram to 3600mhz/c15 and repeated the same tests.
My findings:
On low graphics settings, I gained between 20-30fps just from increasing the ram speed.
On medium graphics settings, I gained between 10-15fps from the ram increase.
On High graphics settings, I gained 0fps across all titles. Zero gain at all.

TL;DR - If you're bottleneck is your gpu, which most peoples probably will be, faster ram saw zero gaming benefit. It was only when I removed the gpu bottleneck that I was able to see gains. If you play on low settings to get max fps, it should help. If you game on high because you want the best quality, you're better off saving your money on the faster ram.

Interesting testing! Was this at 1080p I would assume? One thing I noticed is many of the titles use a fair amount of CPU or can be CPU limited (like Ashes) in the first place.

You also started the testing at 2133 which, for Ryzen 3 is nearly 1100 MHz below (~33%) its base specification. I'd also be interested in seeing the difference between 3200 (Zen 2's base specification) and 3600 so you are testing at the minimum spec for the platform and the sweet spot. I'll bet we find those values shrink on low/medium...

Anyway, here is some testing from TPU across a couple of resolutions. It's all on high/ultra (where most people run, or strive to run their games anyway) though.
https://www.techpowerup.com/review/amd-zen-2-memory-performance-scaling-benchmark/3.html



RE: Intel, I wonder if it would be the same, actually... I would bet the difference would be LESS, honestly, as their memory isn't tied into any "IF" or anything...

EDIT:

Thanks for this topic, it's consistent with other results.

Can you link other results which tested this way?

 

[mackerel]

RE: Intel, I wonder if it would be the same, actually... I would bet the difference would be LESS, honestly, as their memory isn't tied into any "IF" or anything...

It's been tested to death in the past, not in a position to dug up examples right now. Consumer Intel CPUs, being much higher clocked and not having the funny cache situation were much better than earlier Ryzen at high fps situations. Everything is a bottleneck in some way, and for high fps gaming, the CPU can easily be hindered by the ram. Look at it this way:

Intel: ram - CPU - GPU
AMD: ram - IF - CPU - GPU

Any of those links could be the limit. For high fps scenarios, it still feels like IF is more the limit than ram even on Zen 2, but I don't have evidence to back that up. And with Zen 2, you can run slower ram + faster IF.

I also need to take back a previous statement, I don't know that L3 is tied to IF clock. Will leave that open for further research.

 

[EarthDog]

When you can, dig those up. In the meantime: https://www.techpowerup.com/review/...-memory-performance-benchmark-analysis/9.html

Looking at 1080p, again highest settings (and 720p testing is garbage), we see little difference in these titles.

 

[mackerel]

Very quick google gave https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/ as example, older games, older CPU, but a clear difference as speeds increase there. Maybe more modern games have optimised differently. I find them a good resource for this type of testing, along with their youtube channel Hardware Unboxed.

 

[EarthDog]

Clearly it depends on the title(s) used and how they respond to the increase in bandwidth and reduction in latency. But in general, for Intel, it doesn't seem worth the price premium for say DDR4 4000, compared to simply running 3200 or so (which that article you link recommends - DDR4 3000).

 

[mackerel]

In that era, 3000, maybe 3200 was the value performance sweet spot, so I'd go with that. Base speed was 2133 (Skylake), 2400 after, so that is still a decent increase. Even now, the price of 4000 is way above that, even if it has come down somewhat.

 

[EarthDog]

In a couple of those titles there were some notable increases, others, not so much at all. And in in the TPU tests, barely any in those titles at 1080p.

Unless the few titles that respond well show a notable increase, there isn't a reason to buy over the sweetspot. And in many cases, the base spec will do too.

 

[mackerel]

It is for the individual to decide, as a system, what is "good enough". I'd rather have a 4c4t Intel than 1st gen Ryzen for high fps gaming at the time for example, but today's games, that could be different. Even today, I'd rather pair my faster (3000+) ram with Intel systems than Zen 2 ones as they will make better use of it. At the other end of the spectrum, if cash is less of a barrier, you can throw more at the system for diminishing returns.

 

[EarthDog]

Preaching to the choir, Mack!

 

[mackerel]

Mackerel: stating the obvious since... too long

 

[Jeff G]

Interesting testing! Was this at 1080p I would assume? One thing I noticed is many of the titles use a fair amount of CPU or can be CPU limited (like Ashes) in the first place.

You also started the testing at 2133 which, for Ryzen 3 is nearly 1100 MHz below (~33%) its base specification. I'd also be interested in seeing the difference between 3200 (Zen 2's base specification) and 3600 so you are testing at the minimum spec for the platform and the sweet spot. I'll bet we find those values shrink on low/medium...

This was on a 1080p 144hz monitor. I also have a 2560x1440 100hz monitor and 3440x1440 100hz monitors, I could test on either of those if needed.

I started at 2133mhz/c19/1.2v on the ram because that's what it booted to stock. I then enabled XMP which put it at 3600mhz/c15/1.35v. I could dial in 3200mhz and re-run as well, shouldnt take real long. In all honesty, I'll was just happy that I could hit Enable XMP and have zero issues unlike the previous Zen.

If there's other games with built in benchmarks that people wamt to see, I will see if I can get them too. I dont like running stuff like Cinebench because I'm not gaming in Cinebench. I want to keep it as much about gaming as possible because that's pretty much the only thing I built this computer to do.