[Intel] Noticeable FPS gains going from DDR4 2400 to 3600?

Rev. Night

[H]ard|Gawd
Joined
Mar 30, 2004
Messages
1,442
I currently have 2x8gb G.SKILL ripjaw DDR 2400 (15-15-15-35) for my gaming PC, with the rest being a 6700K @4.7ghz/ Gigabyte Z170 HD3P/ Radeon 5700xt. I saw this TechSpot article where it details out how much ram can help Ryzen, but we already knew that. For my setup, specifically my older mobo, would there be any noticeable FPS gains going from my DDR 2400 to a say CL14 DDR 3200/3600? I aim for 1440@100 in all of the games I play, and I mostly get it but not always. I don't mind on buying used here on the FS forum to make pricing even better.

I know there are reviews that say ram is irrelevant for intel gaming, but those are years old and the graphics cards being used are like 7th gen nvidias.

Incidentally enough, yesterday in my bios, I set the ram to 'Relax OC'. Booted up Jedi Fallen Order, and my fps was reduced by half, it never went about 35-40. I set it to Normal as well as Enhanced Performance, and the framerate went back up to the normal 90s.
 
RAM isn't irrelevant for Intel. Skylake and its descendants to benefit somewhat from increase memory bandwidth. Intel isn't sensitive to timings, but clock speed matters. However, you have a somewhat lower point of diminishing returns. It's generally accepted that it's around DDR4 3200MHz or so. I'd say upgrade if you can do it cheaply. I'm not sure that it would translate a whole lot into FPS, but it might make for a smoother gaming experience. You would potentially have less hitching and things like that in games.
 
I'd try loosening timings to something like 16-16-16-40 or so (maybe 16-18-18-40) and then running it at 3200Mhz. Might work and saves you the upgrade cost.
 
I'd have to do so manually in the bios. I have quite a bit of CPU OC experience, nothing with ram.

Curious, I assume the motherboard QVL (qualified vendor list) should be ignored right? Gigabyte hasn't updated it since 2016 and I doubt my mobo can only accept one page worth of ram.
 
I'd have to do so manually in the bios. I have quite a bit of CPU OC experience, nothing with ram.

Curious, I assume the motherboard QVL (qualified vendor list) should be ignored right? Gigabyte hasn't updated it since 2016 and I doubt my mobo can only accept one page worth of ram.

Yes, I wouldn't put too much stock in it. It would be one thing if it were a constantly updated list, but a lot has changed since 2016. Certainly a lot of new ICCs available with different timings, etc.

Yes, you'd have to do memory timing adjustments in bios. Then check for stability, etc. Memtest86 still works pretty well, but then testing it again inside your OS with whatever your preferred method is should be fine.
 
Sounds good! The looser timings would help for higher clocks, and since its Intel, the looser timings wouldn't necessarily be a negative. As opposed to Ryzen which needs tighter timings. I'll try this tonight, benchmarking Far Cry New Dawn, Borderlands 3, and Mordor at 2133, 2400, and whatever the highest 3x00 I can get. I assume the max DRAM voltage to use is 1.400v?
 
I'd have to do so manually in the bios. I have quite a bit of CPU OC experience, nothing with ram.

Curious, I assume the motherboard QVL (qualified vendor list) should be ignored right? Gigabyte hasn't updated it since 2016 and I doubt my mobo can only accept one page worth of ram.
QVL doesn't matter for Intel. Just about everything works with Intel. Hell with ryzen 3xxx it doesn't matter. Only first gen Ryzen had issues.
 
QVL doesn't matter for Intel. Just about everything works with Intel. Hell with ryzen 3xxx it doesn't matter. Only first gen Ryzen had issues.

I test a lot of motherboards. The above statement isn't true on either side. I've seen Intel motherboards that wouldn't work with specific kits. I've seen it far more on the AMD side. While the Ryzen 3000 series CPU's and X570 motherboards are vastly improved over their predecessors when it comes to memory compatibility, it's still an issue. I've got a ton of memory kits on hand for testing for this very reason. Some modules just do not work correctly on some motherboards.
 
Times like these I am glad I keep a competent htpc downstairs. I forgot that I had a pair of Corsair Vengeance LPX 2x8gb DDR4 3200MHz C16. I can easily put that in and see what the improvements are.
 
Times like these I am glad I keep a competent htpc downstairs. I forgot that I had a pair of Corsair Vengeance LPX 2x8gb DDR4 3200MHz C16. I can easily put that in and see what the improvements are.

Like I said, it's not going to be amazing. However, at DDR4 2400MHz, you are just above JEDEC speeds so I imagine it will be something. That's really low by modern standards.
 
I test a lot of motherboards. The above statement isn't true on either side. I've seen Intel motherboards that wouldn't work with specific kits. I've seen it far more on the AMD side. While the Ryzen 3000 series CPU's and X570 motherboards are vastly improved over their predecessors when it comes to memory compatibility, it's still an issue. I've got a ton of memory kits on hand for testing for this very reason. Some modules just do not work correctly on some motherboards.

We've had this discussion before. Generally, you're reviewing a lot of new boards on 1st revision bioses. Most bios revisions have a line about "improve memory compatibility" in the release notes. Do you ever go back and retest the memory after 1 or 2 bios revisions? I wouldn't think you would just for the sheer amount of time involved. I have had very little issue with memory compatibility on either companies platforms since my B350/Ryzen 1700 build at launch. Zen+, Zen2, Intel 6th through 10th Gen have all given me zero issues with whatever memory I threw at it (a lot of Samsung and Hynix). There have been a few caveats...Zen pretty much had an upper limit of 2933Mhz, Zen+ 3200Mhz, etc., but as long as you were inside the mainstream, it worked fine (more so after a bios update or two).
 
man, moving up from 2400 to 3000 is a huge improvement. Jedi Fallen Order, which is notorious for being a bad pc port, is significantly smoother now at loading assets and maps. Far Cry new dawn, the maximum fps hasn't changed but both minimum and average FPS increased by like 5-10fps. The timings are automatic and are loose as hell (18?21?). I plan on trying to move up to 3200 tonight. Voltage is set to 1.400v.

The next question is heat and cooling. How can I see how hot the ram is getting? What C should I be concerned about? I have 2 120mm intakes, should I spend the $12 for a 3rd or glue on heatsinks if needed?
 
I currently have 2x8gb G.SKILL ripjaw DDR 2400 (15-15-15-35) for my gaming PC, with the rest being a 6700K @4.7ghz/ Gigabyte Z170 HD3P/ Radeon 5700xt. I saw this TechSpot article where it details out how much ram can help Ryzen, but we already knew that. For my setup, specifically my older mobo, would there be any noticeable FPS gains going from my DDR 2400 to a say CL14 DDR 3200/3600? I aim for 1440@100 in all of the games I play, and I mostly get it but not always. I don't mind on buying used here on the FS forum to make pricing even better.

I know there are reviews that say ram is irrelevant for intel gaming, but those are years old and the graphics cards being used are like 7th gen nvidias.

Incidentally enough, yesterday in my bios, I set the ram to 'Relax OC'. Booted up Jedi Fallen Order, and my fps was reduced by half, it never went about 35-40. I set it to Normal as well as Enhanced Performance, and the framerate went back up to the normal 90s.
Thanks for asking this I had seriously old RAM before and new RAM makes a HUGE difference. I don't know about your other questions in regards to heat and cooling, good luck!
 
Thanks! I'll post numbers tomorrow since I have an extra set of 3200 c16 I can test and OC too
 
FWIW, I noticed about 5%~10% in benchmarks (FPS) going from DDR4-2400 speeds to DDR4-3000 speeds with extremely tight timings at 2K resolution on my 5960x (I picked up a Quad Samsung B-Die when I did it). At 4K, the difference was much less noticeable, less than 2~3% (as that is a GPU limited resolution), but overall my bandwidth increased and I got my latency way down (for Haswell-e anyway).
 
Latency values don't really matter with Intel systems. Get whatever you want. I'd just aim for DDR4 3200MHz or better.
It depends on the program.
I have found 3000c12 to perform just behind 4000c17 (tweaked subs for both) and much better than 3200c14 XMP with a 6700K.
For most the games I play I find that the final latency is the most important but if you can keep the latency low and squeeze in a little more bandwidth from higher frequency all the better.

The gains from overclocking past 3200c14 XMP were similar to the gains from overclocking the 6700K to 4.7GHz with some games prefering the RAM speed and others preferring the CPU speed.

I guess frequency/bandwidth could become more important with CPU that have more cores fighting over the same amount of bandwidth if the program can make use of those extra cores.

Arma%20III%20cpu%20vs%20ram.png

4000c17 - 55223MB/s Read - 37.3ns
6700k%204.7%204000c17.jpg
3000c12 - 45377MB/s Read - 37.3ns
6700k%204.7%203000c12.jpg
3200c14 XMP - 45620MB/s Read - 40ns
6700k%204.7%20xmp-3200c14.jpg
 
Last edited:
So I have some updates! I recently upgraded my ram from 2400/3000 to new sticks of 3600. Here is the FPS breakdown (min, avg, max) for Far Cry 5/New Dawn. Why this game? Because its very reactive to cpu, gpu, and ram adjustments. Unlike other games were its basically 99% gpu, 1% everything else. Also has internal benchmark.

Gaming PC:
I7-6700K, 4.7ghz delid
Gigabyte G170-HD3P, F22g bios
Radeon 5700xt (undervolt, underclocked to 1000mv and 1900hz)
Windows 10
2x8GB

Far Cry New Dawn, 1440 resolution, Normal settings, TAA

DDR4-2400 stock (G.Skill CL15-15-15-35): 67, 87, 124
DDR4-3000 above OC (G.Skill CL15-15-15-35): 75, 95, 127
DDR4-3600 stock (Corsair LPX CL18-19-19-19): 84, 105, 142

From 2400 to 3600, which is exactly what this thread is about, shows a 20 fps increase in minimum, average, and max fps. Can't ask for more than that considering that I bought them on this forum for $60.
 
I think one of the lacking points that is never particularly presented is the front end to back end of the processor. The coupling or de-coupling of the parts. Intel's "greatest" was this: Prediction, prefetching, decode/microcode, cache bandwidth, LD and S units. Intel's prediction hit rate before mitigation have been known to be "uncanny." The hardware and prediction allowed the flow of data in to the processor to be really fast. After the mitigation, that slowed a good bit (no pun intended). The issue with Infinity fabric is that much of this was a part of the memory "feeding" the processor. Which is also why you see rank always beats with Ryzen-more memory utilization. People think "speed, speed" when it comes to RAM, but memory utilization knowledge is key. You want pages hit, not more cycles refreshing

Anyways, what is really impressive is the Zen 3. So much more fatness to the chip. It is close to some of the newer architectures of Intel's from what I gather. Matter of fact I cannot wait to see this chip's layout and bits per cycle. But also, that fatness will help tremendously with interconnectivity of cores and the Infinity Fabric. It is less pressure on memory and more coherency with the ability to really feed the processor. I would truly love to test this beast or hear David Kanter talk about this processor.

Forgot to mention, so after mitigation there is more contention with Intel's arch. But, IMO, is not worth the cost to try to increase an older architecture. However, you do notice the importance that system memory has overall to the processor but it isn't usually placed in its proper context.
 
Last edited by a moderator:
Back
Top