Whats faster 2 sticks or 4 sticks?

[TepidTech]

Should I go 2 sticks of 32Gb or 4 sticks of 16Gb? Ryzen 9 3900 on a B450 Steel legend, if it matters...

Thanks!

[Fasauceome]

You have made too many threads. You can make a single thread asking for a build suggestion

[TepidTech]

1 minute ago, Fasauceome said:

You have made too many threads. You can make a single thread asking for a build suggestion

Sorry, was just trying to break it down so as not to get into the weeds with unrelated stuff... Will do a "general build" post if needed.

[Valkyrie Lenneth]

4sticks is always better due to more spreading of cores per stick,  its around 10% gain on mem bandwidth ( and with ryzen fps too )

[Fasauceome]

4 minutes ago, Valkyrie Lenneth said:

4sticks is always better due to more spreading of cores per stick,  its around 10% gain on mem bandwidth ( and with ryzen fps too )

You don't gain any memory bandwidth with 4 stick over 2 if it's only a dual channel motherboard, like the Steel Legend mentioned in the original post

[mariushm]

Ignore @Valkyrie Lenneth, he's either trolling you or he doesn't know what he's talking about.

 

4 sticks is not always better.

 

Unless we're talking Threadripper or other very high end Intel processors, majority of processors on the market have a dual channel memory controller, where each memory channel can support a maximum of 2 memory sticks.

So, you get the most performance when you use at least one stick on each memory channel, so two memory sticks. If you install four memory sticks, you won't get better performance, it's still just two memory channels transferring data to the processor cores.

 

In fact, when you have 4 memory sticks installed, there's double the amount of signals going between the memory slots and the cpu socket which means there more 'noise' inside the circuit board and the memory controller inside the cpu has to work harder to receive the data and correct any small errors that may occur.

Because of this, you may find that on cheaper motherboards, you may not be  able to run 4 memory sticks at high frequencies like 3200 Mhz, but the very same memory sticks would run just fine at 3000 Mhz - at lower frequencies each bit of information stays just a tiny bit more on the wires between cpu and socket and the memory controller inside the processor has a bit more time to decode these weak signals and detect correctly if it's a zero or a one bit.

 

So you should always install as many memory sticks as the amount of channels the memory controller inside the cpu has, or double the amount (but double that amount won't give you more performance).

For socket AM4, you have two memory channels. Threadripper has 4 memory channels, so you should install 4 sticks. 

There's some Intel processors which have only 3 memory channels and in the past, AMD had socket AM1 or FM1 if I remember correctly which had only memory channel.

 

 

 

 

[1van]

2 hours ago, TepidTech said:

Should I go 2 sticks of 32Gb or 4 sticks of 16Gb? Ryzen 9 3900 on a B450 Steel legend, if it matters...

Thanks!

Is this the 50th thread with this question for a week? Do you guys even use search?

@Fasauceome@mariushmHave you heard of bank interleaving and rank interleaving? It is not only about channels It has been tested a lot already, there is no need to make stuff up.

[mariushm]

49 minutes ago, 1van said:

Is this the 50th thread with this question for a week? Do you guys even use search?

@Fasauceome@mariushmHave you heard of bank interleaving and rank interleaving? It is not only about channels It has been tested a lot already, there is no need to make stuff up.

Yeah, I've heard.  I put it in the same bin with overclocking memory by a few hundred mhz ... minute performance increases for too much wasted time.

 

Quoting from an old techpowerup article:

 

Quote

DR modules can often be a smidge faster thanks to a capability called “rank interleaving,” wherein the second memory rank can still perform work while the first is being refreshed for use. However, DR modules are often harder for a system to drive to high frequency, which is why most high-performance memory kits use multiple 4GB or 8GB SR memory sticks. The extra frequency achievable by the SR memory modules is often enough to overcome the small performance benefit of DR DIMMs, too.

You can often tell single and dual rank memory apart by looking at the product code, which might say 1Rx4 or 1Rx8 for single rank, or 2Rx4 or 2Rx8 for dual rank. And though you should always verify with spec sheet, it’s a decent shortcut to assume an 8GB DDR4 DIMM is single rank, whereas a 16GB DIMM is almost certainly dual rank.

As we finally come to the data, our results lend credence that—all things being equal—DR memory configurations are a touch faster than SR configs for the purposes of PC gaming. But all things aren’t equal when it comes to overclocking memory, and we’ll explore that in the conclusion.

 

However, this has NOTHING to do with the original question, using 2 memory sticks or 4 memory sticks. This doesn't make a different if you have 2 or 4 ... it applies to each individual memory stick.

Conclusion is the same: you won't get extra performance by installing 4 sticks instead of 2, all other things being the same

[1van]

1 minute ago, mariushm said:

However, this has NOTHING to do with the original question, using 2 memory sticks or 4 memory sticks. This doesn't make a different if you have 2 or 4 ... it applies to each individual memory stick.

Conclusion is the same: you won't get extra performance by installing 4 sticks instead of 2, all other things being the same

Oh really

[Valkyrie Lenneth]

5 hours ago, Fasauceome said:

You don't gain any memory bandwidth with 4 stick over 2 if it's only a dual channel motherboard, like the Steel Legend mentioned in the original post

 

5 hours ago, mariushm said:

Ignore @Valkyrie Lenneth, he's either trolling you or he doesn't know what he's talking about.

 

 

 

 

 

do ur school work ,  if u have more sticks , each core gets assigned to a ram stick,   and i dont know what im talking about?  for all u know i might be designing ram itself, so dont judge before u do the homework

 

u can confirm ur bandwidth gain even on dual channel boards with 2 vs 4 sticks easily, its more bandwidth due to core assign

 

4sticks will always beat 2 , even if it runs in singlechannel mode, simply due to the fact of every cpu these days having 4 or more cores

 

the only downside on 4sticks is that it stresses the memory controller more and reduces ur maximum overclock capabilities

 

ram is also bottlenecking cpus these days, hence why higher mhz gives more fps, or 2 vs 4 sticks ,   hence also why ddr5 will becoming soon

[Fasauceome]

3 hours ago, 1van said:

Fasauceome@mariushmHave you heard of bank interleaving and rank interleaving? It is not only about channels It has been tested a lot already, there is no need to make stuff up.

 

13 minutes ago, Valkyrie Lenneth said:

do ur school work ,  if u have more sticks , each core gets assigned to a ram stick

I haven't been able to find anybody breaking this down in a video format or displaying the impact on performance. Any resources out there that would actually make this more credible?

 

@Valkyrie Lennethwhat does "each core gets a stick" even mean

[mariushm]

1 hour ago, Valkyrie Lenneth said:

 

 

do ur school work ,  if u have more sticks , each core gets assigned to a ram stick,   and i dont know what im talking about?  for all u know i might be designing ram itself, so dont judge before u do the homework

No, ram doesn't get assigned to cores. Cores RARELY get data directly from ram anyway, they get it from L1 cache, then L2 cache, then L3 cache (Ryzen has 24..64 MB of L3 cache).

You have memory controllers with multiple channels, in the case of Ryzen there's 2 channels.

These channels are connect to the Infinity fabric, where the processor cores are also connected. Here's a Ryzen die if you're interested, this one is for 2400g  :

 

 

You can see the Core Complex (CCX0) and you see the four cores with the L3 and L2 cache in the middle, where it's closest to the cpu cores... because the cores always hit the cache first, then as a last resort, go the slower route to the memory controller.

You don't have one particular core reserving one particular memory channel for its exclusive use, and you can't reserve a single memory stick from a channel for a single core. It doesn't make sense.

You can also see in the picture above how both the cpu area and the graphics area (and the multimedia engine, aka hardware encoders and decoders of video)  "hug" the memory controller area,  because all three need to hit the ram a lot.

 

Quote

 

u can confirm ur bandwidth gain even on dual channel boards with 2 vs 4 sticks easily, its more bandwidth due to core assign

No, there's no extra bandwidth. 

Just get Lavasys Aida64 or SiSoft Sandra or whatever benchmark you want, test memory bandwidth with 2 sticks in dual channel, then test with 4 sticks in your computer. You'll see practically no difference in bandwidth.

You'll see more bandwidth when you go from single stick to two sticks, when you basically enable dual channel mode.

single channel = 64 bit x 2 bit per Hz x frequency ...

dual channel = 2x64 bit x 2 bit per Hz x frequency ... 

so of course the memory transfer speed almost doubles... almost because it depends on the size of the data you get from ram how fast you will actually transfer.

 

 

Quote

4sticks will always beat 2 , even if it runs in singlechannel mode, simply due to the fact of every cpu these days having 4 or more cores

It's practically impossible to have  4 sticks in single channel, as most modern configurations will allow a sort of hybrid dual channel mode, when you have memory sticks of different sizes (ex you have a 4 GB and a 8 GB stick, the first 4 GB of each stick will be in dual channel, the last 4 GB of second stick will be single channel, something like that)

 

Quote

 

the only downside on 4sticks is that it stresses the memory controller more and reduces ur maximum overclock capabilities

 

Quote

ram is also bottlenecking cpus these days, hence why higher mhz gives more fps, or 2 vs 4 sticks ,   hence also why ddr5 will becoming soon

The combination of higher frequency and latency (timings) gives more processing speed, which could result in fps increase in some games that are cpu bound, computation heavy.

You can have 3600 Mhz and CL20 or you can have 3200 Mhz and CL16 ... the 3200 Mhz stick may function better in SOME applications you may not care about because they're not games, but computers aren't made just for gamers.

 

Also, NO, you won't get DDR5 soon.  The standard isn't even finalized, it will be done sometime in 2020, and you'll probably have the first processors supporting DDR5 in 2021 or 2022 and they'll require new sockets.

 

Just like DDR4 was compared with DDR3, the initial DDR5 sticks won't be much faster than DDR4, they'll use less power and they'll have higher frequencies, but they'll run at higher latencies, so the performance won't double.

 

 

The only place where you could argue some memory sticks get "assigned to cores" is in processors like Threadripper or Rome, where you have these processors made with multiple dies, and each die has its own memory controller with their own memory channels.

For example, you can see this Epyc CPU made with 4 cpu dies, each with its own DDR4 controller which has 2 memory slots:

 

 

So yeah, in such scenario you could argue that the application must be careful to put the data a particular thread on a particular core needs into the ram stick that's used by that die (the core being in one of the tw CCX in the die).

If your data isn't in those RAM slots, then it's requested from another die through the infinity fabric that connects all dies, and you get higher latency.

This is also why with Epyc and Threadripper you have to install 4 sticks at the minimum, one for each memory channel, so that you won't get a cpu die without memory slots, always being slower as it has to request data from other dies through the infinity fabric.

 

[Valkyrie Lenneth]

25 minutes ago, mariushm said:

No, there's no extra bandwidth. 

Just get Lavasys Aida64 or SiSoft Sandra or whatever benchmark you want, test memory bandwidth with 2

memtest will show it even, and i can tell that u dont know how dual channel works , u can have 2x dual channel on a 4slot dual channel supported board,  aswell as 4x single channel on a 4slot supported board

 

also ddr5 is already out for a while now, just not available to consumers

 

1cas equals around 266mhz average, since u dont know the ratio of cas to mhz

 

and once again, each cores does get assigned to a ram stick

 

, believe what u want but dont confuse ppl who dont know with that bullshit ,   go try it urself on benchmarks ul see about 10% gain from 2 to 4 sticks ( on dual channel supported boards )  if its only singlechannel itl be closer to 5%

 

u also say , its practically impossible ot have 4 sticks in single channel,  u really need to learn about ram,   because EVERY SYSTEM can run all sticks in single channel mode(4sticks or even 8sticks if the mobo has it ),  its dual channel where its required more support

 

 

once again , do actual research on how ram works , rather then cpu dies

[mariushm]

There's two memory channels in a memory controller.

