E5-2696 V2 CPUs for $325!

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

fossxplorer

Active Member
Mar 17, 2016
554
97
28
Oslo, Norway
OMG! I though i was dealing with some kind of big dataset of ~100GB on my MariaDB instances. But 10 freaking TB...:eek:
How do you or Postgres handle memory for this sizes of DB? (I know 0 about Postgres).

I've only tried FreeBSD so far. Performance is just sick. Today I will try the new release of PostgreSQL 10 on a 10TB dataset. Still though, I'm really busy with other things the coming days so it may take a few days before I have some feedback.
 

BackupProphet

Well-Known Member
Jul 2, 2014
1,083
640
113
Stavanger, Norway
olavgg.com
Im gonna give PostgreSQL around 32 GB of ram of 128GB, the rest is being cached through ZFS ARC, with FreeBSD 11.1, you also have ARC compression which helps A LOT, I'm also careful with my index strategy so I can fit most of hot the indexes in ram and don't have to go my SSD's too often. Ive picked up a lot of 800GB Micron SSD's for under $200 recently :D
 

Davewolfs

Active Member
Aug 6, 2015
339
32
28
No one around here buy xeon to do things less than at least 8-12 cores.... if you want an all around cpu at best price performance, 1950x is the way to go. if you want cheapest multithread/$. 2696v2 it is
Mixed workloads are what workstations are all about.

Intel is taking so much time with their W Xeons by the time they are released the next kid on the block will be right around the corner.

The new 8th gen chips seem to have a good speed boost.
 

MBastian

Active Member
Jul 17, 2016
205
59
28
Düsseldorf, Germany
No one around here buy xeon to do things less than at least 8-12 cores.... if you want an all around cpu at best price performance, 1950x is the way to go. if you want cheapest multithread/$. 2696v2 it is
I'd consider that but DDR4 prices are insane. For myself I want a good blend of single and multi-thread performance for my personal workstation. V2 prices are still way to high for my taste, so I'll take my time. All I want is another 1-2 years out of my current setup.

I am just counting GHz here and also omit the small IPC boost from v1 to v2.
v2comp.png
 
Last edited:

Davewolfs

Active Member
Aug 6, 2015
339
32
28
You can probably get $200 for 2x2670s. So under $400 all in for the boost. Not bad really. Basically 25% bump in perf over 2670.

If using all cores expect another 15% for the 2690 and another 15% for the 2696.

Eg Cinebench scores will be ~ 2500, 2800, 3100.
 

Davewolfs

Active Member
Aug 6, 2015
339
32
28
Just took a look at an invoice for the last DDR4 stick I bought. 32GB for $169. They have gone up in price significantly.
 

MBastian

Active Member
Jul 17, 2016
205
59
28
Düsseldorf, Germany
If using all cores expect another 15% for the 2690 and another 15% for the 2696.
I've read some review (cant't remember from which source) that v1 to v2 has aprox 5-10% IPC boost, depending on the task. But I am still unsure if just adding up the max-turbo#cores clock is the the right thing to do.


v1comp.png
 

BackupProphet

Well-Known Member
Jul 2, 2014
1,083
640
113
Stavanger, Norway
olavgg.com
Running benchmarks is hard when almost every system I have overheat :(

On Ubuntu run the following
Code:
sudo apt install python-numpy libopenblas-base
Then try running the following python code
Code:
import timeit
import numpy as np
size = 300
a = np.random.random_sample((size,size))
b = np.random.random_sample((size,size))

def dot_matrix():
    np.dot(a,b)

if __name__ == '__main__':
    print(timeit.timeit("dot_matrix()", setup="from __main__ import dot_matrix"))
Watch your temperature....

With dual 2696 v2 I got 641.057282925 (lower is better) with some slight thermal throttling.
My workstation with i6700K just rebooted every time I ran this benchmark. Will try it on my Haswell gaming pc when I come home.
 

BackupProphet

Well-Known Member
Jul 2, 2014
1,083
640
113
Stavanger, Norway
olavgg.com
It seems like numpy doesn't like too many threads, it scales almost linear up to 8 cores, then it only gets like 20% faster with 16 cores, and from there it gets "slower".

I have the same problem with PostgreSQL, parallel queries scales linear up to 8 cores, about 20% faster with 12 cores, and from there it just get slower.
So there is definitively a lot of applications that scales badly with too many cores.

Other workloads where each thread has a independent task scales a lot better.
 

BackupProphet

Well-Known Member
Jul 2, 2014
1,083
640
113
Stavanger, Norway
olavgg.com
I have another interesting update about these cpu's

Turbo boost works up to 3.5Ghz, but it is not stable there even with one active core, it jumps between 3.3Ghz and 3.4Ghz with some spikes up to 3.5Ghz. When all cores are active and no AVX instructions are in use, the speed is stable at 3.1Ghz. With AVX, speed is stable at 2.7Ghz, but if it gets too hot, over 80 celsius they may downclock to 2.6Ghz and 2.5Ghz.
 

Xfinity

New Member
May 21, 2012
13
0
1
Just wanted to drop by and thank all participants in this thread! Good discussions and facts in this thread! Upgraded my dual 2670v1 rig to 2x 2690v2 @ $300 a piece. Wanted something between the 2696v2 and 2667v2 after researching and what my needs are. Once again, thanks from a lurker ;).
 

wildpig1234

Well-Known Member
Aug 22, 2016
2,198
443
83
49
Just wanted to drop by and thank all participants in this thread! Good discussions and facts in this thread! Upgraded my dual 2670v1 rig to 2x 2690v2 @ $300 a piece. Wanted something between the 2696v2 and 2667v2 after researching and what my needs are. Once again, thanks from a lurker ;).

what MB do you have? good to hear you upgrade ok. i had the s2600cp MB that didn't want to play with the 2696 v2 requiring also a MB upgrade to z9pe-d16.

2690 v2 is not bad. slightly faster at single core, slightly slower multithread due to having 2 less cores compared to 2696 v2...
 

Xfinity

New Member
May 21, 2012
13
0
1
I have the Z9PE-D16 MB so I could have bought the 2696v2 straight away. I need the slightly faster cores for now :).
 

wildpig1234

Well-Known Member
Aug 22, 2016
2,198
443
83
49
Has anyone tried these E5-2696v2 processors in branded servers like HP and Dell?
if the computer originally came with a v1 cpu and didn't come with a v2 cpu, most likely they are locked down to v1 cpu only by a boot block . google and you will see that this is the case for all of the big three hp, Lenovo, dell.. and the boot block is not easily upgradable thr any normal process at all... another ploy to force people to buy new WS without letting people know beforehand that they are at the end of the road with no upgrade path....

if you did have one that does have v2 cpu originally, there is a chance but not a guarantee. I've asked this also in a separate thread but got no answer yet...

Dell T7610 support OEM cpu like e5-2696 v2?

I checked a benchmark site but was not able to see a dell t7610 with the 2696 v2... so the question is still out there..

UserBenchmark: Dell Precision T7610 Compatible Builds
 
Last edited: