|
|
Overclocked Dual Xeon workstation reviewed
First Exclusive INQpressions The Asus Z7S
By Nebojsa Novakovic: Tuesday, 08 January 2008, 7:39 AM
UP TO NOW, the rules were simple: high-end PC, in their " enthusiast" editions were allowed to, and even built for, massive overclocking and other performance tuning modifications. Some of these, like cascaded freeze coolers and liquid nitrogen, almost remind one of souped-up street race cars.
On the other hand, dual-CPU workstation and server boards focused on stability and guaranteed factory-default performance. Overclocking? No way, Jose...
Well, now there is a way to have your cake and eat it - aside from gaming-oriented Skulltrail that Intel is launching soon, another player is out with an overclockable workstation board: Asus' brand new and uninspiringly named Z7S mobo.
Looks
Hey, it's a normal ATX board! Yes, two LGA771 Xeon Harpertown CPUs, heatpipe-cooled i5400 Seaburg chipset, six (asymmetrical) FB-DIMM slots for up to 24 GB RAM, all the power circuitry to feed these plus two PCI-E x16 v2 GPUs - CrossFire supported - and additional 64-bit PCI-X 133, PCI 33 and PCI-E x1 slots: all that in standard PC size. As you can see, that is quite a feat: once you take care of the cooling and such, you could actually fit this into a standard PC casing!
![]()
The mobo supports standard high-end PC 24+8+4 EPS class power supplies - the brand new ones based on 24+8+8 pin scheme could feed it a bit more power though, if it implemented that latest version. I ran it with Thermaltake ToughPower 1000 unit - I guess I don't need to state the PSU's obvious wattage?
The impressive-looking mobo has one thing I like a lot for either workstation or gaming/multimedia use: zero on-board sound. Being tired of crappy quasi-HiFi CPU-bound software audio, I'd rather have the space freed for, say, an extra PCI(-E) slot in this case, and put in a Creative X-Fi XRAM card to take care of that. If paying for a total of 8 expensive cores, why try to save 60 quid and bog down one of those cores with lousy audio?
Overclocking
I put it together with Asus EAH3870 TOP GPU (yes, the CrossFire also worked with two of them), 8 GB of Nanya FBD-800 1.5 v RAM occupying all four channels, a plenty of fans to cool all that hot stuff, and both Win32 and Win64 XP.
What about the CPUs? There was a whole spread of them to test with: I started with a pair of X5472 Harpertown Xeons (3 GHz, FSB1600, multiplier 7.5), then moved to X5482 3.2 GHz FSB1600 with multiplier 'lucky 8', and finally the same thing, but unlocked: QX9775 Extreme Core 2 pair...
Now, the overclocking options on this board aren't as sophisticated those on its desktop Republic Of Gamers cousins, but good enough to give it a go: for one, on the CPU menu, you can adjust the FSB speed and spread spectrum, and on JumperFree, the NB/FSB as CPU voltage - not much, but better than nothing. Also, this board's power delivery doesn't seem to support very high overclocking power consumption.
So, how far could I go then? With the X5472, I gave it the first try - despite its locked multipliers, I easily got 3.2 GHz / FSB1708. Keep in mind this is DUAL FSB1708, i.e. two parallel front side buses, one for each CPU, going to the chipset at that speed, and then running four FB-DIMM channels at an in-sync DDR2-854 MHz.
OK fine, I thought it seems to work, so why not move to X5482 then? Same FSB1708 worked again, but the higher multiplier gave me 3.42 GHz! BIOS booted, Windoze booted, both 32 and 64-bit, and all Sandra & Everest stuff went fine. And, I didn't yet touch any voltages, by the way - all default. Then I upped the NB / FSB voltage to 1.3 volts from 1.25.
I got dual 3.5 GHz / dual FSB1750 running fine - the CPUs were still at 1.22 volts! Not bad at all, a 112 GFLOPs overclocked dual Xeon seemingly stable at default CPU voltage. But dual 3.6 GHz / dual FSB1800 needed a 0.1v CPU jump to complete the 64-bit Sandra runs properly - for 32-bit runs, there was no need even for that. The FB-DIMMs ran at DDR2-900 CL5-5-5 speeds here.
![]()
Now, I tried higher speeds, even with higher voltages, but they couldn't go ahead - part of the reason could be the total power required, or FSB / FBDIMM failing to go up that high.
Since the unlocked QX9775 pair was there too, I decided to give them a try, hoping that Asus BIOS will see - and enable - multiplier settings then. Alas, no luck with that - I can only hope the next BIOS version will support that.
I proceeded to complete the test runs on the QX9775 setup then - here are the benchmarks...
![]()
![]()
As you can see, another set of synthetic benchmark records in this early run - for further better results, we may have to wait either for Skulltrail, or for newer Asus BIOS here. The memory result, at near 8 GB/s, is the highest I've seen for a 4-DIMM FBD configuration. If it had 8 such DIMMs, we'd have 10 GB/s here.
(Im)perfect board
The Z7S is a lovely board, but it's not a supermodel yet - there are a few things I'd comment on. Since there's no space for full 8 DIMM sockets (FB-DIMM bandwidth benchmarks do benefit when there are more DIMMs per channel, unlike desktop DDR2/3), I'd rather cut it back to just 4 DIMMs for fully symmetric memory bank, and tune the latency and speed paths to support those faster CL3/4 T1/2 FB-DIMMs meant for Skulltrail, together.
The saved space could then be used for stronger power circuitry to feed two well overclocked CPUs, as well as an extra fan for the heat-piped yet still hot North Bridge - hey, dual FSB1800 can't be cold...
On the current board, the BIOS needs more flexible memory configuration options, not just the basic four memory latency timings or command rate, but also interleave and ECC options common on Seaburg chipset. Cache snoop filter switching should also be provided, as not all apps like it. Of course, CPU Vcore and Vpll voltage grades should be better-defined, at 0.05 volts or finer.
Overall, prior to the big round of Skulltrail reviews, this is a good beginning for overclockable dual-socket workstation boards. Z7S complements, rather than competes with, Skulltrail. It supports CrossFire instead of SLI, dual v2 PCI-E GPU slots instead of quad v1 PCI-E, has PCI-X instead of more PCI, six memory sockets instead of four, and of course normal ATX size. Yet, they share the same chipset, use the same FB-DIMMs and of course both can support the same processors.
Good: Dual-socket, overclockable, all in ATX size - way to go!
Bad: a bit weak on power & cooling, and BIOS needs a lot more work
Ugly: nothing much, except that odd memory slot configuration?
Beers: Eight
![]() |
|
IP卡
狗仔卡
发表于 2008-1-10 01:03
QQ好友和群
收藏
分享
好贴
烂贴
提升卡
置顶卡
沉默卡
喧嚣卡
变色卡
显身卡
楼主