Intel Broadwell-EP Xeon E5 v4

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Second part.

http://vr-zone.com/articles/broadwe...s-reliability-per-core-performance/66845.html

Sometime in the summer of 2015, very likely close to the IDF fall 2015, Intel is expected to release Broadwell-EP or Xeon E5 2600 v4 — its first 14 nm Xeon processor. What might we see in there?
Continuing the strategy seen in Ivy Bridge-EP and Haswell-EP, Intel will not exactly speed up the cores, but simply pile up more of them on each die. Expect to see up to 18-core Broadwell-EP chips, with multiple die variants: one of them is expected to be eight-to-10 core high performance desktop and fast-core workstation one for situations where less cores but with speeds above 4 GHz (before Turbo, of course) are desirable.​
BDW-EP5.png

BDW-EP8.png

BDW-EP15.png


Simply amazing. And way beyond the reach of any competitors for many years.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,143
136
Very impressive performance gains over the next two years.

Expect to see up to 18-core Broadwell-EP chips, with multiple die variants: one of them is expected to be eight-to-10 core high performance desktop and fast-core workstation one for situations where less cores but with speeds above 4 GHz (before Turbo, of course) are desirable.

Possible 10-core Broadwell-E coming 2015?
 

paul878

Senior member
Jul 31, 2010
874
1
0
From the chart, it seem that performance gains is due to the increase of in the number of cores.
 

Homeles

Platinum Member
Dec 9, 2011
2,580
0
0
From the chart, it seem that performance gains is due to the increase of in the number of cores.
Well, to me that's obviously the point of these chips anyway. But 50% more cores and two generations results in 80% more performance, so it's clearly not only "more cores" doing all the work.

The features certainly sound interesting. It'd be nice to be debriefed on what those features actually do. Bandwidth monitoring in particular sounds like a great tool.
 

mikk

Diamond Member
May 15, 2012
4,293
2,382
136
From the chart, it seem that performance gains is due to the increase of in the number of cores.


Mostly from the cores of course but Haswell itself comes with 5-10% higher IPC and the bandwidth also differs. AVX2 can further improve things (over Ivy Bridge-EP at least if supported). The slide says BDW-EP in OLTP 80% faster than IVB-EP. A 50% cores increase can't reach 80% alone. This is a case maybe where either the bandwith plays a role or AVX2.
 

ClockHound

Golden Member
Nov 27, 2007
1,111
219
106
Fascinating. 160W TDP for the workstation SKU. And some say Intel wouldn't be making hot chips anymore. 160W at 14nm could be cooling fun for overclocking. They'll let us overclock them, right? They could even call it Gulftown II. ;-)
 

MisterMac

Senior member
Sep 16, 2011
777
0
0
How does ORACLE\IBM fare in general these days?

Increasing\Decreasing the ammount of sold bigiron units?
(Not counting %, just raw volume)?


Seems like intel is gunning to takeover all enterprise level workloads from top to bottom.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
Mostly from the cores of course but Haswell itself comes with 5-10% higher IPC and the bandwidth also differs. AVX2 can further improve things (over Ivy Bridge-EP at least if supported). The slide says BDW-EP in OLTP 80% faster than IVB-EP. A 50% cores increase can't reach 80% alone. This is a case maybe where either the bandwith plays a role or AVX2.

OLTP gain is probably due to Haswell's IPC improvements, and the surrounding platform enhancements to allow good scaling with extra cores.

Anyway, its nice but not what I would call "amazing". It something they *require* to fend off increasing competition in the server space.
 

mikk

Diamond Member
May 15, 2012
4,293
2,382
136
OLTP gain is probably due to Haswell's IPC improvements, and the surrounding platform enhancements to allow good scaling with extra cores.


Haswell-EP is rated with 1,35 while Broadwell-EP 1,8. This is a 33% improvement with 28% more cores. I doub't we see a 100% scaling from the cores. Higher Bandwidth could help or Broadwell itself comes with a slightly improved IPC.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
I doub't we see a 100% scaling from the cores.

That's basically what its showing, or close to it.

OLTP is probably not *that* bandwidth sensitive, since we're seeing only 10% faster memory. But beyond what "feature list" shows, there's probably some enhancements the engineers put on various parts of the CPU to enhance scalability(QPI/Ring/Router, etc). What happens is that there's probably lot more headroom to improve performance on OLTP and Database scenarios than ones that are more akin to PC or workstation usage, like SPEC, hence bigger gains.

I don't think there's IPC gain on Broadwell. Last Tock by the same engineers that worked on Nehalem had zero IPC improvements. SpecInt figures show Haswell EP getting figures that's in line with core increase + IPC, but not so for Broadwell.

SpecInt IVB to HSW: 28%(17% gain in cores with IPC improvement)
SpecInt HSW to BRW: 24%(28% gain in cores with loss in scaling)
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
That's basically what its showing, or close to it.

OLTP is probably not *that* bandwidth sensitive, since we're seeing only 10% faster memory. But beyond what "feature list" shows, there's probably some enhancements the engineers put on various parts of the CPU to enhance scalability(QPI/Ring/Router, etc). What happens is that there's probably lot more headroom to improve performance on OLTP and Database scenarios than ones that are more akin to PC or workstation usage, like SPEC, hence bigger gains.

I don't think there's IPC gain on Broadwell. Last Tock by the same engineers that worked on Nehalem had zero IPC improvements. SpecInt figures show Haswell EP getting figures that's in line with core increase + IPC, but not so for Broadwell.

SpecInt IVB to HSW: 28%(17% gain in cores with IPC improvement)
SpecInt HSW to BRW: 24%(28% gain in cores with loss in scaling)

Westmere had IPC gain over Nehalem. Ivy Bridge IPC gain over Sandy bridge.

So there is nothing poiting to Broadwell not having the same. Plus Broadwell contains new instructions too.
 
Last edited:

Ajay

Lifer
Jan 8, 2001
16,094
8,114
136
Westmere had IPC gain over Nehalem. Ivy Bridge IPC gain over Sandy bridge.

So there is nothing poiting to Broadwell not having the same. Plus Broadwell contains new instructions too.

I thought that IPC gains for Westmere were < 3%, except for MT workloads?
In any case, thanks for the Broadwell-EP info!

Once we see the Broadwell-DT benches, we'll have an idea of what to expect from Broadwell-E. It's looking like Broadwell-EP is following HW-EP pretty quickly compared to historical norms. It seems this is driven by Google, et al, who really need a lot of cores and better perf/watt.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,143
136
So is more info about v4 Xeon E5's still expected to come out at the August IDF 2015?

According to the latest rumours Broadwell Xeon E5 is missing from Intel's new roadmaps, they will launch Skylake-EP/EX sooner instead.

Also the slides are outdated, Haswell-EP ended up having 18 cores.
 

PaulIntellini

Member
Jun 2, 2015
58
4
71
I don't think the slides were outdated. more likely secretive/deliberate.

14C is the max for "mainstream" Xeon E5 v3 skus.
 

t0mt0m

Member
Apr 21, 2015
35
2
36
They're showing in the Purley roadmap. But no word about Xeon E5 v4's in rumours? Strange if it's out by Q1 2016?
Curious as the E5 seems fairly well used for servers (and also for the v2 what's used in Mac Pro - so presumably v4 Xeon E5s would be the successor as Mac Pro has skipped v3 so far).
 
Last edited:

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Broadwell-E/EP is just a minorish update compared. Purley get a lot of attention due to its platform.

Intel-Xeon-E7-E5-Skylake-EX-_Purely-Platform_Roadmap.jpg

95c.jpg

big-advance.png