Slashdot Mirror
NVIDIA nForce 4 SLI Intel Edition Launched
Spinnerbait writes "NVIDIA took the wraps off their nForce 4 SLI chipset platform for Intel
Processors today and
there's a full review and showcase with benchmarks up at HotHardware.
As with NVIDIA's AMD version of this chipset, motherboards based on the
technology will support dual PCI Express graphics cards for load sharing in 3D
Gaming applications. What's perhaps even more interesting is how
the new NVIDIA memory controller actually allows the platform to out-pace
Intel's own i925XE in virtually all of the benchmarks."
Here
"I'd rather be a lightning rod than a seismometer." -Ken Kesey
I remember reading on the Inquirer, that on a one on one comparison, the nForce Intel boards weren't able to keep up to the AMD ones, on more than just a processor basis. Was a few weeks ago though, so possible could have been fixed, i.e. driver probs
The one thing the Intel version has over the AMD version of this chipset is RAID 5 support. A RAID 5 controller card by itself is over 100 bucks. Dammit this is going to make me want to turn over to the dark side.
I just double checked on Intel's website, and the best I could find was 8x/8x (3 x8 and 1 x4 PCI express slots (28 lanes total)) And with that it is not possible to have multiple x16 slots (Heck, it's impossible to have 1) (It's possible I missed a better one. I was looking in the server section.)
The main reason that Tyan can do that is because of AMD's superior Hypertransport-based bus design in Opterons, over the shared bus favored by Intel. It's also the reason why Opteron scales a lot better than Xeon.
The other reason Tyan can do that is that Nvidia realized how easy it would be to make very slightly different chipsets that facilitated that. Basically they are just Nforce 4 chipsets, that can operate in parallel, giving 40 Pci express lanes (2-way) or 80 PCI express lanes for a 4-way Opteron. (Note a maximum of 4 x16s, as the other 16 can only be a max of x4, due to the 20 lanes per nforce4)
You can't do x16/x16 with any Intel Processor, as of now. (Though having seen how little x16/x4 or x16/x2 hurts benchmarks (vs standard x8/x8) I'm not convinced it's a big deal at all.)
How about we take an Intel CPU at 3.73 GHz and compare it to an AMD CPU advertised to perform around that same clock speed? That gives us the 3800+, the 4000+, and the FX-55. Traditionally, AMD's estimation has been 5 to 10 percent high in terms of what Intel processors they can match (across the board, as opposed to gaming where the estimates are about right), so we can assume:
- The 3800+ performs at 3.42-3.61 GHz which is too low.
- The 4000+ should perform at about 3.6-3.8 GHz, which is about right.
- The FX-55 is undoubtedly faster than the 4000+, owing to its 2.6 GHz actual clock speed compared to the 4000+ at 2.4 GHz. This is a 1/12 increase, so we can expect the FX-55 to perform at 3.9-4.2 GHz... which is significantly higher than the 3.73 we're targeting.
So the FX-55 is too fast, the 3800+ is too slow, and the 4000+ is about right. An argument might be advanced for the FX-53, but I think those have been discontinued.
Where exactly is the problem? I mean, if you RTFA, they say outright that if all you care about is gaming the AMD outperforms the Intel by around 10%, which gives us that 4000+ they advertise... but if you need to do other things, the Intel outperforms the AMD slightly. Which is roughly what you should expect, given the Intel and AMD core competencies. If the 4000+ specs out in the dead center of the range I estimated, it should come out at 3.7 -- which is just over 99% of the Intel clock speed, and should result in less than a 1% margin of loss. This is about what we see in the results.
In other words, nobody paid me squat, and my own experience with processor comparisons predicts pretty much the exact results we saw. So any claim of "bias" seems ill-considered.
Microsoft cheerleader, blue flag waving, you got a problem with that?
Check out the Gigabyte GA-K8NXP-SLI board which has SATA-150 RAID-5 via an extra chip.
Damien
If you haven't taken a gander at it yet, you may want to take a look at OpenRT and projects using OpenRT such as Quake3 Raytraced. Also take a look at the hardware architecture as well.
.. with traffic, crowds, etc.
Ray-tracing presents a much more detailed rendering of a scene, but was always considerably slower than rasterization. If hardware-accelerated ray-tracing architecture grows in the market, you may see your skyline beautifully rendered in real-time
There has been quite a lengthy discussion on LKML on that topic. IIRC, there are some issues with NCQ and sound support, but it's an interesting thread well worth reading, if you're into nForce4-thingies that is.