• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

My SB-E 3820 Build (C2 Stepping)

Edrick

Edrick

Golden Member
Finally got my system built today. I am still playing with the settings so I will post what I can when I can. I am also going to compare some numbers with my old 2500K build (using same RAM). I am not a crazy OCer so I am not going to throw high voltages at this....most of my tests will be @ 4.0Ghz (for now).

i7cpuz.png


Temps are in the low 50's (running LinX) with a Corsair H60 and IC Diamond 7 compound. The Vcore is set to 1.28 currently, but I am going to see how low I can get that. *Edit* Vcore down to 1.236.
 
Last edited:
i5 2500K vs i7 3820
(Edit after I fixed a BIOS issue - See Post #14 for updated screenshot)

2500vs3820.png



One thing to note here, my P67 MB would not allow me to run 4 sticks of RAM @ 2133mhz for some reason. 2 was no problem, but not 4. With the X79 MB, I can run all 4 (same RAM) @ 2133mhz.

Sandra Benchmarks

Memory Bandwidth: 2500K = 24.5GB/s | 3820 = 45.49 GB/s
Memory Latency: 2500K = 19.4 | 3820 = 20.7

L1 Cache Bandwidth: 2500K = 569GB/s | 3820 = 549GB/s
L1 Cache Latency: SAME

L2 Cache Bandwidth: 2500K = 423GB/s | 3820 = 323GB/s
L2 Latency: SAME

L3 Cache Bandwidth: 2500K = 214GB/s | 3820 = 169GB/s
L2 Latency: 2500K = 15.7 | 3820 = 19.3

I am not sure if HT on the 3820 is playing a role in these numbers. As the 3820 has 4MB more L3$, I did expect the latency to increase, but I did not expect the bandwidth to decrease.
 
Last edited:
ill run my 2600k at your sandy e settings to compare.I have 4 sticks and my p67 board runs them fine.
 
gramboh said:
Question to OP: why did you switch platforms from SB to SB-E?
Click to expand...

My goal from the beginning was to go with an X79 system. I only bought a 1155 system to hold me over (I like to upgrade yearly). I did not go with the original 6 core C1 steppings as I did not need all that CPU power, so I decided to wait for the 3820. My plan is to drop in a IB-E chip sometime in Q3/Q4 before upgrading to Haswell sometime in Q2 2013.

Plus I like having 40 lanes of PCIe 3.0 the extra L3$. I was hoping for better Quad channel memory performance, so I am a little disappointed so far.
 
Edrick, I wonder how well this chip fares in games such as BF3. I would like to see the CPU usage if you ever have the time to test. (and if you have the game)

Especially with HT off. I want to see if a quad core is actually the bottleneck itself. I want to get one of these chips but those memory bandwidth results are making me cringe.
 
SLK said:
Edrick, I wonder how well this chip fares in games such as BF3. I would like to see the CPU usage if you ever have the time to test. (and if you have the game)

Especially with HT off. I want to see if a quad core is actually the bottleneck itself. I want to get one of these chips but those memory bandwidth results are making me cringe.
Click to expand...

I have BF3, so I will test that for you.

Also, I have not stopped cringing since I ran my first benchmark today. D:
 
its becasue the memory controller is made for server ecc ram,no worries man,in real programs that need the bandwith you will kill your old setup.

there has to be something wrong with aida since you are showing less bandwith with double the channels

if those numbers are right you have worse memory than a 2600k at 4.0ghz ?

wtf is going on there
 
grkM3 said:
its becasue the memory controller is made for server ecc ram,no worries man,in real programs that need the bandwith you will kill your old setup.

there has to be something wrong with aida since you are showing less bandwith with double the channels

if those numbers are right you have worse memory than a 2600k at 4.0ghz ?

wtf is going on there
Click to expand...

The throughput "doubles" (roughly) in multi-threaded accesses, but the access speed (i.e. latency) of a memory access per thread is much slower on SNB-E.
 
Fixed the memory bandwidth issue. Now it looks like I expected it to look like. The issue was I had to disable Channel and Rank Interleaving in the BIOS. (I had them enabled on P67 platform). Now I am not cringing as much.

i7_41fixed.png



And look what happens with the Sandra Benchmarks!!!

(Before) Memory Bandwidth: 2500K = 24.5GB/s 3820 = 18.78 GB/s
(After) Memory Bandwidth: 2500K = 24.5GB/s 3820 = 45.49 GB/s !!!!!
 
Last edited:
could you not run the memory at the same timings ? 8-9-8-27 vs 9-11-11-28
lots of cycles left on the table maybe.
 
rgallant said:
could you not run the memory at the same timings ? 8-9-8-27 vs 9-11-11-28
lots of cycles left on the table maybe.
Click to expand...

8-9-9-27 was at 1866mhz
9-11-11-28 is at 2133mhz

Both are as fast as my RAM will run at each speed. I simply could not run my RAM at 2133 on my old P67 system.
 
LinX GFLOPS (before RAM fix) @ 4.0Ghz = 59.8 w/HT

LinX GFLOPS (after RAM fix) @ 4.0Ghz = 86.83 w/HT

Can someone please post a 2600K or 2700K score with HT?
 
I have a feeling that you will jump back to 1155 when ivy hits retail.Intel made ivy with even better latency.If you dont need the 45gb bandwith I see no reason for anyone going x79 right now

ivyzu.jpg
 
grkM3 said:
.Intel made ivy with even better latency
Click to expand...
Then I look forward to IB-E with improved latency on the quad channel controller. 🙂

I have a feeling I may jump back to the mainstream platform only when Haswell hits the streets only because I am impatient and cant wait for the Haswell-E. I really wish they would release Haswell high end first.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top