Dual channel happens where the memory controller issues reads and writes on both memory channels at same time.

Each memory channel has a 64 bit data path and there's 2 bits transferred on each pin (rising and falling clock), so when the controller reads and writes from one channel, you transfer 128 bits for every Hz.

In dual channel mode, it transfers 2 x 128 bits on every Hz. 

If you have 4 sticks of memory, the processor can still only request data from ram through those two channels. It can't magically make 4 separate requests, one for each stick, your limitation is the number of channels.

Yeah, the channel may interleave requests between sticks in the same channel so that while one stick prepares the data to send it, the other stick sends data ... but the performance increases by doing this are extremely small.

 

However, the latency... it means there's a bunch of these Hz where the memory controller does nothing but waiting for the memory sticks to get data ready and make it available on the pins. Once data is available, it bursts and goes to the CPU in a continuous stream.

If you have big chunks of data, like let's say when you're encoding a video so you take 1-10 MB of raw video frames and load them in cache and then work in cache with them and then read a bunch more frames, latency matters less, because those few Hz spent waiting for ram is nothing compared to the tens or hundreds of thousands of Hz that are needed to transfer those tens of MB.

If you have an app. that works with very small and random amounts of data, like databases, latency can be a problem.

 

Last message I'm gonna post here, not worth wasting my time.

[Focusrite]

On 11/15/2019 at 4:32 PM, Fasauceome said:

You have made too many threads. You can make a single thread asking for a build suggestion

Thanks a bunch.

[mike0559]

I found the discussion above to be perplexing. Each "episode" of the discussion seemed to reveal that the opposite conclusion is emphatically the correct one! 

 

In researching this myself, I found the following article that tests it empirically:

https://www.techspot.com/article/1971-more-ram-modules-better-for-gaming/

 

The results appear to show that 4 sticks can be faster than 2 in some situations, and I don't see any of their results ever showing 4 sticks to be slower than 2 (so maybe the conclusion is "throw in 4 sticks just in case it helps"). But then they also claim, without empirical testing, that using 4 sticks might lead to worse performance sometimes due to a possible decrease in the clock frequency in some situations. I wish they had tested this rather than speculating theoretically, since apparently this thread and others show that theoretical opinions can be divergent with conviction.

[mariushm]

1 hour ago, mike0559 said:

The results appear to show that 4 sticks can be faster than 2 in some situations, and I don't see any of their results ever showing 4 sticks to be slower than 2 (so maybe the conclusion is "throw in 4 sticks just in case it helps").

It can be slower if the motherboard is cheap and can't handle 4 sticks running at high frequencies, and you have to lower the frequencies to boot or get the ram stable.

For example, you try 4 3600 mhz sticks and find out you can only do 3200 mhz with all four installed. Or 2933/3000 Mhz.

 

Reviewers don't test with a 60$ budget motherboard, they test with whatever they fancy, a mid-high end that also supports overclocking, to make benchmarks and tests simpler and to basically be able to reuse board with different processors and video cards.

[Tankkeepers]

I just wanted to put some fact down gor one you can see fps gain of 4 to 10 percent using 4 vs 2 sticks it is not an increase in bandwidyh or core assinment its duel to there being an extra rank avalible as 4 single rank moduels will run in duel channel duel rank mode when installed vs 2 single rank moduels running in duel channel single rank its has to do with the timing of the ram while rank one is being given data two rank can prisses the request to send data and have it rdy once rank 1 is done which in therory halfs the cls timing of the ram but as said real world it 4 to 10 percent just like duel vs single channel

[ImAyaanKhan]

On 11/15/2019 at 7:31 AM, TepidTech said:

Should I go 2 sticks of 32Gb or 4 sticks of 16Gb? Ryzen 9 3900 on a B450 Steel legend, if it matters...

Thanks!

Get a 2x16gb kit, as getting 4 sticks will see you without dual rank

[Mister Woof]

I went from 2x8 single rank to 4x8 single rank a year ago because I wanted cosmetically to fill all my DIMMs

 

there may have been improvements in benchmarks but they sure as hell weren't perceivable