• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

RDIMM vs UDIMM

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
J

jawadesign

Registered
‍
Joined
Oct 4, 2013
Location
Chicagoland (lol)
Ok, I searched this forum for my question and found pretty much nothing...
I feel pretty confident asking you guys this question with the hope there is an answer. Which memory scenario below is faster?

1. 12800 - DDR3-1600MHz - ECC - RDIMM

2. 10600 - DDR3-1333MHz - ECC - UDIMM

To my knowledge, number 1. is faster, RDIMM is also automatically ECC.
But what I've read is RDIMM is slower than UDIMM, the nature of the reading/writing process.


Is the scenario above a wash...?
The price difference is HUGE between the two. I need ECC, well I should say I prefer ECC while running calculations. Comparing UDIMM ECC vs UDIMM non-ECC, that cost difference is not very large when the bandwidth/speed numbers are similar.

TIA!
 
Last edited:
R = Registered, U = Unregistered

Sometimes "Buffered" is used instead of "Registered", but this is not the same as "Fully Buffered".

http://en.wikipedia.org/wiki/Registered_memory
  • Registered is NOT automatically ECC.
  • Registered is theoretically faster than otherwise identical unregistered modules in a multi-channel setup.
 
petteyg359,

Thanks for trying to help... after posting I realized that RDIMMs are NOT all ECC.

I'm still trying to figure out which configuration above is faster, based on my intended setup, #1 should be fastest.

My mobo supports, UDIMM, RDIMM, ECC, non-ECC, DDR3, 4-channels, eight 240pin DIMM sockets. What's interesting is my mobo excepts a faster RDIMM ECC than UDIMM ECC, but the RDIMM ECC cost more than double! I'm going for 32GB of RAM total and will not have to upgrade for a very long time, buying RDIMM in 4-packs is definitely cheaper, thus the reasoning behind 32GB. Also, this will allow 1 DIMM per channel, I will fill all the blue or black sockets, every other one... correct me if I'm wrong, but the way I'm understanding the Wiki link, this is the fastest configuration.


...
 
Last edited:
Yes. RDIMM alleviates some burden from the memory controller, which is why faster speeds are possible with it. Obviously, it is priced accordingly :)
 
^Thank you :)

If I were to get number #1 vs #2, do you have any idea how much faster it would be, would it be noticeable? I'm trying to justify the cost of the RDIMM and make sure it's not overkill. At some point, I will run simulations in SW.
I really don't know if I'll ever need more than 32GB of RAM, SolidWorks currently recommends a max of 16GB. My system allows up to 64GB of UDIMM and 128GB of RDIMM. I know this, I definitely want the extra security ECC in whatever I choose, it's just peace of mind...

I'm also wondering, if I ever decide to try and OC my Xenon E5-1650, will the 1600MHz-ECC-RDIMM-1600will play nice at 4.5GHz???


I looked at this link for a better explanation of why RDIMM is faster than UDIMM.
search/google: (RDIMMs maximize server performance, reliability, and scalability) if link does not take you there...
http://www.eetimes.com/document.asp?doc_id=1279507


...

This is the brand and model of RAM I'm interested in at the moment...
http://www.kingston.com/dataSheets/KVR16R11D4K4_32I.pdf

Lenovo's (brand) equivalent is $379 per DIMM! yea...
KINGSTON'S is less, much less!
 
Last edited by a moderator:
I had always read that UDIMMS were faster the RDIMMS -- because the buffering introduces additional latency.
 
magellan said:
I had always read that UDIMMS were faster the RDIMMS -- because the buffering introduces additional latency.
Click to expand...

Wikipedia said:
Nominally, there is a performance penalty for using registered memory. Each read or write is buffered for one cycle between the memory bus and the DRAM, so the registered RAM can be thought of as running one clock cycle behind the equivalent unregistered DRAM. With SDRAM, this only applies to the first cycle of a burst.

