Intel Xeon E3-1280 v2 Review: Ivy Bridge Goes Professional

Status
Not open for further replies.

bit_user

Polypheme
Ambassador
Skip.

I went with a Sandybridge E5-1620 + discrete graphics. Twice the memory bandwidth. Twice the PCIe lanes. Comparable price. And the raw performance of the cores is only a couple % slower. A good tradeoff for GPU compute.
 

mousseng

Honorable
Apr 13, 2012
672
0
11,060
Okay, I'll take your word for it that a $600+ Xeon can be better value in certain scenarios than an i7. But how exactly is it better value than the ~$230 E3-1230v2, which (as far as I can tell) is exactly the same, only clocked a few hundred MHz lower? Is the need to squeeze every ounce of performance out of the server-class CPUs so great that Intel can demand a $400 price hike for 300MHz?
 
G

Guest

Guest
The 3Ds Max test doesn't make any sense unless you mention which renderer you're using (Mental Ray? Vray? Scanline?). Also it would be nice if you compared against desktop processors to see if it's worth splashing out on the Xeons
 

PreferLinux

Distinguished
Dec 7, 2010
1,023
0
19,460
[citation][nom]Skeletor1[/nom]The 3Ds Max test doesn't make any sense unless you mention which renderer you're using (Mental Ray? Vray? Scanline?). Also it would be nice if you compared against desktop processors to see if it's worth splashing out on the Xeons[/citation]
You don't buy Xeons for performance, you buy them for reliability. The performance for clock speed is exactly the same.
 

PreferLinux

Distinguished
Dec 7, 2010
1,023
0
19,460
[citation][nom]mousseng[/nom]Okay, I'll take your word for it that a $600+ Xeon can be better value in certain scenarios than an i7. But how exactly is it better value than the ~$230 E3-1230v2, which (as far as I can tell) is exactly the same, only clocked a few hundred MHz lower? Is the need to squeeze every ounce of performance out of the server-class CPUs so great that Intel can demand a $400 price hike for 300MHz?[/citation]
If you need the single-threaded performance, you need it. You can't get that performance by combining multiple systems. In servers or render farms, you can just add a few more machines to make up for the lesser performance, because they are dealing with tasks that are extremely well threaded – so you don't buy the fastest option, you buy the best value option. But in some cases, the single threaded performance is more important (certain workstation tasks) or you are limited to one system (many workstation tasks), so the performance matters more than value until the performance stops making a significant difference.

And I wouldn't say that it is better value, rather I'd say that it is necessary for the extra reliability.
 

mandrilux

Distinguished
Nov 11, 2011
2
0
18,510
Nice review, but i'd like to view a comparasion between E3-1245v2 or E3-1275v2 versus I7-3770 or I7-3770K over a motherboard with chipset Z77 like Asrock. Because the E3 is cheaper than I7 and supports same socket.
Thanks.
 

mandrilux

Distinguished
Nov 11, 2011
2
0
18,510
Nice review, but i'd like to view a comparasion between E3-1245v2 or E3-1275v2 versus I7-3770 or I7-3770K over a motherboard with chipset Z77 like Asrock. Because the E3 is cheaper than I7 and supports same socket.
Thanks.
 

ekho

Honorable
Aug 3, 2012
51
0
10,630
Intel doesn't compete hard these days.
It does whatever it wants.
AMD or ARM-BASED are not serious competitors at least for about next 2 years I guess.
 

silverblue

Distinguished
Jul 22, 2009
1,199
4
19,285
ARM doesn't feature in the workstation space, whereas AMD's Vishera/Delhi, whilst not perfect, could still be a good CPU - its performance gains over Zambezi/Valencia eclipse Ivy's over Sandy. Still, it'll only be aggressive pricing from AMD that really makes them stand out this year against the i7s.

I'm liking the v2 moniker; instead of inventing new codes, is it so hard to just attach a suffix like a version number of an a/b/c etc.? That's enough to convince people that it's comparable to an older model in speed, socket type etc. but the version number will denote improved performance.
 
Nice review Chris and thanks! :)

Translation to Real World - One thing that has often disturbed me is the duration of many of these benches, my experience is that they often either aren't relevant or worst aren't a good measure to real world jobs which often last for HOURS not 1~2 minutes. For comparison sake and perhaps scaling it would be nice to have a 'Part 2' with E5's and UP/DP/MP.

It took me a half cup of coffee to figure out why you choose the E3-1290, I got it once I realized the clocks.

Using Stock clocks the Ivy Bridge is a good step in the right direction, but other than it's Litho it's hard for me still to consider it a 'Tock'. I'm hoping the Haswell will correct some of the IB shortcomings.
 
