Intel is introducing their first new line of EX 4-socket devices in years, welcome to the new Xeon E7 v2 2800, 4800, and 8800 lines. There are 20 new devices that make up the new chips so bear with us as we go through the SKUs as best we can.
Close followers of Intel know that the EX line of chips is the big one, traditionally 4+ sockets although they have recently come down to 2S configurations. These beasts are also on a two-year release cadence rather than the 18 months for the 1-2S EP line or the single year updates found in consumer parts. The last EX release was Westmere-EX from early 2011.
These chips were affectionately known as Xeon E7 2800, 4800, and 8800 for 2S, 4S and 8+S configurations. The new Ivy Bridge-EX “v2″ line has exactly the same names just without the v2 after the E7. This may lead one to believe that they are socket compatible and just updates to the Westmere based E7 line. If you deduced this on your own, you would be dead wrong, the two lines have absolutely nothing in common other than socket counts and the name. Intel marketing scores another own goal with confusing naming schemes, but that is about the extent of the negatives on this new line.
What do you want an E7 for? If you can get 2S and 4S devices in E5 or E7 versions, why would you want the more expensive E7? RAS, expandability, and throughput for the technical side but the real reasons are a bit less complex. The types of work that you need to do with a large EX system are characterized by lots of memory, very high throughput, and the need for extreme reliability.
For most tasks if you can’t do things just as well with four 2S EP servers, 4S or 8S machines will probably not satisfy your needs. That said they do get you far closer to enough performance and that is worth a lot to many customers. In short there are reasons for this line to exist, and if you need the capabilities, you are more than happy to pay for them.
What types of workloads is the E7 family aimed at? Intel lists such day to day tasks as HPC, OLTP, in-memory analytics, business intelligence, Crysis, CRM, ERP, data warehousing, high frequency trading, and virtualization consolidation. All but the last are known for large data sets that like to be memory resident, high throughput, and extremely high costs of downtimes. Virtualization consolidation is a cost of management play, and Crysis is an in joke among reviewers. In short E7 is aimed at big jobs, ones that take a bigger hit when split among many boxes.
So what is Intel offering with their E7 lines? That is a bit complex, starting with three lines plus two more “segment optimized” versions of the same chips. There are only three dies which are turned into 20 official SKUs with a bewildering array of options. These range from a mere $1223 for the 6-core E7-4809 v2 to a 15-core $6841 E7-8890 v2. As you might have guessed the core count ranges from 12-15 on the 8800 line, 6-15 on the 4800, and back up to 12-15 on the 2S 2800 line, all v2s. Power ranges from 105W to 155W, clocks start at 1.9GHz base and top out at 3.4GHz base with turbo allowing a few SKUs to hit 3.6GHz on one core.
20 options from rolling three dice
What differentiates the lines? In typical Intel fashion nothing coherent but there are some general trends. The first big difference is cache, the xx80 and xx90 parts all have 2.5MB of L3 cache per core, basically slices like Sandy and Ivy bridge E/EP. All of the others except the HPC oriented 8857 v2 have 2MB of cache per slice. This means the three die are likely a 10 core, 2MB cache per slice version, a 2MB per slice 15 core variant, and a 2.5MB 15 core monster. This could be incorrect since Intel didn’t say what the versions were, but this seems to be the logical breakdown.
In classic confusing marketing fashion the Basic line has Turbo fused off just because, QPI capped at 6.4GT/s, and memory speeds hamstrung to 1066MHz. Why anyone would want this part is beyond us but Intel makes them if you do. Luckily there is only one variant, the E7-4809 v2, so crippled.
The Standard line has five models that range from 2-8 sockets but most SKUs are in the 4S range. QPI is upped to 7.2GT/s and all have 2MB cache per core slice. Memory speed is raised to 1333MHz and the maximum core count is 12. These parts should make up a large number of sales with prices ranging from $1466 for the 4820 v2 to $3059 for the 8850 v2. This range looks good for those who outgrow an EP system but their task doesn’t scale well between boxes.
The real heart of the E7 v2 line is the Advanced segment, as we said earlier more is better in this game. Advanced brings more to the table, 12-15 cores and 2.5MB cache per slice for eight of the 15 SKUs. QPI is unlocked to the full 8GT/s and memory now maxes out at DDR3-1600. In short it is not crippled in any way.
Those three are the major categories, the last remaining models are the four segment optimized SKUs in three categories. There are two labeled as frequency optimized for enterprise, one low power, and a lone HPC frequency optimized version. All are 8S, have 2.5MB cache per core, full speed 8GT/s QPI, and support DDR3-1600. In short they are simply outliers that fit a certain task better than the mainstream models and are priced accordingly.
Reassuringly expensive the E7 v2 line may be when compared to the more mainstream E5 CPUs but that is nothing compared to the competition. Intel lists the competition for the E7 as IBM’s Power line and Sun’s Sparc chips, neither of which are known for rock bottom pricing schemes. Compared to those two the E7 v2 line looks pretty cheap.
For 4S systems, Intel says a top end 4490 is 1.8x faster than a 4S Power 750 Express with a Power 7+ CPU while being 74% cheaper. Against a Sun Sparc T5-4 4S machine the same 4490 is said to be 1.28x faster and 61% cheaper. Moral of the story #1: If you don’t want x86, the Sun T5-4 is your best bet. Moral #2: Compared to mainstream x86 devices, the E7 line may be painfully pricey but the non-x86 competition will make your eyes water. Moral #3: Although it wasn’t said by Intel, all of these machines will likely run software packages with price tags that dwarf the hardware so morals #1 and #2 are irrelevant.
In summary there is now a new line of high-end Intel server chips meant for 4-8 socket and beyond machines. While the new Xeon E7 v2 line will support 2S machines, the tasks this family of 20 new models are aimed at really starts above that count. With such huge core counts, performance numbers, scalability, and throughput, the prices scale to match. Then again for those who need these massive systems, the performance boost is welcome and the price is, relatively speaking, quite reasonable.S|A
Have you signed up for our newsletter yet?
Did you know that you can access all our past subscription-only articles with a simple Student Membership for 100 USD per year? If you want in-depth analysis and exclusive exclusives, we don’t make the news, we just report it so there is no guarantee when exclusives are added to the Professional level but that’s where you’ll find the deep dive analysis.
Latest posts by Charlie Demerjian (see all)
- What does Qualcomm’s server SoC look like - Apr 15, 2015
- How does Qualcomm’s SenseID fingerprint scanner work? - Apr 9, 2015
- How fast is Qualcomm’s 64-bit Kryo server core? - Apr 7, 2015
- Amazon is spending lavishly on game development - Apr 6, 2015
- What is the name of Intel’s Cannonlake +1 server platform - Apr 6, 2015