Back in 2011 Intel introduced its Sandy Bridge architecture, which went into a lot of the company's mobile, desktop, and server products. Before that, the Nehalem design did the same thing, driving a litany of products with familiar code names: Clarksfield, Bloomfield, Lynnfield, Gainestown, and Beckton should all bring back some fond memories. In fact, if you look back in history, there are a great many examples of Intel leveraging its work across multiple segments, optimizing each configuration as needed. With Haswell, that effort continues. We've already looked at the architecture's significance in the mobile and desktop spaces. Now it's time to visit the workstation and entry-level server market with Xeon E3-1200 v3.
A number of the product portfolio's features remain persistent between the Core and Xeon markets, just as they have since the Sandy Bridge era, when Intel introduced its first Xeon E3-1200 CPUs:
Similar clock rates with increasing IPC and lower overall system power each generation
- Dual-channel DDR3 memory support
- The ability to accommodate 32 GB of RAM
- Unbuffered memory only
- On-die graphics available
- ECC support for the Intel Xeon E3 line
- PCIe control is built into the CPU package and consists of 16 lanes
- Pricing relatively comparable between server and consumer desktop parts
In shifting from the LGA 1155-based Xeon E3-1200 to the LGA 1155-based Xeon E3-1200 v2, and now to the LGA 1150-based Xeon E3-1200 v3, processor speeds remain similar, with slight boosts in certain areas. Similarly, the amount of work each successive architecture can get done in one clock cycle continues improving slightly, but not staggeringly so. Memory support holds steady at up to 32 GB of unbuffered ECC.
Comparatively, Intel's high-end Xeon E5 brand, intended for more compute-intensive workloads, supports up to quad-channel memory configurations and registered DIMMs. That gives those LGA 2011-based platforms the ability to address hundreds of gigabytes of memory. Back when Sandy Bridge first surfaced, 32 GB seemed like a lot of RAM for a small server or workstation. In 2013, we see high-end desktops sporting that much (particularly easy across eight memory slots).
As you shop for business-class hardware, it's important to keep your application in mind. Do you need multiple processors in a machine? Is power consumption a big concern? How about price? From the Xeon E3 through the E5 line-ups, there exists a spectrum along which certain target environments fall. Xeon E3 is most definitely about small business servers, entry-level workstations, and density. Think Web hosting, light design work, and centralizing data in a branch office.
A notable feature that the workstation-oriented Xeon E3 models offer is the availability of Intel's HD Graphics P-series engine on-die. Hardware-wise, it's very similar to what you get on the desktop Core i7s. But Intel provides a special driver with the Xeon that is certified for applications from Autodesk, Adobe, SolidWorks, and Siemens. Naturally, discrete GPUs make the most sense in a high-end workstation. But in a more mainstream box, HD Graphics P4600 and ECC memory come together as a solid combination. It's only unfortunate that you can't get a professional derivative of the Iris Pro 5200 in the Xeon E3 line-up.
Intel employs the same naming convention as the previous two generations, illustrated in the graphic below. Today's story centers on the company's Xeon E3-1275 CPUs, but you can clearly see (and appreciate) that Intel is maintaining nomenclature consistency. This might have been difficult to decipher at first. Three architectures in, though, we know right where everything goes.
During the course of this piece, we'll look at the key similarities and differences between Sandy Bridge, Ivy Bridge, and Haswell, and what they mean for each workstation chip's performance. We're also digging into the platforms from Supermicro that we used for testing. After all, they're equal citizens in the quest for reliability in a professional environment.
Intel maintained processor interface compatibility between the Sandy and Ivy Bridge-based Xeon E3s, both of which dropped into LGA 1155 sockets. In most cases, upgrading from one to the other was achievable through a firmware update. With the introduction of Haswell, notable changes (such as the fully-integrated voltage regulator) necessitate a new interface called LGA 1150, which naturally requires updated motherboards and platforms.