G

Guest

Guest
This is a server chip. Where are the IO and database tests? You know, server tasks.
 
G

Guest

Guest
Would have loved to see the Xeon v2 coupled with a Z77 chipset board, just to see if there is any performance degradation compared to the C-series chipsets. AsRock, Gigabyte, MSI all support the Xeon V2s on their H77 and Z77 boards.
 

mousseng

Honorable
Apr 13, 2012
672
0
11,060

Okay, so basically it is that thing I said (the need for performance being that great). And yeah, I worded the whole 'value' bit pretty poorly, but you seem to have caught on to what I was getting at. Thanks!

Being a consumer with no knowledge of the enterprise/server sector of hardware, it's a bit difficult to see how something so seemingly small can be worth so much, but I often forget that businesses have a lot more money to spend than individuals like myself.
 

iamtheking123

Distinguished
Sep 2, 2010
410
0
18,780
[citation][nom]bit_user[/nom]Skip.I went with a Sandybridge E5-1620 + discrete graphics. Twice the memory bandwidth. Twice the PCIe lanes. Comparable price. And the raw performance of the cores is only a couple % slower. A good tradeoff for GPU compute.[/citation]
E5 is the full server version of Sandy Bridge. The equivalent won't be released for Ivy Bridge until next year, so the comparison isn't valid. This is just some re-badged client Ivy Bridge parts with minor enhancements.

FYI you wasted money if you bought E5 for home use. Overclockability, which Xeons don't have, is more important that memory bandwidth or pci-e lanes.
 

bit_user

Polypheme
Ambassador
[citation][nom]iamtheking123[/nom]E5 is the full server version of Sandy Bridge. The equivalent won't be released for Ivy Bridge until next year, so the comparison isn't valid.[/citation]What do you mean it isn't valid? I'm talking about what I can buy today.

And the E5-16xx series is targeted at workstations, as much or more than servers. You can find plenty of workstations on the market that use them.

[citation][nom]iamtheking123[/nom]FYI you wasted money if you bought E5 for home use. Overclockability, which Xeons don't have, is more important that memory bandwidth or pci-e lanes.[/citation]If you read my post, I said it was for GPU compute. Feeding hungry GPUs over 32 lanes of PCIe 3.0 requires just about as much memory bandwidth as you can throw at it. I don't overclock, because correctness and stability are more important to me than a little extra performance (which is mostly dependent on the GPUs, anyhow).
 

iamtheking123

Distinguished
Sep 2, 2010
410
0
18,780


Because you're saying "look a client part isn't better than a last gen server part" (Ivy Bridge E3 is a client part with a Xeon badge). Well duh. You're judging what Ivy Bridge can do before they've even done anything yet.

@PreferLinux,

FYI it's a myth that server parts are "more reliable" than client parts. It's different hardware like more cache, more IO's, etc but the testing process that each go through is the exact same. Both use +/- 3 sigma tolerances for determining failing limits (after the process matures more, the limits are tightened up).
 

bit_user

Polypheme
Ambassador
[citation][nom]iamtheking123[/nom]Because you're saying "look a client part isn't better than a last gen server part" (Ivy Bridge E3 is a client part with a Xeon badge).[/citation]Those kind of labels are irrelevant to me. I'm only looking at the facts: features, specs, and price. Based on that, I chose the E5-1620 as the best CPU for my GPU-compute workstation build.

Since you're hung up on labels, check out HP's Z400 series workstations. They use Intel E5-16xx series CPUs.

[citation][nom]iamtheking123[/nom]@PreferLinux,FYI it's a myth that server parts are "more reliable" than client parts. It's different hardware like more cache, more IO's, etc but the testing process that each go through is the exact same. Both use +/- 3 sigma tolerances for determining failing limits (after the process matures more, the limits are tightened up).[/citation]I can't speak to that, but I will point out that Xeon chipsets have RAS features like ECC support. That's certainly a source of additional reliability.

Server chips & chipsets tend to lag their mass market counterparts that are often just crippled versions of the same product. I always assumed the delay was to shake out some of early design errata and improve the manufacturing process to produce better quality wafers.
 
[citation][nom]A Bad Day[/nom]Intel: Compatibility? Standards? Screw that.[/citation]

Seems reasonable enough to me. I don't see why Intel should complicate things for motherboard manufacturers by using a 2000+ pin socket when they don't have to. Adding traces like that is expensive and server owners rarely upgrade them, they just buy whole new machines 3-5 years down the road. With that said, I'm not aware of any plans to bring Sandybridge to the LGA 1567 platform.
 
Status
Not open for further replies.