Today Nvidia releases the nForce 4 chipset, and like with the VIA K8T890 it is a rare bird. So rare in fact that unlike with the K8T890 we do not have a board to test. That puts us in the position of posting up an article without benchmarks which is not what we like to to, and to add fuel to the fire one site has already broken NDA. Ok, so AMD PCI Express chipset launches haven't been what we had hoped for, but we are still holding out hope that November will bring actual boards you can buy.
That aside we begin to poke around at the wonders found on the Nvidia press ftp, and will bring our notes from the nForce 4 conference call, and our look at the nForce 4 Ultra and SLI demo we had at Nvidia's Austin offices.
Nvidia provides plenty of shiny images of the nForce 4, and the above three are of the budget nForce 4, the Ultra enthusiast version, and the high end gamer SLI version. These are of course the polished pr images, and you won't find the fancy logo on the chip on the board when you pull the heatsink off.
So we see Nvidia is rebranding their chipsets, and moving from the 150 and 250 branding to adding Ultra and SLI to describe, and differentiate their chipsets. So what are the basic features of the nForce 4, and what sets each chipset apart? For this we turn to the oddly colored nForce 4 family of MCPs feature chart.
So here we see a direct comparison of features for each nForce 4 chipset, and we also see the changes from the nForce 3 250GB. Of course you now have PCI Express, SLI, SATA 3GB, an enhanced firewall, and a revamped performance utility.
The plain nForce 4 chipset gets the least of these features, and is targeted at boards costing from $50 to $80. It comes with 20 fixed PCI Express lanes, and does not have SLI support. It also does not have the 3GB SATA support, and it does not have the high end firewall features. You also have Nvidia's advanced RAID, and access to their system utility.
When you step up to the nForce 4 Ultra Nvidia expects you to be paying $100 to $150 a board. For your bucks you retain the fixed 20 lanes of PCI Express without SLI support. You do gain the 3GB SATA, and you do have the advanced networking features.
Stepping up to nForce 4 SLI gives you all the features of the nForce 4 and nForce 4 Ultra, and you get the coveted SLI feature. If you are shooting for this you are a gamer, but DCC boxes with SLI Quadro's will come to market. Nvidia would not comment on dual nForce 4 chips on one board, nor whether 2P Opteron boards are coming. SLI steps you up for $200 or more for your board purchase so it won't be the comparatively cheap upgrade I enjoyed back with 3dfx and my two Pure 3D 2s.
So what does a nForce 4 board look like? Here are a few pictures.
This is the Nvidia money shot of their nForce 4 demo board. You are probably familiar with the look of the nForce 4 demo boards already.
Now here is a shot of the MSI nForce 4 SLI board that comes with two 8X PCI Express slots. You can see a card between the slots, and this will need to be flipped up or down depending upon you wanting to be in SLI mode, or if you want to be driving 4 monitors. Now while I've recently switched to dual monitors, I think I'd be going SLI all the way. Also of note with this board is the lack of 1X PCI Express slots. This doesn't bode well for cards showing up to support it any time soon.
Lastly we have our picture of the nForce 4 demo board we took during a demonstration over at Nvidia.
Now we have gone over the main features of the nForce 4, its different flavors, and a look at some boards. Now we will throw a barrage of screenshots courtesy of the friendly Nvidia FTP at you. Buckle up.
Nvidia's firewall now includes Active Armor, and shifts almost all functions over to to hardware taking a load off of the CPU. They have plenty of diagrams to demonstrate how it works. Above you can see packets streaming in over the network. They are parsed, and sent to the firewall where bad packets are dumped, and good ones move onto the placer. Nvidia says this process has no load on your CPU. They then move on to the application. Unrecognized packets to place some load on the CPU as they are policy checked. If they are recognized as bad they are rejected, and good packets again move on to the application. Here some of the load falls on the CPU which Nvidia estimates at 10%.
With this diagram Nvidia illustrates the same processes, but when the firewall falls mostly on the CPU. Here you can see CPU usage is much higher.
Here is the above two examples simplified with both scenarios.
Here we have an example of processing packets without ActiveArmor. You can see packets have to travel around before they get to the application.
With active armor intercepting packets it makes the process of packet sorting much less painful, and you gain in saved CPU cycles as less burden falls on the processor.
So now we have seen how Nvidia handles the packets. Now lets look at the web based interface for the firewall.
The main page of the Network Access Manager links you to all of the settings including basic configuration of the firewall and ethernet, logs, and administration. There is an expandable menu on the left, and the image links in the main body of the page.
This firewall configuration screen shows available protocols, and allows for enabling or denying access. You can also add rules in, or purge the table.
This screen shows the firewall wizard that allows you to set up for game servers, telnet, SSH, FTP, DHCP, chat programs, e-mail, file sharing and more. If you can run it, then you can set it up here.
Here the wizard asks if you want to use an FTP client or server on your system.
Here is the finish screen for the FTP setup, and notice it warns that TCP/UDP, IP, DNS, and ARP had been enabled.
