Ram for a 5600x?

[Naff8000]

Reading through these threads I feel completely lost! I'm obviously an uninitiated noob when it comes to ram.

I've installed my 5600x and purchased 2 kits of Ballistix 3600 C16 BL2K8G36C16U4W for 4x8Gb which I assume is all good? £140 can't really go wrong or should I be getting a different kit?

Cheers

p.s. I'm not really looking to tweak much as the HU video showed (to me anyway ) that 4 ranks of decent speed ram were similar to tuned ram of the same caliber?

 

[Cyber-Mav]

2 sticks be more likely to hit higher clocks than 4 sticks.

 

[Mache]

Ive just watched that video twice! and im still confused between single and dual rank ram.

Im in same boat, im wanting the white RGB Corsair Vengence pro Ram (2x16gb cl16 3200mhz) for £160 but by the video i think i would be wasting money for the performance gains so may aswell get 16gb for same performance.

Or... go 3600mhz (cl18?) Man it is so confusing!

 

[Boxfreshchris]

Ive just watched that video twice! and im still confused between single and dual rank ram.

Im in same boat, im wanting the white RGB Corsair Vengence pro Ram (2x16gb cl16 3200mhz) for £160 but by the video i think i would be wasting money for the performance gains so may aswell get 16gb for same performance.

Or... go 3600mhz (cl18?) Man it is so confusing!

A 16GB kit (2x8GB) will likely be single rank memory. Having said that I recently bought 2x16GB Corsair Vengeance Pro RGB 3600Mhz CL18, and that turned out to be single rank too, however being single rank means it should overclock better, so I should be able to reduce the timings down from the CAS 18 they come as stock.

Then there's the added confusion of your motherboard memory controller, whether its daisychain or t-topology. A daisychain board prefers 2 dimm slots being populated, whereas a t-topology board prefers all 4 dimms. However, a daisychain board will overclock better (in general).

As you can see, it's a bit of a minefield.

Personally, I have an Asus Strix X570, which is daisychain, so I've stuck with my 2x16GB Corsair RGB 3600 CL18 kit, and I hope the fact that it's single rank means I can drop it down to CL16, and it cost £145. Looking around, 32GB of 3600 CL16 is expensive. Add to the fact that a lot of ram is out of stock.

 

[ZeroG]

It is indeed a minefield. I've been looking at ram for an upgrade lately based on an R5 5600X. I was aiming for 2x8GB of 3600mhz ram @ CL16. I was getting a bit confused because there's dual/quad channel and single/dual rank. People expected to get dual rank but getting single and such.

What with stock being hard to come by and prices fluctuating quite a lot I may just wait until early next year to upgrade.

 

[Mache]

All my bits are on order except ram (same as you, 5600x). I might just wait until i get my 5600x and go from there, more reviews should be out by then too.

I suppose theres no hurt in emailing the manufacturer and asking which ones which

 

[sabinx]

I purchased some ram from crucial 2x16gb 3600 C16 for £160 for my 5600x on a B550-I Strix. Turns out it is single rank indicated by the code on the actual stick, if the last characters says FB1 its single rank, if it says FE1 its dual rank, so now way for me to have known when ordering. Looking at some videos, if you're just gaming 1t 1440p or 4k it hardly makes much difference?

 

[MrPils]

I purchased some ram from crucial 2x16gb 3600 C16 for £160 for my 5600x on a B550-I Strix. Turns out it is single rank indicated by the code on the actual stick, if the last characters says FB1 its single rank, if it says FE1 its dual rank, so now way for me to have known when ordering. Looking at some videos, if you're just gaming 1t 1440p or 4k it hardly makes much difference?

Its the number before the letters after the product code that matters. 8 for 8 chips (single rank), 16 for 16 chips (dual rank). Either way your motherboard matters more than the memory ranks and nobody has all of the motherboards to be able to say which are better with 2 sticks and which are better with 4 sticks. Some boards are terrible with 4 sticks, some boards are terrible with dual rank sticks, some boards are terrible with everything. On top of that you have silicon lottery with your memory sticks themselves too.

Moral of the story, don't get hung up on memory ranks. Two single rank sticks is always the safest option, additionally if you want to tune them manually you will always get better timings using single ranks unless silicon lottery deals you a bum hand when picking within the same model of memory.

 

[Mache]

So what if we didnt want to tune atall and just load the xmp profile and away we go type of thing?

Just get what we can afford?

Ive never been so hyped up on memory til now. Ive always just gone for the speed i could afford. As ive found out in the past lower latency was always more expensive.

 

[Boxfreshchris]

If you just want to turn on XMP/DOCP and be done with it, ideally, you want high speed low latency 2 dimms of dual channel on a daisychain motherboard.

Now you just have to try and find all those things, A) in stock, B) not silly money.

 

[MrPils]

So what if we didnt want to tune atall and just load the xmp profile and away we go type of thing?

Just get what we can afford?

Ive never been so hyped up on memory til now. Ive always just gone for the speed i could afford. As ive found out in the past lower latency was always more expensive.

All XMP/Auto combinations are terrible. If you don't manually tune you give away at least 20% of your memory performance when choosing any kit and frequently way, way more.

For example my G.Skill F4-4000C15D-16GVK kit is considered to be high end and right up near the top of the performance charts (its a B-Die 7.5ns bin). If I set XMP timings @ 3800mhz on my single ccx 5600x it takes around 165 seconds to complete the Ryzen calc memtest easy preset. By simply setting tRRD_S, tRRD_L and TFAW to 4/4/16 from the defaults of 7/10/44 the time goes down to just over 130 seconds. Setting tWTR_S and tWTR_L to 4/4 from 5/14 reduces that time further to around 125 seconds. Turning off gear Down mode and using CMD rate 1T dips the time to just below 120 seconds. These are timings not defined by XMP, they are set by your motherboard adding up to over a 33% loss in memory subsystem performance. It doesn't matter what kit you buy, the XMP timings mean nothing at all if you don't manually tune. If you aren't going to manually tune (and that is fine!!) buy the absolute cheapest you can find at 3600mhz, plug it in and away you go. Dual rank, single rank, who cares? Expensive memory is just a pig with lipstick if you don't know or want to know what to do with it, so save your money and don't buy it. Dual Ranks per channel (however you get there - 4 sticks or 2 sticks) will be faster than single rank per channel if you plug and play, but you'll still be competing in the special olympics. There's some brilliant examples of this in the 8 pack thread where people are running 14-14-14 xmp timings and have just increased mhz to 3800 and declared done, but the benchmark result is up to 50% slower than the same timings when tuned.

To give more perspective, fully tuned out the above mentioned kit does just under 100 seconds on the memtest easy preset on a single ccx cpu when fully tuned. This translates to up to 500 points in 3dmark and ignoring edge case games at most 10fps compared to XMP set and forget. If you put crappy 3600mhz memory in there it wouldnt be that much worse, but the money you saved over the £200 G.Skill kit would upgrade a 3070 to a 3080 and get you lots more fps. Spend wisely dudes.

 

[Mache]

All XMP/Auto combinations are terrible. If you don't manually tune you give away at least 20% of your memory performance when choosing any kit and frequently way, way more.

For example my G.Skill F4-4000C15D-16GVK kit is considered to be high end and right up near the top of the performance charts (its a B-Die 7.5ns bin). If I set XMP timings @ 3800mhz on my single ccx 5600x it takes around 165 seconds to complete the Ryzen calc memtest easy preset. By simply setting tRRD_S, tRRD_L and TFAW to 4/4/16 from the defaults of 7/10/44 the time goes down to just over 130 seconds. Setting tWTR_S and tWTR_L to 4/4 from 5/14 reduces that time further to around 125 seconds. Turning off gear Down mode and using CMD rate 1T dips the time to just below 120 seconds. These are timings not defined by XMP, they are set by your motherboard adding up to over a 33% loss in memory subsystem performance. It doesn't matter what kit you buy, the XMP timings mean nothing at all if you don't manually tune. If you aren't going to manually tune (and that is fine!!) buy the absolute cheapest you can find at 3600mhz, plug it in and away you go. Dual rank, single rank, who cares? Expensive memory is just a pig with lipstick if you don't know or want to know what to do with it, so save your money and don't buy it. Dual Ranks per channel (however you get there - 4 sticks or 2 sticks) will be faster than single rank per channel if you plug and play, but you'll still be competing in the special olympics. There's some brilliant examples of this in the 8 pack thread where people are running 14-14-14 xmp timings and have just increased mhz to 3800 and declared done, but the benchmark result is up to 50% slower than the same timings when tuned.

To give more perspective, fully tuned out the above mentioned kit does just under 100 seconds on the memtest easy preset on a single ccx cpu when fully tuned. This translates to up to 500 points in 3dmark and ignoring edge case games at most 10fps compared to XMP set and forget. If you put crappy 3600mhz memory in there it wouldnt be that much worse, but the money you saved over the £200 G.Skill kit would upgrade a 3070 to a 3080 and get you lots more fps. Spend wisely dudes.

