dashbarron

Distinguished
Sep 9, 2007
187
0
18,680
If X38 northbridge won't directly support Full x16 support ... and as the new article on the front-pagereviewing the X38 chipset says there isn't full bandwidth support for SLI at the present moment, then what are the Full X16 boards I see on the market?

Some offer x8 x8, and other times I see ones listing themselves as being "Full x16" and it at least assumes the roll of a full x16 bandwidth support for both lanes.

This is quite contradictory and confusing, can someone explain this to me? Are the boards listed now actually supportive of Full SLI x16 support?
 

cyph0r

Distinguished
Mar 9, 2006
80
0
18,630
The x38 has 16x-16x dual pci-express on the northbridge.

also as of now thex38 doesnot support SLI, it only supports Crossfire.
 

dashbarron

Distinguished
Sep 9, 2007
187
0
18,680
Ok :)

Now what about the motherboards we see now, that offer "Full x16 bandwidth, what is with those?

Stats on the Penryn don't seem impressive. Seems I should get a 3.0GHZ Duo Core, or spend a lot of money and get the top notch Q6850
 

cyph0r

Distinguished
Mar 9, 2006
80
0
18,630
from what I understood with the AMD models yes they have 16x on both slots BUT its a piggy back link on the data.

for example here is a crappy little diagram


CPU----16x data lane-----NB-------16x data lane-----SB----
\ \
16x LANE 16x Lane


So even though the slots are truly runnning at 16x unfortunatly the return lane form the NB getting the data to and from the CPU is still only capable of handling 16x worth of data and not 32x. does this make sence? That is why we didnt see any performance boost on alot of the amd motherboards that came out with this "True dual 16x" Tech.

since the release of C2D I stoped keeping up with amd baised and nvidia chipsets. here is why I stoped looking at nvidia motherboards. Nvidia motheboards are bottlenecked as far as harddrive transfer bandwith goes. THG proved this a while back when they hooked up 4xraptors up to a nvidia chipset and then to a intel chipset. on the nvidia the speed capped at 120mbps give or take while the intel chipset was able to hande up to 600+mbps. I have confirmed this my self for I own 4x150 raptors.

Now suposedly this design handles both PCI-ex on the NB and has more than enough lanes between the cpu and the NB to handle the data.
 

dashbarron

Distinguished
Sep 9, 2007
187
0
18,680
"Now suposedly this design handles both PCI-ex on the NB and has more than enough lanes between the cpu and the NB to handle the data."

By this design, you're referring to X38?

So what you're saying is, it's pretty much a waste of money to spend extra money buying a motherboard now that supports "Full X16" because the north bridge or CPU still can't handle the full bandwidth lanes, correct?
 

cyph0r

Distinguished
Mar 9, 2006
80
0
18,630

hehe what part is confusing?
ill try to explain better
 

4745454b

Titan
Moderator


Are you sure about your chart? AMD systems use a bi directional hyper transport bus to take data to and from the NB. While I haven't taken the time to do the math, I would think it would be faster then a single PCIe 16x lane. When I get back from work, I'll do the math to double check.

Edit: Ok, I'm back from work. I'm using this link to find the speed of the HT bus.

http://en.wikipedia.org/wiki/HyperTransport

HyperTransport supports an auto-negotiated bit width, based on two 2-bit lines to 32-bit lines. The full-sized, full-speed, 32-bit interconnect in each direction has a transfer rate of 20.8 GByte/s (2.6 GHz * (32 bits / 8)), aggregating 41.6 GB/s bandwidth in two directions, making it much faster than many existing standards.

Ok, break out your calculator. 2600 * 4 * 2 20.8GB/s (the 4 is the result of the 32bits / 8, while the 2 comes from the fact that HT is like DDR. Read the paragraph above the one I quoted for that.) The problem with this is that many boards are based off the slower HT bus that runs at 1GHz. If you do the math for that, you get 1000 * 4 * 2 =8GB/s. So how fast is two PCIe 16x lanes?

If my memory serves, a single PCIe lane is ~250MBs. 16 of these together is (250MB * 16) 4GB/s, so two cards should be able to send (at a maximum) 8GB/s. Notice that this is the exact speed of the most commonly found HT. As some like to point out however, we don't normally use the entire PCIe bandwidth. We are just now being able to max out the bandwidth available with AGP8x, and not in all applications. Don't forget that HT is bidirectional. This means even if the CPU is receiving 8GB/s from the video cards, it could still be sending data to the NB.

This isn't a problem with HT. It might be an issue with the slower FSB, perhaps someone else would like to check that.
 

drysocks

Distinguished
Sep 2, 2007
178
0
18,680
Cyph0r:


 

ira176

Distinguished
Mar 19, 2006
240
0
18,680
I have an MSI board with an AMD 580X Chipset. It offers full X16 signaling to both physical X16 PCIE slots. Of course it's only Crossfire compatible.