Enterprise, Hardware

Nehalem-EP Part 2: 16-thread Task Manager pixellated


Overkill VR is a fast paced VR action game

Following our first preview article, we continue today with article focusing on the BIOS options, as well as the initial Vista 64 findings on this huge 48 GB machine before going into the benchmark realm in the Tuesday’s Part 3.

Naturally, this being a workstation machine which can also be used as a departmental server or visual supercomputer cluster node, Super Micro has included a whole range of BIOS options that you’d never, and I mean never, expect to see on any desktop motherboard, no matter how high end. We included some of the BIOS screenshots in the first part of our article, and we’ll go into more advanced options of the BIOS here.

DIMM Socket Distance pitch... 425 (mm?) from the CPU socket

DIMM Distance pitch… 425 (?) between two DIMMs

Just look at some of them, such as DIMM socket distance pitch, which is useful if you also have the ability to fine tune the RAM timings (well OK, that pitch would be fixed at the motherboard design stage – you can’t exactly move the DIMM sockets around!). I have not yet tried to manually adjust the altitude and temperature sensitivity settings for the DIMMs to see what impact, if any, they have.

Advanced options for enabling Virtualization...and PCIe Gen1 compatibility selector.

Advanced options for enabling Virtualization…and PCIe Gen1 compatibility selector.

The usual choice of managing so many memory channels is there, by either interleaving them for maximum bandwidth – which is always the choice on a workstation or extreme desktop – or instead creating memory mirroring; or even lock-stepping the two CPU’s memory arrays for selected mission critical use. There are also more CPU and North Bridge options than usual, including added virtualization settings including virtualized I/O support.

Advanced BIOS options... workstation owners rejoice, enthusiasts - demand that Intel comes out with V16

Advanced BIOS options… workstation owners rejoice, enthusiasts – demand that Intel comes out with V16

Super Micro is known for its very conservative performance settings, as there are practically no adjustments possible here: no changes in CPU multiplier [even if you had an unlocked processor], no voltage or even fan speed control changes, no base system clock adjustments, and of course no memory latency or voltage controls. As mentioned before, the option to force the memory speed to a higher clock than the default can be selected, but doesn’t work. Also, since the system clock here is 133 MHz and not 133.3 MHz, the CPUs end up working at 3,192 MHz instead of 3,200 [or probably 3,208 on an ASUS] so the benchmark results may end up this little bit slower than some other 3.2 GHz W5580 systems – more about that on Tuesday.

Continued on the next page…

Now, the Vista 64 SP1 installation disk handles the on board SATA and there is a little confusion here; according to the Super Micro manual, this board features an ICH9R but SiSoft Sandra 2009SP3 sees it as an ICH10R. Due to the lack of time, we did not remove the ICH heat sink, and the install was done flawlessly. In the future, we’ll be having another install done using the hardware RAID controller as well.

I ran the newest beta of SiSoft Sandra Business 2009SP3 version, which fully supports the Nehalem-EP platform. Prior to the benchmarks, here are some fairly comprehensive system info screenshots from Sandra:

Yes, the obligatory money-shot is here – 16 threads in Task Manager!
3.08 GB taken after boot and opening Sandra alone...This means Vista cached everything that it can.

3.08GB of RAM taken after boot and only by loading Sandra 2009. This means Vista cached everything that it can. Still, responsiveness is on the side of Windows XP x64 and upcoming Windows 7…

It is funny looking at the Vista Welcome Screen showing 48.0 GB RAM, and even funnier is seeing that just the OS itself blew up to a 3 GB "chunk" of it before any apps are running. Even though Vista likes to cache most commonly needed files, praxis showed that you don’t exactly get a boost from it… especially given the fact that our machine is powered by SSDs. At least on this machine, it’d be very hard even for Vista to run out of memory – maybe swap space should finally be disabled for good.

Also, please take a look at the average temperature while running the Task Manager (and a benchmark in the cut window portion). The other CPU is quite a bit hotter as not only it is farther from the main system fans, but also it is very close to the hot 4870X2 card – even though the Asus "Harley Davidson style look" card is far cooler than the ATI reference one. Either way, the CPU cooling is overdue for a replacement with something stronger.

