E5-2678v3 - $90

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

zer0sum

Well-Known Member
Mar 8, 2013
849
474
63
The seller accepted $90 for a single CPU, and it's the same as an E5-2680v3.
It's not an amazing deal, but might be the best V3/V4 bargain right now along with the 2660v3 @ ~$80


12 Cores
24 Threads
2.50GHZ base
3.30GHZ turbo
30M Cache
120W TDP

Non AVX boost speeds
1588697856190.png

AVX boost speeds
1588697871425.png
 

awp

New Member
Dec 23, 2015
5
3
3
@zer0sum where did you get those charts? I might want to swap our E5 2667 v3s with these, but I'm curious on how they stack up. I've been considering switching them to 2699s since they're ~$800, but these @ $100 might be a decent bump for dirt cheap.
 
  • Like
Reactions: Samir

frogtech

Well-Known Member
Jan 4, 2016
1,482
272
83
35
So do people generally agree idle temps between chips will be about the same? For comparison I would be replacing dual E5-2630L V3, which obviously has a lower core count.
 
Last edited:
  • Like
Reactions: Samir

VMman

Active Member
Jun 26, 2013
128
47
28
Good price I purchased 10 last week from them and they would only take $95/ea :-(
 

zer0sum

Well-Known Member
Mar 8, 2013
849
474
63
@zer0sum where did you get those charts? I might want to swap our E5 2667 v3s with these, but I'm curious on how they stack up. I've been considering switching them to 2699s since they're ~$800, but these @ $100 might be a decent bump for dirt cheap.
Straight from Intel :D

The information here is also really good!
 
  • Like
Reactions: aloe and awp

aloe

Member
Aug 17, 2019
53
11
8
Bought one at $95, will be a nice upgrade from my 5930k for Proxmox server. thanks for letting us know about this deal.
 

e97

Active Member
Jun 3, 2015
323
193
43
Do you know if this is supported with Supermicro Mainboards and what is the impact on the power-usage for all turbo?
The gist of the mod is remove the CPU micro code at boot so the CPU initializes in a state where all cores are enabled for turbo, then load the microcode at OS time.

If the SuperMicro board doesn't require a signed BIOS firmware and you can modify it - it should work.

Re: power consumption, keep in mind my sample size is 2 and my E5-2678v3 is water cooled.

The power usage is about the same or a bit lower due to under-volting based on Kill-A-Watt readings. I was testing PSU efficiency at the time so I dont have exact data but remember at the wall usage being lower with all PSUs after unlock + under-volt at idle and load.

Might be a silicon lottery thing and water-cooling, too many confounding variables.
 

frogtech

Well-Known Member
Jan 4, 2016
1,482
272
83
35
that is a reasonably good price for a good CPU. Purchased some for 103 in December.

Added bonus is that with the right motherboard you can use DDR3 memory with this proc.
I believe you might have said somewhere it's because this CPU has both a DDR3 and DDR4 controller right? Does the existence of the DDR3 controller in the silicon have any performance/latency impacts you think? I'm assuming it's completely inactive on a DDR4 only board?
 

itronin

Well-Known Member
Nov 24, 2018
1,237
796
113
Denver, Colorado
hmmm not sure I said that originally. I did look these up on ARK once upon a time.

As for performance, I have a 73 and a 78 using DDR3 memory in "cheap chinese motherboards". They have radically different workloads than the 78's / DDR4 I have in servers so I don't even have a performance "feel" between the two of them.

I would hope that the chipset does some detection on what's going on and that the uCode is loaded appropriately to activate the correct controller but honestly IDK and am just guessing. Someone like @e97 would likely be more in tune with this.
 
  • Like
Reactions: frogtech

e97

Active Member
Jun 3, 2015
323
193
43
The chips were created for customers during the DDR4 shortage. Generally the hybrid design has poorer performance than a specific variant.

DDR3 and DDR4 pins are different. How they connect to the socket and how they connect to the IMC are factors. ARK shows the socket pinout. Without access to the chip design or X-rays, we can only check experimentally.

I use this motherboard with DDR3:

Screen Shot 2020-05-09 at 4.10.11 PM.jpg

There's a DDR4 version, X99-D8 :

Hc099439660d343ddaeec43ff996143721.jpg

If someone has it we can compare benchmarks.



Aside: when looking at Total Cost of Ownership (TCO) = Cost, Setup/Validation, Power => if going DDR4, modern chips have significant power savings and a smaller number of systems for the same work load.
 
Last edited:

BeTeP

Well-Known Member
Mar 23, 2019
657
430
63
Before CV19 struck I was thinking about getting Huananzhi X99-TF board to play with. It is an interesting and quite popular board which has both DDR3 and DDR4 slots. But then I got sidetracked and now I got a pretty long backlog of other things to do. So it is unlikely I would get back to this one before it would become obsolete.
 
  • Like
Reactions: Sleyk

andrewbedia

Well-Known Member
Jan 11, 2013
699
255
63
Do you know if this is supported with Supermicro Mainboards and what is the impact on the power-usage for all turbo?
Not sure about power impact, but I did this on X10SRH-CF today with a E5-2666v3 and it now all-turbos to 3.5GHz. Trying to see if there is a way to integrate the maximum effort loader as a DXE in the firmware image. I have no idea if this will flash normally--I used a chip clamp and a Raspberry Pi 4 to do it completely out of band.
 
  • Like
Reactions: Sleyk and itronin

nthu9280

Well-Known Member
Feb 3, 2016
1,628
498
83
San Antonio, TX
IIRC, it should reduce the idle consumption. I was looking try this last year when I saw the thread on AnandTech. I wasn’t sure if the 2u cooling is enough for the full turbo for dual procs and did not venture any further. @Marsh has been using the all-core turbo unlock.
 
Last edited: