IPC Difference in generations of chips?

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
Looking at my next move. I was ready to go thread ripper but the cost of ECC UDIMM DDR4 and DDR4 memory in general is making me balk.

I need at least 8 cores at 3+ghz because my workloads can use the cores sometimes and the ones that can't love higher frequency (VC1 Transcode from Plex, World Of Warcraft). Ryzen is on-edge or out due to Linux issues (iommu on my R7 1700 was a mess). The IPC difference is getting decent between the newest gen and the V2 but is it really that much?

Looking at these 8 core chips:
E5-2667 V2 3.3Ghz

+4Ghz single core
+Cheap ECC Ram
+Under $400
-Old tech
-IPC lower

E5-1680 V2 3.0Ghz
+3.9Ghz single core
+Cheap ECC Ram
+Overclockable
-Expensive
-Old tech
-IPC lower

E5-1680 V3 3.2Ghz
+3.8Ghz single core
+May be overclockable
+Newer Tech
+Decent priced motherboards
-Expensive
-Expensive ECC Ram

I7-7820X 3.6Ghz
+4.5Ghz single core
+Highest IPC
+Overclockable
+Price ($549 at microcenter with $30 off motherboard)
- No ECC Ram
- Expensive Ram
- Expensive Boards, mainly gaming boards
- Runs hot

Any STH opinions?
 

MiniKnight

Well-Known Member
Mar 30, 2012
2,997
907
113
NYC
What's the difference in price at the same configuration level?

V2 is Ivy Bridge and i7-7820x is Skylake?
Sandy Ivy Haswell Broadwell Skylake

Each jump is 6-8% over previous gen in IPC but it's cumulative. Ivy to Skylake is 1.06 ^3 to 1.08^3 or 19% to 26%
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,045
1,583
113
CA
I've always thought IPC was simple, 6% improvement = 6% improvement but now I'm wondering...

Does it depend on what the CPU is actually doing? Or is it an across the board 6% improvement?

IE: Some is 6% improvement, but other tasks remain the same?
 

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
What's the difference in price at the same configuration level?

V2 is Ivy Bridge and i7-7820x is Skylake?
Sandy Ivy Haswell Broadwell Skylake

Each jump is 6-8% over previous gen in IPC but it's cumulative. Ivy to Skylake is 1.06 ^3 to 1.08^3 or 19% to 26%
Ivy is $375 for CPU, $270 for board, I have the RAM
Haswell is $600 for CPU, $270 for board, $400 for DDR4 ECC (64GB)
Skylake is $549 for CPU, $250 for board, $500 for DDR4 NON-ECC Ram (64GB), $100 for cooler

It is around $600 more to go 19-26% more IPC. That is a pretty big price to pay for the performance difference. The advantage to the newer stuff is for resale down the road it will be worth more than the older tech. I have a bad habit of building, using for a while and flip for something else.
 

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
I've always thought IPC was simple, 6% improvement = 6% improvement but now I'm wondering...

Does it depend on what the CPU is actually doing? Or is it an across the board 6% improvement?

IE: Some is 6% improvement, but other tasks remain the same?
I think it is a "it depends". The 1680 V3 is slower than the 2667 v2 clock speed wise, but would that 6% make up for a few hundred mhz?
 

MBastian

Member
Jul 17, 2016
72
12
8
Düsseldorf, Germany
For what it's worth Anandtech had a IPC roundup from Sandy Bridge to Skylake. Interesting but I am a bit sceptical, especially that "Dolphin" benchmark seems skewed(AVX?).
I for myself will will wait patiently until the E5-2667v2 ot maybe the elusive E5-2673v2 will come down in price. Imgo the E5-2690v2 or E5-2697v3 are just not worth it if you're not desperate for more cores and cache.
 

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
For what it's worth Anandtech had a IPC roundup from Sandy Bridge to Skylake. Interesting but I am a bit sceptical, especially that "Dolphin" benchmark seems skewed(AVX?).
I for myself will will wait patiently until the E5-2667v2 ot maybe the elusive E5-2673v2 will come down in price. Imgo the E5-2690v2 or E5-2697v3 are just not worth it if you're not desperate for more cores and cache.
The dolphin bench may use CPU extensions not in the V1/V2 chips. Skylake looks good, the new X299 chipset is meh and it sort of sucks that new Xeon generation went a different path so it won't be like the E5's that you can put in consumer boards.

ECC is not a killer for me, I prefer it though.
 

cliffr

Member
Apr 2, 2017
77
31
18
42
I have a bad habit of building, using for a while and flip for something else.
That's the way they get you. You can buy old and refresh every 24 months at 20% of the IPP. Or you buy new and have it last 60 months at 100% IPP.
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,045
1,583
113
CA
I think it is a "it depends". The 1680 V3 is slower than the 2667 v2 clock speed wise, but would that 6% make up for a few hundred mhz?
I'm not sure we're talking about the same thing.

What I was trying to say was assuming CPU1 = CPU2 in terms of frequency, but one is a generation newer than the other if they're doing the exact same thing on BOTH, is there any "task" that the older gen could do just as good as the newer? OR, is the 6% IPC an across the board increase no matter what "task" the CPU is actually doing. I don't know enough about CPU architecture to answer this myself.

Now, regarding your question would 6% make up the difference of a few hundred mhz... well you could figure that out near exactly if you go with the 6% IPC gain, know the exact mhz difference, # of cores, etc... However, if my concern above about some tasks might not matter... then that may be a factor too?

I'm curious :)
 

Nanotech

Active Member
Aug 1, 2016
595
99
28
40
Just as a heads-up, we should have a piece comparing 3 similar Intel Xeon E5 v3, V4 and Scalable processors tomorrow. It is written, just may get preempted by some other coverage we will have.
Patrick will you also include V1/V2 for generational ipc differences? I think it gives a better perspective seeing as how V1/V2 processors are still somewhat popular (due to price/performance).
 

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
For now I decided to go with the E5-2667 V2 setup. I have the ram & cpu cooler, picking up a X9SRA board and the CPU for around $650 total. This should hold me off for a bit to watch how Threadripper and X299 work out and not get bit as an early adopter. Hopefully DDR4 will drop a bit in price.

Edit:
Passmark wise the E5-2667 V2 is pretty stout compared to newer chips
PassMark - CPU Performance Comparison
 

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
Got the E5-2667 V2 in. It runs real nice. CPUZ's benchmark put it a tiny bit faster than a stock 5960X (Haswell-E) and a little bit slower than a 6900K (Broadwell-E). Overclocked those two would beat the 2667 v2 hands down but I am not OC'ing this system.
 
  • Like
Reactions: cactus

Nanotech

Active Member
Aug 1, 2016
595
99
28
40
Got the E5-2667 V2 in. It runs real nice. CPUZ's benchmark put it a tiny bit faster than a stock 5960X (Haswell-E) and a little bit slower than a 6900K (Broadwell-E). Overclocked those two would beat the 2667 v2 hands down but I am not OC'ing this system.
It's definitely a nice Ivy Bridge-EP processor. It's only 100mhz slower than the E5-2687W V2 but turbo's on 1/2 cores to the same speeds as the 2687W V2. Considering that the 2667 V2 is only 2-3 generations behind and can match the 5960X and come close to the 6900K with only a 300mhz/100mhz advantage that's still impressive for a 2013 Ivy Bridge-EP. It's a shame CPU-Z removed older processors from the benchmarks that used to be in the older versions.
 

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
It's definitely a nice Ivy Bridge-EP processor. It's only 100mhz slower than the E5-2687W V2 but turbo's on 1/2 cores to the same speeds as the 2687W V2. Considering that the 2667 V2 is only 2-3 generations behind and can match the 5960X and come close to the 6900K with only a 300mhz/100mhz advantage that's still impressive for a 2013 Ivy Bridge-EP. It's a shame CPU-Z removed older processors from the benchmarks that used to be in the older versions.
Cinebench R15 (mult/single core)
E5-2667 v2 - 1294 / 141
E5-2687W v2 - 1227 / 141
I7-5960X - 1300 / 147
I7-6900K - 1476 / 179
Dual X5650 - 1279 / 93

The Cinebench score for the 6900K shows more of a difference than the CPUZ bench, but we are still under 15% difference over all those generations. The E5-2667 V2 beats the Westmere-EP by a larger margin than the 6900K beats the 2667 V2.
 
  • Like
Reactions: jfeldt

Nanotech

Active Member
Aug 1, 2016
595
99
28
40
Cinebench R15 (mult/single core)
E5-2667 v2 - 1294 / 141
E5-2687W v2 - 1227 / 141
I7-5960X - 1300 / 147
I7-6900K - 1476 / 179
Dual X5650 - 1279 / 93

The Cinebench score for the 6900K shows more of a difference than the CPUZ bench, but we are still under 15% difference over all those generations. The E5-2667 V2 beats the Westmere-EP by a larger margin than the 6900K beats the 2667 V2.
Interesting result for the 2687W V2 multi-threaded Cinebench performance. I would think that the 2687W V2 would be the faster processor but it's slower by 67 points in Cinebench R15 multi-score. Considering both turbo to 4.0Ghz on 1/2 cores I wonder why the 2667 V2 is better?
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,045
1,583
113
CA
Curious, with this small variations in performance why do you guys care?

Are you doing something that matters?

Or, more of just "I have the fastest" ?

Genuinely curious :) as I have lots of not needed performance ;) ;)
 

TType85

Active Member
Dec 22, 2014
597
173
43
Garden Grove, CA
Interesting result for the 2687W V2 multi-threaded Cinebench performance. I would think that the 2687W V2 would be the faster processor but it's slower by 67 points in Cinebench R15 multi-score. Considering both turbo to 4.0Ghz on 1/2 cores I wonder why the 2667 V2 is better?
I think the results I found were a bit low. Anandtech article below puts it dead on with the 2667 v2 where it should be.

Intel Xeon E5-2687W v3 and E5-2650 v3 Review: Haswell-EP with 10 Cores