| Processor | Architecture | Manufacturing Process | Interface | Introduced |
|---|---|---|---|---|
| Intel Xeon E3-1200 | Sandy Bridge | 32 nm | LGA 1155 | 2011 |
| Intel Xeon E3-1200 v2 | Ivy Bridge | 22 nm | LGA 1155 | 2012 |
| Intel Xeon E3-1200 v3 | Haswell | 22 nm | LGA 1150 | 2013 |
In addition to the interface change that happened between Ivy Bridge and Haswell, Intel also transitioned to 22 nm manufacturing between Sandy and Ivy Bridge, true to its tick-tock cadence. The primary beneficiary of this technology advancement was the mobile segment. But because Intel's Xeon E3-1200 series leverages the same architectures as Intel's Core i3, i5, and i7 CPUs, those mobility-oriented enhancements (primarily affecting power) make their way into the Xeon line-up, too.
Up until recently, the Xeon E5s were all based on the older 32 nm Sandy Bridge-EP/EN architecture. Intel introduced its v2 family at IDF earlier in the month. But the company is still one generation behind in the Xeon E5 series compared to these Haswell-based E3s.
In terms of specifications, the following table compares all three generations:
| Xeon E3-1275 | Xeon E3-1275 v2 | Xeon E3-1275 v3 | |
|---|---|---|---|
| # of Cores | 4 | ||
| # of Threads | 8 | ||
| Base Clock Rate | 3.4 GHz | 3.5 GHz | |
| Max. Turbo Boost Frequency | 3.8 GHz | 3.9 GHz | |
| Shared L3 Cache | 8 MB | ||
| Instruction Set Extensions | SSE4.1/4.2, AVX | SSE 4.1/4.2, AVX 2.0 | |
| Thermal Design Power | 95 W | 77 W | 84 W |
| Max. Memory | 32 GB, Unbuffered ECC DDR3 | ||
| Max. Memory Data Rate | DDR3-1066/1333 | DDR3-1333/1600 | |
| # of Memory Channels | 2 | ||
| Max. Memory Bandwidth | 21 GB/s | 25.6 GB/s | |
| Processor Graphics | HD Graphics P3000 | HD Graphics P4000 | HD Graphics P4600 |
| Graphics Base Frequency | 850 MHz | 650 MHz | 350 MHz |
| Graphics Max. Dynamic Frequency | 1.35 GHz | 1.25 GHz | 1.25 GHz |
| # of Displays Supported | 2 | 3 | |
| PCI Express Revision | 2.0 | 3.0 | |
| Quick Sync Video | Yes | ||
| vPro | Yes | ||
| VT-x with EPT | Yes | ||
| VT-d | Yes | ||
| TXT | Yes | ||
| AES-NI | Yes | ||
| TSX-NI | No | Yes | |
The changes from one launch to the next are apparent, and the most noteworthy evolution, ironically, might be processor-based graphics. Intel is leaning on its HD Graphics 4600 implementation; again, there are no Iris Pro 5200-equipped Xeons. The HD Graphics P4600 engine sports 20 execution units, or four more than the generation prior.
Supermicro set us up with two workstation platforms for our exploration. The first is LGA 1155-capable for the Xeon E3-1275 and -1275 v2 CPUs. The second, which we'll look at on the next page, is an updated version for the Haswell-based Xeon E3-1275 v3.
The SuperWorkstation 5037A-iL employs Supermicro's CSE-732D4F-500B mid-tower chassis. It's painted black, as the end of its model number suggests. From the outside, we only see a few key features; everything else is conservatively covered. Right off the bat, we notice the front panel's vented 5.25" optical drive bay covers, a pair of USB 2.0 ports, two USB 3.0 ports, and front-panel audio connectors.
This is a workstation case, so the top of the chassis doesn't have a fan mount. Instead, all airflow is directed from the front to the rear of the enclosure. In the back, we see the 500 W, 80 Plus Bronze-certified power supply vent, a 120 mm fan below it, and externally-accessible expansion slots. The side panel employs a combination of screws and a toolless latch to keep the system sealed up.
Inside the chassis we see the toolless 3.5" drive mounting solution, which can pivot out by simply activating a lever and giving the cage a nudge. Doing so provides quick access to four hard drive bays.
Disks are snapped into a plastic carrier; that carrier then slides into the chassis. This quick mounting mechanism eliminates drive screws.
Supermicro taps its X9SAE motherboard for use in the barebones platform. This is an interesting ATX motherboard because it centers on Intel's C216 chipset, enabling support for LGA 1155 processors and their integrated graphics engines. The X9SAE includes VGA and two HDMI outputs. As you likely noticed already, there's a big difference between the X9SAE's layout and what enthusiasts expect to find on desktop-oriented boards.
