|
ServerWorks are an altogether more interesting company than Intel. Beavering away almost unnoticed by the regular computing masses, Intel don't really see ServerWorks as a threat, more of a chipset partner much like VIA is to AMD. This is why they are so interesting, more so than Intel. Lack of a public face but a massively high reputation in the server space, ServerWorks have been suppliers of high quality multi processing platforms for a long time. Indeed, for a long time while Intel were reluctant to push their own products in this arena, instead focussing on their battle at the time with AMD, ServerWorks had the multi-processor Pentium 3 server market sewn up pretty much from day one. Till this day, with companies who invest in the server space reluctant to change hardware that works, ServerWorks have been supplying solutions that do just that. Their older GrandChampion Pentium 3 solutions are famous for reliability, performance and support in a market that is notoriously tight with its money and which doesn't like to spend on untested products. So ServerWorks have enjoyed recent success without shouting about it and that's enabled them to take a shot at the high end workstation segment now that AMD has vacated the premises for the time being.
With Intel granting ServerWorks a Pentium 4 bus license from the get go, their P4 products have carried weight in a market that simply wouldn't have tolerated VIA's P4 fumblings, had VIA indeed decided to play up here.
So is GCWS competitive with E7505? Let's have a look.
Taking our information from ServerWorks' own product matrix here we can see that GCWS is a seemingly complex beast.
With either 3 or 4 bits of silicon making up GCWS, lets take a look at them in turn like we did with E7505.
CMIC-LE is the equivalent to Intel's E7505 MCH and is what you'd normally call the Northbridge. Featuring a similar pair of memory controllers to E7505, CMIC-LE supports more memory than E7505 does at 16GB. While the product matrix on the ServerWorks site is incomplete and mismatched with the actual hardware specifications, the controllers on CMIC-LE are DDR266 compatible giving the same ~4.2GB/sec memory I/O bandwidth as E7505. ServerWorks have a reputation for excellent memory performance from their memory I/O subsystems and it's fair to say that will be carried forth to the CMIC-LE bridge.
The processor support is identical to E7505 with a shared max ~4.2GB/sec CPU I/O path (533MHz again with 64-bit data bus per processor) to the CMIC bridge. No independant busses like with 760MPX and AMD but the bus protocol the Pentium 4 uses to share the bus with another processor means that bus contention is at a minimum.
CSB6 is the southbridge I/O processor for GCWS and it's slightly less up to date than ICH4 in terms of high speed serial I/O with no support for USB2.0. Only the legacy 1.1 spec is implemented. However the PCI bus provided by CSB6 is of the 64-bit, 33MHz variety giving double the bandwidth of ICH4's solution out of the box before any extra peripheral I/O is taken into account. That's the price Intel pays for using a consumer positioned bridge on E7505 and even ICH5 doesn't address that with Intel relying on other packages to give higher speed PCI to their solutions. Not a problem given the spec of any E7505 board but something worth noting. Another win over ICH4 is the provision for a triple channel IDE controller on CSB6 for support of 6 IDE devices. Lastly the bridge does some other high-end work with monitor support for LCD monitoring and key/touchpad support for security. So a loss with no USB2.0 support but a win in the PCI and IDE sides of things compared to ICH4.
ServerWorks doesn't explicitly spec their PCI-X segment bridge, CIOB-X2, with GCWS but there is no shipping GCWS solution that doesn't use it. It's identical in spec to 82870P2 with a pair of 64-bit PCI-X segments that can run up to 133MHz for a full 1GB/sec of bandwidth available to each segment. It also fully supports PCI2.2 like 82870P2 so no worries there.
ServerWorks use a dedicated package for the AGP bus interface, called CIOB-G2. The chip sits external to the CMIC northbridge but has a full data path to it for full speed 533MHz 2.1GB/sec AGP8X support. This gives them some flexibility in product arrangement since they were able to pair a CMIC bridge with CIOB-G (AGP4X AGP bridge) until their G2 bridge was ready meaning no upgrade work needed doing to the CMIC bridge and an easy phase in of the G2 bridge when it was done. So more flexibility in that regard but more cost and complexity in terms of silicon and number of chips needed.
The 3rd piece of silicon in a GCWS solution and the 4th overall if you count the PCI-X bridge is CIOB-E, the Ethernet I/O controller. CIOB-E is special in that it sports a pair of full duplex Gigabit Ethernet MAC's on one bridge along with the PHY physical front ends. It's also not quite clear if it sits on the PCI-X bus of a CIOB-X2 or if it provides one of its own on the physical package, but either way you get full duplex operating on both MAC's. So a massive win over E7505 which doesn't spec GigE in this range of solution.
Finally, like E7505, the CMIC-LE does ECC and active detection/correction of memory errors letting the system carry on even in the event of a full memory bank failure. ServerWorks do the same with 8-bit d/c over two controllers and 4-bit over one and extensive reporting back to the host OS via the bridge and NMI and MCE on the CPU.
So what about boards based on these chipsets?
|