However, this performance penalty is not universal. There are other factors involved in memory access speed. For example, the Intel "Westmere" 5600 series of processors access memory using interleaving, wherein memory access is distributed across 3 channels. If 2 memory DIMMs are used per channel, this "...results in a reduction of maximum memory bandwidth for 2DPC (DIMMs per channel) configurations with UDIMM by some 5% in comparison to RDIMM." [1] (p. 14). This is because "...when you go to 2 DIMMs per memory channel, due to the high electrical loading on the address and control lines, the memory controller uses a “2T” or “2N” timing for UDIMMs. Consequently every command that normally takes a single clock cycle is stretched to two clock cycles to allow for settling time. Therefore, for two or more DIMMs per channel, RDIMMs will have lower latency and better bandwidth than UDIMMs."[2]
Click to expand...
 
But you can also interleave UDIMM's as well, as was done w/FPM DRAM a decade ago. I don't think any consumer motherboards even offer 3 DIMM's per channel. BTW, on the Westmere systems I've worked with (IBM x3550-M3's) the maximum operating frequency for the RDIMM's is 800 Mhz when installing 3 RDIMM's per channel. I also think RDIMM's are much less tolerant of overclocking than UDIMM's.
 
My (possibly incorrect) understanding is that you avoid unregistered latency issues by keeping to a single-channel setup, but thereby reduce maximum bandwidth; while the registered modules might have greater latency in single-channel, they have better latency in dual-channel, and you keep the dual-channel bandwidth advantage.
 
Based on my latest research, you're both right... this data appears to be a few years old, but not pre DDR3. So take it with a grain of salt, but it does make sense.

• For one DIMM per memory channel UDIMMs have slightly better memory bandwidth than RDIMMs (0.5%)
• For two DIMMs per memory channel RDIMMs have better memory bandwidth (8.7%) than UDIMMs

And this is where it just gets interesting…
So as you add more DIMMs to any memory channel, the memory speed drops. This is due to the electrical loading of the DRAMs that reduces timing margin, not power constraints.
If you don’t completely fill all memory channels there is a reduction in the memory bandwidth performance. Think of these configurations as “unbalanced” configurations from a memory perspective.

Source:
http://www.vlsiencyclopedia.com/2012/07/difference-between-rdimm-and-udimm.html


So… I chose RDIMM because my Lenovo mobo excepts a higher speed when using this type of DIMM. 1600MHz (RDIMM) vs 1333MHz (UDIMM), don’t ask me why it’s like this?? The way I looked at it originally, the RDIMM speed is greater than the UDIMM’s (0.5%) bandwidth advantage, when using one DIMM per channel.

So the question to you guys, do I move forward with placing only one RDIMM stick in each available channel, there are four total or do I place two RDIMM sticks in each channel, side-by-side, increasing bandwidth but taking a hit on speed? Very confusing… now you know why I created the thread, lol!


...
 
petteyg359 said:
My (possibly incorrect) understanding is that you avoid unregistered latency issues by keeping to a single-channel setup, but thereby reduce maximum bandwidth; while the registered modules might have greater latency in single-channel, they have better latency in dual-channel, and you keep the dual-channel bandwidth advantage.
Click to expand...

The main reason that the IBM x3550M3 (Westmere) supports RDIMM's is so that you can have more memory installed. You can only have 3 DIMM's per channel if you go w/RDIMM's. With UDIMM's you're limited to a maximum of 2 UDIMM's per channel @ 1066Mhz bus speed. The x3550M3 has 9 memory slots per CPU, w/3 slots per channel. The only, er, performance advantage, to having 3 RDIMM's per channel is that you can do 3-way interleaving. The trade-off is that you can only run the RDIMM's in this configuration at 800Mhz -- regardless of the rated speed of the RDIMM's. Registered DIMM's reduce the electrical loading of the memory data bus over UDIMM's.

You can have two single-rank or dual-rank RDIMMs installed at 1333Mhz, but only one UDIMM at this speed, so in this case, you could have greater interleaving possibilities for the RDIMM, so I guess it's possible you could get better performance in this scenario.
 
Last edited:
Ok, I'm trying to figure out clearly what your suggesting.

I have a Lenovo mobo, it has a two sockets per channel, colored black + blue. It's a typical setup, four sockets on each side of the processor, 4 channels total. Should I install four DIMMS together, filling two channels completely? Leaving four sockets empty for the time being.