The platform's DIMM slots align with the case's front to back (or east to west) airflow. Enthusiast motherboards tend to have memory slots oriented top to bottom (or north to south), exploiting the airflow of those systems, which often involves exhaust up top. Clearly, this is a configuration for entry-level servers and workstations. That's why we're dropping Intel's Xeon E3-1275 and -1275 v2 processors into it.
Ethernet connectivity is handled by Intel 82579LM and 82574L controllers. If you're not already familiar with server and workstation networking, the 82574L is a roughly five-year-old controller. Its BOM cost is significantly higher than what a motherboard vendor would pay for a competing solution, but the benefit is great operating system support, speed, and stability. Even hardware-limited environments like Solaris and VMware ESXi work out of the box with Intel's 82574L because it's such a dominant building block in the server space. The 82579LM PHY, on the other hand, is integrated into the platform. Since it made its introduction alongside the Sandy Bridge architecture, many operating systems need additional drivers to get it working.
Storage support is typical for C216-based motherboards. There are four SATA 3Gb/s and two SATA 6Gb/s ports on-board. This is one area where Haswell-class hardware improves on the generations prior.
As a workstation, the A in Supermicro's model name indicates that the X9SAE includes up to 7.1-channel integrated audio, facilitated by Realtek's ALC889 codec. The board also includes a S/PDIF header for optical output.
Add-in cards are accommodated by one 16-lane PCI Express 3.0 slot, two second-gen PCIe x4 slots (one of which is exposed through a x8 slot), two, single-lane links, and a legacy 32-bit PCI slot. Coupled with Supermicro's SC732 chassis, the platform includes a retention mechanism for large discrete graphics cards.
Other notable platform features include four USB 3.0 ports and up to eight USB 2.0 ports for peripheral connectivity. Although there is only one fan that comes bundled, there are five on-board fan headers. A 64 Mb EEPROM with AMI firmware is also included, which should handle most system needs.
Since Intel keeps its Haswell-based platforms fairly stable in terms of interface size, memory configurations, PCI Express support, the motherboards look pretty similar (aside from the power circuitry integrated onto the die). Consequently, many vendors had little trouble updating their previous-gen products. This goes for Supermicro's SuperWorkstation 5038A-iL. As you can probably guess from the incremented model number (5038 versus 5037), this is an updated platform.
The chassis is a lot like what we saw from the 5037A-iL. Looking at the front, there are some obvious changes. Up front, we still have two USB 3.0 and two USB 2.0 ports, along with an audio interface. That I/O is augmented by two FireWire connectors on the updated case. Our test unit came with a 5.25" bay adapter, which you can see in the second external bay from the top.
Inside is Supermicro's X10SAE motherboard. This is where you'll find some of the most notable changes. For instance, the memory configuration is now top to bottom, which is more typical of consumer motherboards. That'd be a major change on a passively-cooled server board. But for the workstation-oriented X10SAE, it works fine.
In the newest generation, Ethernet comes from Intel's i217LM and i210AT controllers. This is notable because Intel is transitioning its customers to newer silicon. The i217LM is found on C226-based platforms, while the i210 succeeds Intel's 82574L. We've already seen a lot of Haswell-based server boards with these controllers. You may not find drivers built-in to your operating system, but support for most environments is at least available from Intel.
Supermicro enables eight SATA 6Gb/s ports, compared to the previous version's two 6 Gb/s and four 3 Gb/s connectors. Six of those ports are tied to Intel's PCH, while the other two come from ASMedia's ASM1061 on-board controller. The key takeaway is that there are more higher-throughput ports, which is a good thing from a workstation. We also find the same toolless 3.5" drive sleds that were available from the LGA 1155 platform.