If you are a gamer you certainly will want to play over the network or the internet.
Here the example game is Doom 3, and you are allowed to enter a port range for the game to use.
The monitoring page takes log data into a scrollable box. You can see each log entry tagged with an icon indicating if there is a warning about the traffic.
This packet page shows data on how many packets coming in, and going out showing how many were successful, and how many were blocked.
This graph shows how many UDP packets where allowed or denied inbound or outbound.
Lastly here is a look at the application access control panel. You can allow local and remote access password protected, and access can be designated by IP.
Now we move forward to a brief look at the new system utility now called nTune. We will look at it in more detail at a later date, but here it is in brief.
It is a multifunction utility that includes plenty of overclocking functions, as well as benchmarking features. You also have system monitoring, and bios update functions.
The main screen links you to benchmarking and system tuning, performance troubleshooting, profile management, online support, system monitoring, and the bios update feature. Yes you can tune your system, check to see if dual channel isn't enabled, or memory isn't set to the highest speed, and you can save your profiles.
Here is a look at the performance section where you can benchmark the system, and you can tune for performance based on what you want to achieve with your system.
Now on to SLI which is the new feature generating the most buzz for the nForce 4. Essentially like with the 3dfx you take two video cards, and each shares half of the load.
Here Nvidia has a graphic that illustrates setting up SLI with two video cards. They need to be the same model video card GPU, but not necessarily the same brand. There will be a connector on the top of the cards in the same spot, and also the card to flip to move from SLI mode to quad monitor mode.
We will take a closer look at SLI when we have an opportunity to test it. Now we move onto the RAID features of the nForce 4 that have been greatly enhanced.
First off we have a look at the reference board. When a drive in your RAID array fails a monitoring utility will indicate which drive has failed by marking it. In this case it is red, and Nvidia has a picture of their reference board. Board manufacturers should substitute the actual model board with their layout when implementing this feature.
NCQ is a hot buzzword in hard drives, and the nForce 4 supports it. As you can see from this diagram it allows for the data head of the drive to move from track to track to grab the next bit of data instead of having to rotate around and not jumping tracks of data. NCQ should improve performance for drives, but the jury is still out on the overall benefit, and in what sort of situations it is utilized in.
Here is a shot at the typical RAID controller you will find onboard motherboards today. You will find a single storage controller integrated into the VIA Southbridge as an example.
With the nForce 4 Nvidia has integrated two storage controllers. The benefits of this are that twice as much bandwidth is available, and with 3GB SATA support this becomes important.
On the left we have RAID 1 or mirroring that has two drives with the exact same data so if one drive fails the other takes over. On the right is RAID 0 where data is striped across each drive increasing performance, but decreasing the fault tolerance.
Lastly we have RAID 0+1 where you combine both features for mirroring and striping. You get data backup, and we get the advantages of striping that increases performance. The downside is at least 4 drives are needed.
Lastly nvRAID supports RAID across SATA and PATA drives. So if you have an old PATA drive, and you want to add a SATA drive you can use both in a RAID array.
Now we have seen what type of RAID the nForce 4 chipset supports. Now there are some new features Nvidia brings to the table. Amongst these are the ability to morph from one type of RAID to another. You can do the following actions.
Morph from RAID 0 to RAID 0 adding a disk to the stripe set
Morph from RAID 0 with one disk to RAID 1 with two disks
Morph from RAID 0 to RAID 0+1
Morph from RAID 1 to RAID 0
Morph from RAID 1 to RAID 0+1
Morph from RAID 0+1 to RAID 0
Morph from RAID 0+1 to RAID 0+1
So as you can see Nvidia has some powerful morphing features that make RAID more useful, and manageable. In fact you can even set disks to a spare for these RAID options. We could go on for page after page to list them all.
Also being a RAID user myself I asked Nvidia if they would support moving from another RAID chip to theirs without having to reinstall completely. The answer is no, but I urged them to look at providing that feature in the future. What we did learn was that you can upgrade to the nForce 4 RAID from the nForce 3 RAID and you will not have to reinstall from scratch. You can look at the following pages for many screenshots showing these various configurations looking through the nvRAID utility.
For now we have a first look at the nForce 4 features. While it isn't as exciting without a board to test we do hope to have one soon so we can compare against the K8T890 for performance. So far features seem to be the strong point of the nForce 4, but we will need to test it before we can conclude anything. Later this month look to tests from us with the nForce 4, K8T890, and the ATI Athlon 64 chipsets with SiS added into the mix in November.
Athlon 64 FX53 Review Athlon 64 2800+ Review Socket 939: Athlon 64 3800+ and Athlon 64 FX53 Review Prescott Survival Kit Review Dual Core 90nm Opteron First Glimpse AMD's Dual Core 90nm Opteron Demonstration Dissected Athlon 64 3000+ Socket 939 90nm Review Athlon 64 4000+ and Athlon 64 FX55 Review