2 Matched E5-2620v2 Pairs

I have a total of 4 E5-2620v2s that I'm looking to part with.
They were previously used in my ESXi hosts but I've since upgraded the hosts and no longer need these processors.

I'm asking for $300 Ea. or $500 for a pair.

You can see a pic of all 4 CPUs here http://i.imgur.com/8NMytP9.jpg

 

I'd probably suggest eBay to sell these, as anyone on these forums will likely be buying the sub $70/each e5-2670 v1's before spending $300 on an inferior CPU (higher clock speed and higher core count). You could probably sell these on eBay if you get the right buyer.

 

Just asking... what do you mean by matched pairs ?

You have 2 sets of 2 or a total of 4 of the same CPU.

 

Hello @William, based off the post and the picture, he has (4) e5-2620 v2 CPU's for sale.

 

hehe yes

I have heard that term "Matched Pairs" many times before in the OC crowd. People seem to think there is actually such things and matched pairs of CPU's will perform better. So I was just wondering what he was referring to

 

Jebus... I didn't realize 2670v1s were so cheap now. I knew they were coming down in price but $70...
I guess I'll toss them on Fleabay.

Sorry for causing confusion. By matched pairs I meant something more along the lines of "These two CPUs worked together fine previously and should keep doing so"

 

I always thought Matched Pair meant they were of the same stepping.

 

The OC guys would say they are special matched CPU's by Intel and would perform better. They might also say stuff like CPU Batches that come from the center of the wafer OC better and would start calling out batch numbers. In some cases they might work out that way but more often than not it was a simple crap shoot on CPU's for better performing ones.

We do know that at least as far as AMD was concerned that binned CPU's that had high leakage on transistors would OC better. They would get far hotter than normal CPU's and could take LN2 temps and scale good. They marketed some of these called Tweakers for a while. Yes they OC'd very good and were very bad for air cooling systems.

 

Interesting side story regarding the OC guys' opinions on performance and matching. I was a chip designer at Intel/IBM/Sun for 14 years, with 10 years focusing solely on microprocessors, and a technical basis exists for saying that die from the center of a wafer would have tighter optical control, which could result in a higher speed bin or greater ability to overclock because timing variations across the die would probably be minimized relative to die on the perimeter of the wafer. Notwithstanding the general principle that die at the center are probably tighter controlled, individual variations among die at the center could easily be a greater factor in performance variation than proximity to the center of the wafer. I had only a little exposure to what manufacturing data was externally knowable, but I would be surprised if the location of a particular die within a wafer was programmed into that die or publicly available from the manufacturer (but I don't know that this is not possible).

Strongly agree, also, that transistors with higher leakage can perform better. In MOS transistors, as the turn-on voltage lowers, leakage rises as the square of the voltage change. Additionally, switching current increases as the square of the voltage change. A higher switching current allows individual nodes to switch quicker because node capacitances discharge quicker.

All of this is probably more than anyone cares about but I've learned so much from this group on system issues (which I had no exposure to before) that I do what I can to pay it back.

 

Good info and thanks for sharing Dave

Much of what the OC guys were saying back then was only guess and them trying to come up with solutions. I really don't think any of them knew anything about where certain batch numbers come from on any part of a die. I would think that batch number only referred to a series of die's made in a batch and not a single wafer. It was known that a certain batch of a specific CPU might perform better. An example is I had heard that a certain batch of i7-920's OC'd very well. I found a seller on eBay that had a bunch of these, non ES chips so they were retails in boxes. I grabbed one and paid a hefty price for it but it did OC very good and I was able to get several WR's with it on a EVGA Classified motherboard. I think that was pure luck tho on the batches.

To me the biggest factor was like you said how leaky they were. This often meant they OC'd badly on air or water cooling but could really take off on Phase/Cascade and LN2. I did this many times back then when I had all that equipment. I only have a monster phase unit left from those days now.

The single biggest thing that everyone went for back then were ES Chips and the whole ES black market that was going on in those days. I learned this the hard way also. I had a Q9650 I had gotten from newegg and put it under LN2 and did pretty good with it, I thought. I then went to the bot and I was very low ranked I noticed all the top scores were done with ES chips and after I learned all about what was going on I stopped doing scores on the bot as it appeared things were rigged. That's another story tho LOL

But back to matched pairs. I think this was a marketing term some company used to up prices on equipment. I believe what they did was match the VID or Core Voltage between two CPU's and call them a matched pair and it became a buzz word. As far as I know Intel never sold CPU's in pairs calling them matched. There might be some merit in this but I would think that on a dual processor system manufacturing tolerances negate this for the most part and the differences between Xeons of the same type are very small and do not warrant a matching process.

 

This makes a lot of sense, William. Identifiers on the package likely correlate to wafer lot, and a wafer lot is a series of wafers fabricated at essentially the same time, so that these wafers have very similar electrical characteristics. Die-to-die variation is very low within a single wafer, and is pretty low between wafers in the same lot.

Yeah, I don't have specific experience with ES chips but I would expect that they were unlocked because engineering samples are put through a "shmoo" process where functionality is checked over a wide span of voltage and frequency characteristics. The word "shmoo" comes from the graphical representation of the results often looking like the shmoo character. By finding out where holes or marginalities existed in the shmoo result, engineers saw where the design sensitivities were. Those sensitivities were probably smoothed over through timing adjustments before first production silicon.

From a chip designer's perspective, a "matched pair" would only matter for performance purposes if both processors were required to run at the same supply voltage and clock frequency. I don't know enough about dual-processor boards to know whether that is the case. But if it was the case, we all know that higher supply voltages and higher operating frequencies (all other things being equal) lead to CPU performance, so performance of the pair would be limited by the processor whose max supply voltage or operating frequency was lower.

Another angle on this issue that presumably doesn't apply here is that processors on some (mainframe) systems operate in lockstep, where 2 or more processors execute the same instructions at the same time, and instruction results are compared on an instruction-by-instruction basis. This system is conceptually similar to parity checking in memory, but applied differently of course. But I digress ...

 

Thanks again Dave.
For the most part you confirm what I thought what I thought all along about these things.