Sandra’s 2009 System Info
Our Nehalem is clocked at 3.2 GHz, but turning Sandra on enables Turbo mode and the CPU goes to 3.33 GHz

Our Nehalem is clocked at 3.2 GHz, but turning Sandra on enables Turbo mode and the CPU goes to 3.33 GHz

There are a few interesting things to note in this otherwise pretty self explanatory screenshot series. The motherboard and chipset are detected as both SLI and Crossfire capable. This would go nicely if Intel didn’t pull a VIA and decided to act against nVidia. Given the fact that nVidia wants court orders preventing Intel from using nVidia patent portfolio, X58 losing SLI might be the first thing to go. In fact, the Tylersburg 36D chipset is identified as the X58, which it basically is, except for added extra QPI link & logic to handle two CPU and optionally dual IOH chips.

Nehalem-EP in the nude, Gainestown core

Note that the integrated memory controller is working at 2.66 GHz

Note that the integrated memory controller is working at 2.66 GHz

The CPUs are detected as running pretty much all the time (when there’s a load) at 3.33, not 3.2 GHz due to the 1-step Turbo benefit. This is a bit conservative compared to Core i7 Extreme CPUs, but bear in mind this is a workstation motherboard. It will be interesting to see what the competition will unlock.

More system details, this time memory
Finally, no more crappy GTL+ FSB and its limited bandwidth... if it states 25GB/s, there is a pretty good chance it will be 25GB/s

Finally, no more crappy GTL+ FSB and its limited bandwidth… if it states 25GB/s, there is a pretty good chance it will be 25GB/s

Also, the newest CPU-Z 1.50 information and notice some lovely things: firstly, even at the 3.45 GHz operation in parallel with running a benchmark, the CPU seems to run at just tad above 0.9 volts. There were pretty much constant variations there.

Clockin’ it to 3.45 GHz CPU-Z
Gainestown core, 3.45 GHz

Eight cores at 3.45 GHz is all you can reach on this "non-overclockable" motherboard, but impressive number is the Core Voltage… less than 1.0 Volt

Sometimes, at a Turbo level below this, 3.33 GHz, the CPU voltages would vary between 0.9 and 1.2 volts too, but if the D0 stepping is that good, we got a really nasty upward performance margin to make use of here. Then, the memory has a pretty decent CL7 latency setting for registered DDR3-1066 mode, and a T1 command rate was stable despite 2 DIMMs per channel.

X58+ICH10R combination

Specification mixup - Super Micro claims motherboard has ICH9R, CPU-Z detects ICH10R chip

Specification mixup – Super Micro claims motherboard has ICH9R, CPU-Z detects ICH10R chip

48GB of DDR3-1333 memory with 1T Command Rate
New high-density DDR3 memory modules started to lower latencies... will we see CAS5 modules?

New high-density DDR3 memory modules started to lower latencies… will we see CAS5 modules?

So, overall good stuff – dual 3.33 GHz performance at stock voltage sounds good to me, and yes, it is easier than ever for vendors like Asus or DFI to move a rank up from ROG or LanParty single-socket X58 (Tylersburg 36S) mainboards to dual-socket i5500 (Tylersburg36D) ones, rather than just depending on the shaky situation with the Intel’s successor of Skulltrail, the infamous V16 project. The cooling and per-socket wattage are same, ie up to 200W per socket in a dual-CPU version of, say, Rampage II Extreme; memory systems can be identical just doubled for two CPUs, and both the performance and I/O scaling (quad-SLI/Crossfire) gains are there to be taken. The only thing missing, that the Intel Enterprise division should take notice of and offer the "High Performance Edition" multiplier unlocked flavor of W5580 to match these boards, and we’re all set for some fun times: how does dual 5 GHz system potential sound?

In the meantime, let’s look at the benchmarks on the gear we got right now, and watch out for the Tuesday’s Part 3!