Thats what i needed. I think thats what we all need sometimes, a bit of grounding/back to reality.

Maybe one day i will dive into the world of ram oc. I just want to upgrade with the right parts (like we all do) and get the abosultely max performance out of my budget but not overspend on the unnecessaries!

Maybe i will go ahead and buy the 16gb of RGB Corsair Pro... got a white/gold build going on think it should look quite nice when done. Just need the 5600x to turn up!!

 

[Naff8000]

Very interesting @MrPils , especially on the settings you found most useful. How would the reduction from 165 seconds to 120 seconds translate to gaming? And is the general idea (for gaming) to get memtest as low as possible?

I ask because although you say 20-33%, the HU video above seems to give marginal game benefits.

 

[sabinx]

I think it depends on resolution, if your not GPU bound and playing at 1440p or 4K I can't imagine it's a significant difference?

 

[RBImGuy]

I think it depends on resolution, if your not GPU bound and playing at 1440p or 4K I can't imagine it's a significant difference?

rarely a difference worth mentioning.
2x8gb 3600mhz cl16 amd recommended, call it a day.

 

[MrPils]

Very interesting @MrPils , especially on the settings you found most useful. How would the reduction from 165 seconds to 120 seconds translate to gaming? And is the general idea (for gaming) to get memtest as low as possible?

I ask because although you say 20-33%, the HU video above seems to give marginal game benefits.

I think it depends on resolution, if your not GPU bound and playing at 1440p or 4K I can't imagine it's a significant difference?

It depends very much on the games tested, the resolution you're using, CPU, GPU - the list is practically endless. The less GPU limited you are the more of an impact it can have, but how the game engine reacts to that situation is just as important. E-Sports style games and high fps racing games are mostly the ones which see the largest improvements but you do find other more random games (like Hitman 2 in the Hardware Unboxed video) which react well to memory performance tuning. You should be buying super fast memory for learning or improving on your memory overclocking skills though, not for gaming. The price premium gives you such a big upgrade elsewhere in your system that its not even worth considering unless the rest of your pc is already "endgame". In terms of a percentage fps increase in all games from fully maxed out memory the number is usually very low single digits - the games where you see the biggest gains are usually already at very high fps to begin with (so while the fps increase might be high, the gain overall is low).

Buying pc parts is always a balancing act, but as long as you pick a CPU that is "good enough" you want to make sure you have a suitable size SSD and a good power supply next. Budget for some cheap 3600mhz memory, a solid mid range motherboard and your case and cooling. Spend as much as you need to of what's left over to afford the highest model range of GPU you can - RX580/3060ti/3070/3080 - doesn't really matter what just get the fastest core gpu model you can afford. Anything left over you can then sprinkle on things like maybe moving up a core count on your CPU for a bit more longevity, perhaps some improved case or CPU cooling, a step up to a GPU with a better cooler. Only when you've got the rest of your setup optimal and you have a bit of budget leftover should you consider looking for faster memory if you aren't going to be playing around with manually optimising timings. If you did want to learn its definitely fun but also infuriating at times and you can be headbutting against a wall between "breakthroughs" when learning, however its very rewarding to get that big a gain from your memory in benchmarks. Its for the achievement of doing it though more than the empirical gains as once you're spending money on top end memory the rest of your system is usually fast enough that you really don't notice in games.

When you have your system built sure feel free to tune the memory you have bought, the timings I outlined above are the ones you want to look at first because you'll get the largest gains there. Once you've tuned those then you should look at your Cas Latency and TRCD timings. Have a bios profile saved and if you're nervous about messing up your OS boot off a USB stick into memtest86 for a couple of minutes before loading windows (free @ memtest86.com) to make sure things are looking ok. Be warned though, that's when the bug starts to bite and then you start looking at expensive memory kits again!

 

[ollies]

@MrPils Really useful and informative posts / advice, thank you. Other than your own posts on here (particularly in the 8 pack thread ) is there any particular material you'd point to in order to learn more about it? I remember dabbling in memory timings when I first built my current system 8 years ago and failed to get a stabble system so gave up, but have a new build in the works and would like to revisit it.

 

[Naff8000]

Yes an excellent post @MrPils. Thanks a lot, that's all the info I need and I'll be giving this ram tuning a miss for now!

 

[Mache]

Anyone going for 16gb or 32gb?

Not sure what to get. Im running 16gb with my 6700k and does ok vut was wondering what everyone else has chosen

 

[Kaotc]

im going with 2x16gb allows me to add the same again in the future if required