backtop


Print 96 comment(s) - last by Archmaille.. on Dec 14 at 4:24 PM

ASUS L1N64-SLI WS to be the first 4x4 motherboard

DailyTech has obtained a couple of images of an upcoming motherboard for AMD’s 4x4 enthusiast platform. The motherboard is an ASUS  L1N64-SLI WS powered by two NVIDIA nForce 680a MCPs. Two socket-1207 processors are supported with four memory slots—two slots per processor. With two nForce 680a MCPs the ASUS L1N64-SLI WS features 12 SATA 3.0 Gbps ports and one PATA for storage connectivity. There’s also an additional e.SATA port on the back I/O as well.

As far as PCI Exress goes the L1N64-SLI WS features a grand total of four PCI Express x16 slots for plenty of SLI and SLI physics processing power. Due to space limitations the L1N64-SLI WS only has one PCI and PCIe x1 slots.

Networking features of the ASUS L1N64-SLI WS include dual Gigabit Ethernet ports. The onboard Gigabit Ethernet features NVIDIA’s FirstPacket, DualNet, Teaming and TCP/IP offload technologies. Audio is powered by a high definition audio codec with optical and coaxial S/PDIF outputs. It is unknown which codec ASUS has equipped the L1N64-SLI WS with, though it could be Analog Devices or Realtek. The board is not legacy free and still has PS/2 and parallel ports.

Due to the complex design of AMD’s 4x4 platform, the ASUS L1N64-SLI WS uses an eATX form factor which will not fit in smaller cases. Since dual processors require a little extra power, ASUS has equipped the L1N64-SLI WS with an 8-pin EPS12v and Molex power connectors.

U.S. distributors claim the ASUS L1N64-SLI WS will have an MSRP of $480 without bundles, but the street price will probably be much less.

Expect AMD’s 4x4 enthusiast platform to arrive later this month with three Athlon FX processors—FX-70, FX-72 and FX-74. It is unknown if the ASUS L1N64-SLI WS will be compatible upcoming AMD Stars processors.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

If you're going to do this...
By Anonymous Freak on 11/20/2006 5:45:09 PM , Rating: 2
...you might as well go for a dual-socket Xeon board. Probably cost about the same, and you can throw two quad-core Xeon 5300 series processors in there. Significantly less power draw, and at least equal performance.




RE: If you're going to do this...
By Thorburn on 11/20/2006 5:47:38 PM , Rating: 2
The FB-DIMM's wouldn't be cheap though I fear.


RE: If you're going to do this...
By VooDooAddict on 11/20/2006 6:16:39 PM , Rating: 1
The problem with the Xeon route is that afordable motherboards aren't there for gaming. Many can't handle a single PCIe x16 ... let alone 2 (or 4?!)

FB-DIMM does command a premium, too. Though the motherboard premiums for something that can truely support x16 seem to be the killer.

Otherwise I agree ... two quad core Xeons are a very tempting option for a gaming/workstation system. Even at 1.6 or 1.8 Ghz it's still a pretty fast chip in gaming. (Considering you can get two of those chips for the cost of one Core 2 Quad Extreme.)


RE: If you're going to do this...
By Thorburn on 11/20/2006 7:04:26 PM , Rating: 2
If the rumours of an NVIDIA nForce 680i based Xeon chipset are correct then things could certainly get interesting though.
The Xeon really needs something like that as standard DDR2 and 16x PCI-Express would really benefit it in the workstation (and maybe even enthusiast) market.


RE: If you're going to do this...
By Fenixgoon on 11/20/2006 7:33:52 PM , Rating: 4
FB DIMMS are also horridly inefficient. if im not mistaken, some good DDR2 > FB DIMM RAM. FB DIMM has greater theoretical power, but in practice (as per Anandtech's review of the Mac Pro), it's simply not 100% useable.


RE: If you're going to do this...
By JeffDM on 11/21/2006 5:57:16 PM , Rating: 2
The difference is that FB-DIMM scales up better, which is why the workstation chipsets use it. You can put up to 16 of them in a workstation. I think all said, a quad (2x2) Xeon system probably ends up being comparable in cost as a quad (2x2) AMD. I don't understand why this is helping, because it looks like rebranded Opterons + Opteron boards at Opteron pricing, though with less memory capacity and more PCIe slots.


RE: If you're going to do this...
By Nighteye2 on 11/20/2006 7:25:08 PM , Rating: 2
Xeons do not support NUMA, 4x4 does.


RE: If you're going to do this...
By afkrotch on 11/20/2006 7:44:27 PM , Rating: 2
4x4 uses ccNUMA. Which is crap.


RE: If you're going to do this...
By Kiijibari on 11/21/2006 5:56:35 AM , Rating: 2
quote:
4x4 uses ccNUMA. Which is crap.
Sounds like you dont have any clue what you are speaking about ...


RE: If you're going to do this...
By JumpingJack on 11/21/2006 12:06:25 AM , Rating: 3
quote:
Xeons do not support NUMA, 4x4 does.


Interestingly, do you know what NUMA is??? Non-Uniform Memory Access.... in essense, it is forced by virute of AMD's implemenation of HT links and IMC. This does not make NUMA itself superior, as a call from non-local memory must induce at least one hop from one CPU to the other, this is a latency burden step. AMD deservese credit for getting the latency low as possible, but uniform memory access (such as Symmetric Multithreading) does not suffer from latency due to this extra hop.

NUMA is by no means superior, it is a memory access method that is actually burdened by virtue of AMD's architectural decisions. On the single socket desktop, there are not extra CPU nodes between memory and CPU, so 1P is not NUMA --- however, 4x4 being two sockets is NUMA. This is just as crippling as sharing an FSB in a MCM approach, perhaps more so as the physical distance between nodes is much much greater in a 2 socket design.


RE: If you're going to do this...
By drank12quartsstrohsbeer on 11/21/2006 9:46:42 AM , Rating: 1
I don't think you know what you are talking about. Go look at the throughput numbers of existing dual processor opteron boards.

Then come back here and talk about how you actually meant to say something else, or how you last post was misunderstood.


RE: If you're going to do this...
By Nighteye2 on 11/23/2006 10:15:14 PM , Rating: 2
Exactly. A good example of NUMA performance can be seen here: http://www.techreport.com/reviews/2005q2/opteron-x...


By JumpingJack on 11/30/2006 1:27:12 AM , Rating: 2
quote:
Exactly. A good example of NUMA performance can be seen here:


I said nothing about bandwidth, and a sysnthetic BW measurement makes not a representation of realworld performance.

Please link up one showing Woodcrest kicking butt on Opty, and having lower overall 'Sisandra Memory Bandwidth', then explain what you mean.


By JumpingJack on 11/30/2006 1:24:53 AM , Rating: 2
quote:
I don't think you know what you are talking about.


Actually, yes I do. I said nothing about bandwidth, and it works great if you are shuttling large chunks of data combined with a NUMA aware OS. However, you suffer from a fit ignorance and you are mis-interpreting what I said. This is common among people who really know nothing about computing, there is a difference between throughput and latency and how those combine to really produce speed in terms of data rate.

NUMA, non-uniform memory access, by definition means access to any given part of memory will differnt in access time depending on the location of that memory. For example, CPU1 has local connection to MEM1 and CPU2 has local connection to MEM2, if CPU1 cache hit requires data located on MEM2 CPU2 must fetch the data, shuttle to CPU1 via the cHT link --- this will by the fact that it is now across a bus induce extra latency.

IBM measured the affect:
http://www-03.ibm.com/servers/eserver/opteron/pdf/...
See figure 4, where main memory calls split into two discrete values for latency.

Hmmmm, did I not predict that Kentsfield would have better memory metrics:
http://www.extremetech.com/article2/0,1697,2065497...

Hell -- an FX-62 is showing better memory performance on the small stride cache line traces (where latency has the biggest affect on speed through the system). 4x4 is suffering from this....

Jack


By JumpingJack on 11/30/2006 2:02:41 AM , Rating: 2
Oooops here is one more:

quote:
Secondly, Quad FX platform is often slower than the regular Socket AM2 system with a single CPU because of the higher memory subsystem latency. NUMA technology that proved highly efficient in servers turned out to do more harm than good in the desktop space.

http://www.xbitlabs.com/articles/cpu/display/amd-q...


"This is about the Internet.  Everything on the Internet is encrypted. This is not a BlackBerry-only issue. If they can't deal with the Internet, they should shut it off." -- RIM co-CEO Michael Lazaridis

Related Articles
AMD Beyond "Brisbane"
November 14, 2006, 6:31 PM
AMD Q4'06 Dual-Core Roadmap
October 3, 2006, 8:23 AM
Realtek Audio Codecs Not Rendering True EAX
September 26, 2006, 10:45 PM
nForce 590, 570, 550 Announced
May 23, 2006, 4:41 AM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki