Intel recommend using two identical CPUs in their socket-2011 dual-socket motherboards; it is theoretically possible to use non-identical CPUs, though, and Intel’s rules for that are that they must be:
- Of the same processor family
- Have the same number of cores
- Have the same cache size at each level of cache
- Able to find a common QPI link frequency
Given the above, Intel states that you can mix CPUs with different core frequencies – the faster CPU will clock down to match the slower one, however. The same applies to the QPI links.
We haven’t yet had the opportunity to test differing CPUs in other brands’ dual-socket motherboards yet – we will update if or when we do. This isn’t something which we would usually recommend without finding someone who has successfully tested your intended combination just in case, but it is handy to know that it is theoretically supported.
On the top of any modern Intel CPU you will see the country of assembly printed, and it will be either Malay or Costa Rica. Back in the day this could mean the difference between a chip with excellent overclocking potential and a dud; how about now?
In a nutshell, no more difference than you get with normal batch-to-batch variation. Intel’s Copy Exactly program apparently works as intended – controlling every possible controllable variable involved in the assembly process to eliminate any differences. You can read more about Copy Exactly on Intel’s website (link)
– the origins of this process actually date back to the 1980s.
So if you end up with one or the other, relax – neither one is going to automatically mean a better chip.
We got our first Intel X520-T2 10Gbe cards in this week; benchmarks and a review won’t be available for a couple of weeks yet but we’re expecting big things…
Stay tuned for more!
For those moving from 1155/1156/1366 you might be a little surprised at just how much bigger the new CPUs are; it’s little wonder that they require a twin-lever clamping system on the motherboard.
Customer’s old CPU on the right (Intel i7-870, Socket 1156) and their new CPU on the left (i7-3820, Socket 2011). That’s a lot
of extra CPU real-estate for all those extra pins – almost 900 more. It’s quite a bit heavier in the hand, too.
We finished our first Ivy Bridge 3770 build today; here’s a few teaser shots before a full write-up:
More to come soon!
Intel have released some information about ECC in their networking cards;
Some of the older chips like the 82571 – found in cards like the Pro/1000 PT single/dual/quad port NICs – actually do have error correction on the in-band traffic, which is good news. You can see that the latest generation (i.e. i-350, i-540 etc.) have ECC on both the in-band and out-of-band (management) traffic – that’s on top of benefits like lower TDP and CPU consumption.
Of note is that the more common chips for onboard Intel Gigabit LAN ports – e.g. 82574L – doesn’t feature ECC or parity at all on either in-band or out-of-band traffic. For those who are trying to create the most stable, secure system possible this is a consideration which may prompt you to look at some of the newer network cards which do have those features. I would think that for the majority of users it’s unlikely to have a significant impact on your data long-term but if you’re buying new, it pays to have all the facts. Also, for those of us who are paranoid about data corruption, well, there’s now one more place you can have ECC for your peace of mind…
One of the more common quad-port network cards that pops up online is the Pro/1000 PT. This is a 2006 Intel design – discontinued in 2009 – which generally seems to be priced anywhere from AU$120-160 as a server pull. It uses two Intel 82571 chips, each controlling two gigabit ports, and is PCI-E 1.0 x4. The chips are widely supported and work out of the box with most operating systems and hypervisors. The card is also compatible with PCI-E 2.0 and 3.0 lanes, and there are both low-profile and full-height variants.
One of the appealing features these cards have – beyond four gigabit ports, of course – is teaming (aka trunking or bonding, falling under IEEE spec 802.3ad). This allows you to create links between the card and other devices which span more than one gigabit port, allowing you anywhere from a single gigabit connection to using all four to pipe your traffic through. Note that you won’t get 4Gb/s in a single transfer that way – it will be four gigabit pipes rather than one 4Gb pipe. This is particularly useful if the network traffic going into your server is exceeding a single gigabit link and can result in a noticeable improvement in network performance – so long as your switch supports 802.3ad too.
The card’s TDP is 12.1W and the chip heatsinks get reasonably hot to the touch if you don’t keep adequate airflow over the card during operation; they’re cards designed for use in high-airflow server environments so if you’re putting them in a quiet home server you may want to consider how much airflow the cards will get. The controllers can be passed through for hypervisors such as ESXi; keep in mind that you’re passing through one or both controllers rather than the ethernet ports themselves, so you can only pass the ports through in pairs.
This card represents quite good value for the home/SMB user who needs more ports, either to separate network traffic or to alleviate bandwidth congestion. The wide compatibility is also an advantage for those using motherboards without existing Intel network controllers and it should work out of the box with just about any modern OS, including ESXi. At about a third of the cost of one of the new variants (i350-T4) these are definitely a card to consider if you’re looking for an Intel NIC but don’t want to buy new.