Audio gets upgraded to Realtek's ALC1150. As with its predecessor, the X10SAE includes optical S/PDIF connectivity.
Expansion includes two third-gen PCIe x16 slots that either run at 16x/0x or 8x/8x, three PCIe 2.0 links in x4 slots, and two 32-bit PCI slots. The X10SAE's support for two x8 links lets you hook up high-end SAS card or a second GPU. That second PCI slot is interesting too; it may have been added due to other PCIe-based controllers that were already deployed on-board.
Rounding out the board's feature set is support for six USB 3.0 ports and up to 10 USB 2.0 connectors. Both generations accommodate two additional ports compared to the X9SAE model. The X10SAE has eight fan headers that can control speeds based on thermal readings and target acoustics. A 128 Mb EEPROM with AMI firmware is twice as large as the X9SAE.
The bottom line is this: the X10SAE is an upgraded platform that adds faster interface speeds and support for LGA 1150.
To test the processors, we ran them though the standard Tom's Hardware Windows 8 benchmarking suite.
Test configurations:
| System Test Configurations | |
|---|---|
| CPU | Intel Xeon E3-1275 (Sandy Bridge) and -1275 v2 (Ivy Bridge) LGA 1155 |
| Intel Xeon E3-1275 v3 (Haswell) LGA 1150 | |
| CPU Cooler | Intel Retail LGA 1155 and LGA 1150 |
| Platforms | Supermicro SuperWorkstation 5037A-iL with X9SAE Motherboard (LGA 1155) |
| Supermicro SuperWorkstation 5038A-iL with X10SAE Motherboard (LGA 1150) | |
| RAM | 16 GB (4 x 4 GB) Kingston Unbuffered ECC DDR3-1333 (Xeon E3-1275) 16 GB (4 x 4 GB) Kingston Unbuffered ECC DDR3-1600 (Xeon E3-1275 v2 and v3) |
| Common | |
| System Drive | Samsung 840 Pro 256 GB, SATA 6Gb/s SSD |
| Power | Supermicro, 500 W, 80 PLUS Bronze (included) |
| Software and Drivers | |
| Operating System | Windows 8 Professional x64 |
| Graphics Driver | Intel HD Graphics Driver |
| Benchmark Configuration | |
|---|---|
| Audio/Video Encoding | |
| HandBrake CLI | Version: 0.98, Video: Video from Canon Eos 7D (1920x1080, 25 frames) 1 Minutes 22 Seconds, Audio: PCM-S16, 48,000 Hz, Two-Channel, to Video: AVC1 Audio: AAC (High Profile) |
| iTunes | Version 10.4.1.10 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format |
| LAME MP3 | Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677, MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG2), Audio: MPEG-2 (44.1 kHz, Two-Channel, 16-Bit, 224 Kb/s) Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Abobe Creative Suite | |
| Adobe After Effects CS6 | Version 11.0.0.378 x64:Create Video, Three Streams, 210 Frames, Render Multiple Frames Simultaneously |
| AdobePhotoshop CS6 | Version 13 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| AdobePremiere Pro CS6 | Version 6.0.0.0, 6.61 GB MXF Project to H.264 to H.264 Blu-ray, Output 1920x1080, Maximum Quality |
| AdobeAcrobat X Pro | Version 10.0.0.396: Print PDF from 115 Page PowerPoint, 128-bit RC4 Encyption |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.95: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| Autodesk 3ds Max 2012 | Version 14.0 x64: Space Flyby Mentalray, 248 Frames, 1440x1080 |
| Blender | Version 2.64a, Cycles Engine, Syntax blender -b thg.blend -f 1, 1920x1080, 8x Anti-Aliasing, Render THG.blend frame 1 |
| Compression | |
| 7-Zip | Version 9.28, LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5" Benchmark: THG-Workload-2012 (1.3 GB) |
| WinRAR | Version 4.2, RAR, Syntax "winrar a -r -m3" Benchmark: THG-Workload-2012 (1.3 GB) |
| WinZip | Version 17.0 Pro, Syntax "-a -ez -p -r" Benchmark: THG-Workload-2012 (1.3 GB) |
| Synthetic Benchmarks and Settings | |
| 3DMark 11 | Version: 1.0.1, Performance Suite |
| PCMark 7 | Version: 1.0.4, System, Productivity, Hard Disk Drive benchmarks |
| SiSoftware Sandra 2013 | Version: 2013.01.19.11, Processor Arithmetic, Cryptography, Memory Bandwidth Benchmarks |
We'll see a pattern emerge over the next few pages: mainly, Ivy Bridge improves over Sandy Bridge, and Haswell is faster still. No generation saw the big gains that were evident in the mobile segment, but performance is up even still.
3DMark 11