Again, I went with RDIMM's, which I purchased in a 4pak, 8GB a piece. My intent was to install the DIMMS in every other socket, but that doesn't seem like the fastest configuration anymore, because it's RDIMM's. And if I'm understanding you correctly, the RDIMM's in pairs will run at 1333MHz.


...
 
Last edited:
I'm sorry but I didn't notice that JD has an E5-1650 CPU, which is LGA2011(Sandy Bridge-EP). Everything I wrote only applies to LGA1366 (Westmere-EP), an older model Xeon. Since it's a newer architecture, it might not have the same limitations the Westmere-EP did. If you only have 4 RDIMM's I'd probably try to populate all channels so I would get channel interleaving, which I've read is more effective than rank interleaving within a channel.
 
Thanks Megellan, I'll consider filling all the channels at some point... I may have to do some type of bench test for RAM to figure out the best configuration.

My only problem now, because I'm using two 8GB sticks initially, two channels will be open. I'm not sure if this is going to cause some type of slow down, maybe I should have purchased four 4GB sticks instead?? Basically I jumped on a good deal, I purchased RDIMM ECC memory for a UDIMM non-ECC price - from a brand name. My thought was I could sell the other two sticks at some point and if I couldn't... my fall back position would be to install all four sticks, never hurts to have more memory long term :)


...
 
Last edited:
JD, I'd like to see the results of your bench tests.

On my LGA 2011 system, I can do channel interleaving, but not rank interleaving, because I have single rank DIMM's. Supposedly both channel interleaving and rank interleaving are best for performance.

I've also read that having more than one DIMM per channel and having multi-rank, high density DIMM's can limit your memory overclocks as well.
 
My order for 32GB of RAM got canceled, out of stock, lol... So now I have a second chance at picking up some different RAM if I want, single rank vs dual rank, Kingston makes both types with same GB/speed stats. I only bring it up because single is faster than dual, so I've recently read.
Maybe I should go with the same spec'd RDIMM ECC, but instead 4 sticks of 4GB (single rank)? Filling all the channels and hitting 16GB, I would be good for a few years probably. I could buy another round of the same sticks (fill the other sockets for a total of 32GB), if I ever really needed it. I'm guessing I wouldn't, but looking 10yrs back, 4GB seemed like plenty, hahahaha!


...
 
Last edited:
 
Go with RDIM always if you can. You can't mix UDIM and RDIM. It makes sense to go with RDIMM if you plan on expanding to higher GB capacity later down the road. UDIMM will always allow less memory in the board. Don't waste your money twice.

IE: UDIMM you can do 24 gb, RDIMM you can go 32. Wouldn't you hate a year from now you maxed out with 24gb and wanted to upgrade to 32 just to find out you have to buy memory all over because the only way to reach 32gb is with RDIM. Hope that helps.
 
Based on everything I read, here and elsewhere, it seems if you plan to ONLY use 1dimm per slot, UDIMM will give you 0.5% speed increase. Pretty small, but it's there.

If you use 2 or more dimms per slot (Dell has some 3 per slot) then DEFINITELY should use RDIMM for 8.7% speed increase.

You should get RDIMM now for only a half a percent speed difference, you will have much more compatibility (and speed) when you upgrade later.

At least that's what I would do
~Wookie
 
You must log in or register to reply here.

Similar threads

HankB
Comments on this build?
2
Replies
21
Views
961
DaveB
DaveB
R1S8K
X79 vs X99 what is the difference ? also differences in Xeon CPUs and RAM types.
Replies
0
Views
1K
R1S8K
R1S8K
Yomama
Ryzen 5800x vs. 5700x
Replies
8
Views
984
Cisco Kid
C
D
  • Question
Need timings to finish "overclocking" RAM to run at actual speed (1600Mhz) when motherboard only accepts 1333Mhz unless "overclocked"
Replies
7
Views
1K
WhitehawkEQ
WhitehawkEQ
U
360mm good quality AIO vs custom loop for ~250-300W CPU
2
Replies
38
Views
2K
User5566
U
Back