3DMark 11 is a gaming benchmark, and not a core area of focus for the Xeon E3-1275's more workstation-oriented HD Graphics component. Still, we do see performance improve from one generation to the next.
Arithmetic

A look at our Dhrystone and Whetstone benchmarks suggests what we can expect in many of the subsequent benchmarks. Floating-point and integer performance improves in small increments. The other thing we see is that the 100 MHz clock rate difference and memory speed enhancement moving from the original -1275 and -1275 v2 yield an outsized advantage for the v2 part.
Multimedia

Haswell's AVX2 support translates to big gains in the integer module. That's perhaps the largest theoretical boost we'll see in moving to the -1275 v3, though it necessitates properly optimized software.
Cryptography

Our cryptography benchmarks are relatively similar from one machine to the next. Of course, because all three setups support AES-NI, they are as fast as their memory subsystems allow them to be. Hashing performance scales more predictably according to the architectural speed-ups from one generation to the next.
Memory Bandwidth

In terms of memory bandwidth, there is a clear fall-off for the Sandy Bridge part. This can be explained simply. When the Sandy Bridge-based Xeon E3 series launched, Intel validated it with DDR3-1333 memory. The subsequent generations added DDR3-1600 to the mix, even with ECC support.
Cache Bandwidth Results

Predictably, Intel's Xeon E3-1275 v3 shows one of its most decisive benchmark victories. The L1D cache results are a direct consequence of a pathway widening to 64 bytes per cycle in Haswell, from 32 bytes per cycle previously.
Adobe Photoshop CS6

As we would expect, we do see per-clock gains with each successive generation. The major question you have to ask yourself is whether those improvements are worth buying Xeon E3-1275 v3 for if you already own a Sandy Bridge-based part. Of course, seconds add up throughout the day. So, saving a bit of time here or there can definitely have a meaningful impact on your daily productivity.
Adobe Premiere Pro CS6

As we saw in Photoshop, the Adobe Premiere Pro results show a similar trend line. In fact, there's a pretty significant jump as you go from Xeon E3-1275 to -1275 v2. The 100 MHz frequency different and lower memory data rate slow Sandy Bridge down. The delta isn't revolutionary, but is definitely quantifiable.
Look at it this way: a $100/ hour artist gaining 15 minutes of working time a day is equivalent to a $25/day benefit.
Adobe After Effects CS6

After Effects CS6 likewise conveys a nice improvement from the Sandy Bridge to Haswell generation. Although the jump from Ivy Bridge is significantly more muted, we certainly do see a nice performance advantage with the new Intel Xeon E3-1275 v3.
Content creation applications generally provide very consistent results that are interesting from a number of angles. First, they are clearly a standard use case for workstation processors. Second, they tend to be sensitive to processor frequency, IPC, and memory data rate changes. Third, the benchmarks tend to scale well with many cores even across multiple interfaces. Those attributes make these types of benchmarks very attractive.
3ds Max

Our 3ds Max workload demonstrates a theme that is repeated often in this review: the Xeon E3-1275 v3 is the fastest, and then there are consistent performance increases over the years.
Blender

Here we appear to be measuring the effect of a 100 MHz speed bump and faster memory as we glide from the Xeon E3-1275 to the -1275 v2.
Cinebench

Cinebench demonstrates a steady speed-up with each successive architecture. This benchmark also lets us dig into the impact on single-threaded and parallelized performance. We like Cinebench because it's based on a real-world engine, and because it leverages as many cores as we throw at it (even from multi-processor Xeon and Opteron configurations).
Our productivity tests represent tasks commonly performed on desktop workstations.
ABBYY FineReader 10 
In FineReader, which is really well-threaded, Intel's Haswell architecture performs a little better than were expecting it to. The speed-up is only a few seconds, granted, but it's definitely quantifiable.
Adobe Acrobat XI

Looking at the PDF creation benchmark using Adobe Acrobat XI, these numbers are certainly solid. Picking up 15 seconds in a single-threaded test from Sandy Bridge to Haswell is nothing to scoff at. As someone who regularly converts PowerPoint presentations to PDF before I send them along, this metric resonates with me especially. It's worth minutes per day of my time.
Visual Studio

Our Visual Studio compile test runs noticeably faster with Haswell, opening up a multiple-minute lead over previous generations. This is certainly one of those tasks that is affected by a new processor architecture; it's both long and really well-optimized for threading.
Fritz

Fritz speeds up most during the transition from Sandy Bridge to Ivy, where it picks up an extra 100 MHz and support for DDR3-1600 on top of the minor tweaks Intel made to the architecture itself. Similar improvements moving to Haswell further nudge performance forward.
These applications are certainly important in the workstation space, where compressed data helps speed-up file transfers and minimize storage requirements.
WinRAR

WinRAR doesn't fully utilize four Hyper-Threaded cores, so it's probable that much of what we're seeing in our workload comes from optimizations for IPC in the architecture. Regardless, there's a nice transition from Xeon E3-1275 to -1275 v3.
7-Zip

7-Zip is better able to exploit available resources, and it appears to reward the 100 MHz-higher of the Ivy Bridge generation, in addition to reflecting per-clock tweaks to each architecture. Haswell operates at the same frequency, so its advantage is purely architectural.
WinZip

The WinZip results are in-line with our expectations on each set of tests, with the Xeon E3-1275 v3 clearly performing best.
Media work is exceedingly common in the workstation space. While a lot of the heavy lifting happens on multi-processor Xeon E5 and Opteron machines, single-socket Xeon E3s are still ample for more mainstream workloads.
TotalCode Studio

Our threaded TotalCode Studio test conveys an approximately 13% improvement moving from the Xeon E3-1275 to the -1275 v3. Of course, if you're working with lots of data, that speed-up can be quite significant in absolute terms.
HandBrake

Getting a Handbrake benchmark to work well across 32+ or more cores is more difficult than running it on a quad-core, Hyper-Threaded CPU, which it's already tuned to utilize.
As we've seen before, the largest speed-up happens from Xeon E3-1275 to -1275 v2, since we get all of the Ivy Bridge improvements, plus 100 MHz and DDR3-1600 support. Haswell maintains a lot of the same specs, but is still tuned for better per-core/clock performance.
iTunes

iTunes is a decidedly consumer-oriented piece of software, and our benchmark is single-threaded. So, the double-digit performance gains we measure going from Sandy Bridge to Haswell are attributable to a small frequency increase, memory bandwidth, and, most significantly, IPC improvements.
LAME

Our LAME results typically mirror iTunes, since both are single-threaded metrics. In this case, there's a little more improvement from Xeon E3-1275 to -1275 v2, while the shift from v2 to v3 conveys IPC enhancements.
With each successive generation, from Sandy Bridge to Haswell, Intel made deliberate efforts to improve platform efficiency, measured in performance per watt. Ivy Bridge benefited from a transition to 22 nm manufacturing. Haswell incorporates some specific operating state improvements that help bring down idle power use, though we noticed on the desktop that efficiency isn't always significantly better. Those same trends carry over to our comparative look at Intel's Xeon E3-1275 across three generations.

At idle, the Haswell-based Xeon E3-1275 v3 idles at very low power. Like desktops, workstations spend a lot of their time doing little, making this an important measurement.

This chart was an interesting one to generate. In 3DMark 11, we were only seeing our samples pulling 60 W from the wall. The key is that both the CPU and GPU need taxing workloads thrown at them simultaneously in order to present a worst-case power figure.

Our noise figures are less a reflection of Intel's Xeon E3 and more an indication of how Supermicro is contending with the thermal load of each platform. To measure this, we took measurements at idle and under duress using both barebones configurations and all three CPUs.
Not surprisingly, the Sandy and Ivy Bridge-based configurations fall within a margin of error. They demonstrate similar power consumption characteristics and the rest of the supporting platform doesn't change. Haswell isn't much different, either.
What we can say, however, is that all three setups operate quietly. Because there are fans involved, they aren't silent. But they're certainly not boxes you'd need to keep in a server room. Supermicro clearly designed these things to behave in an office environment.
Summed up, these are low-power and low-noise workstation processors, unlike the Xeon E5s that trade some elegance for massive performance. The simple fact is that the GPU-enabled Intel Xeon E3 chips are designed to be used desk-side. As such, they'll take up more space than some of the denser server-oriented options out there. In datacenters, the technology improvements introduced alongside Ivy Bridge and the operating state advantages that accompany Haswell should prove very valuable. For example, I pay $20/ month for each amp in my colocation facility, so saving a few watts to stay within a lower power pricing tier can save me $240 each year. There are colocation providers that charge a bit less, and a lot more than what I pay. There are also many providers that increase charges for spikes above a pre-set threshold. In those cases, the latest Xeon E3s offer a very tangible benefit that might not be as apparent in the workstation world.
After many years of working with these platforms on a daily basis, and many hours over the past few weeks running benchmarks, it's clear than Intel's Xeon E3-1200 family is progressing along in an evolutionary manner. By that we mean successive generations are yielding small but consistent performance increases. You can use our benchmark data to decide whether a Xeon E3-1200 v3 CPU is a worthwhile update based on your existing IT infrastructure.
The technologists in us wonder if this will become a pain point, though. Intel's competition has already expressed an intention to enter the server space with ARM-based designs that continue picking up performance at a rapid rate. Fortunately, from what we've seen in testing the Ivy Bridge-EP-based Xeon E5s and the Avoton SoCs leveraging Silvermont, Intel is prepared to battle.
What does all of that mean for Xeon E3? It'll almost certainly come under increasing pressure from within Intel and without. The latest CPUs certainly have their merits, though. In terms of power, Intel's new v3 generation shows off the architecture's emphasis on reducing idle consumption. Performance from the HD Graphics P4600 engine is also quite a bit better. We know the on-die subsystem is quite a ways off from what a discrete graphics card can do. But a growing collection of professional certifications at least make the integrated graphics component viable for entry-level design work. We'll tell Intel's workstation team the same thing we told the desktop crew: we'd love to see a version of the Xeon E3 with GT3e on-board, certified for professional applications.
A majority of Xeon E3s are sold into server environments where the HD Graphics component isn't needed. In those applications, improvements made to the Haswell architecture are really significant. Furthermore, as density continues to be emphasized by micro-servers for the cloud and other applications (like HP's Moonshot), having fewer components down on the motherboard helps enable smaller designs. Haswell's fully-integrated voltage regulator facilitates this. The architecture's advantages really map over well to the datacenter, then. Lower idle power use, better performance, and fewer on-board components are all great differentiators.
GPU-enabled versions, such as Intel's Xeon E3-1275, are meant for mid-range workstations requiring ECC-capable memory and no discrete graphics card. I can't help but think back to Chris Angelini's The Core i7-4770K Review: Haswell Is Faster; Desktop Enthusiasts Yawn, where his conclusion weighed the modest performance gains against existing LGA 2011-based solutions. The fact of the matter is that the entry price for a new LGA 2011-based CPU is similar to the Xeon E3-1275 v3 if you're planning to use an add-in graphics card. The bonus you get from either a Sandy Bridge- or Ivy Bridge-EP-based setup is that the number of cores you can get your hands on, total PCIe connectivity, and memory throughput are all greater than what the LGA 1150 platform supports.
At the end of the day, the fact that we have Intel's Xeon E3-1275 as a persistent offering through three generations means there are professionals buying this CPU. The addition of ECC support is notable, as are the improved power characteristics. Of course, purchasing a workstation CPU with on-die graphics over one without means that subsystem is important to you. And in that sense, the success of Xeon E3-1275 v3 is largely contingent on Intel's ability to add professional application certifications, continue enhancing performance, and expound on features like OpenCL support.









