ASUS Unveils World's First PCI-Express 3.0 Motherboard for AMD Processors

[btarunr]

ASUS did the unthinkable yet simple, by innovating the first AMD platform motherboard that features PCI-Express 3.0 x16 slots, the Sabertooth 990FX/GEN3.0 R2.0. Long naming aside, the board provides you a couple of gen 3.0 slots by using PLX-made 48-lane PCI-Express Gen 3.0 bridge chip.

While the board features four PCI-Express x16 slots, only two similarly-colored slots can be used at a time, of which two are PCI-Express 2.0 x16, wired to the 990FX northbridge, and two slots being x16/NC or x8/x8-capable, being wired to a PLX 48-lane PCI-Express Gen 3.0 switch, which in turn takes two PCI-Express 2.0 x16 links from the northbridge.



Apart from this unique feature the socket AM3+ Sabertooth 990FX/GEN3.0 R2.0 features AMD SB950 southbridge, eight SATA 6 Gb/s ports, two eSATA 6 Gb/s, 8-channel HD audio, six USB 3.0 ports, and a zesty ASUS-exclusive feature-set. The new board could be released to the market very soon.

View at TechPowerUp Main Site

 

[Prima.Vera]

To be honest I didn't know that AMD can't do PCIex 3.0...Or this is what I am understanding from article..

 

[btarunr]

My explanation of the PCIe clusterfoo on this board.

 

[seronx]

I'll just wait for the new LGA platforms next year from AMD with native PCI-E 3.0.

 

[Maban]

According to the manual, the first and third PCIe are 3.0 x16, the third switches to x8/x8 with the fourth, and the second is 2.0 x16. It may not exactly be as Tarun diagrammed but it's still some kind of voodoo. I think they just used a single 2.0 x16 link from northbridge to the switch, and the rest is just marketing.

 

[Sasqui]

My explanation of the PCIe clusterfoo on this board.

http://img.techpowerup.org/130109/bta234dsf.jpg

Is that a robot named 990FX or Cluterfoo?

 

[btarunr]

According to the manual, the first and third PCIe are 3.0 x16, the third switches to x8/x8 with the fourth, and the second is 2.0 x16. It may not exactly be as Tarun diagrammed but it's still some kind of voodoo. I think they just used a single 2.0 x16 link from northbridge to the switch, and the rest is just marketing.

http://content.screencast.com/users...1b-4d5c-8e2e-4117004a2025/2013-01-09_1223.png

In that case it's a supermassive marketing fail. Just one 2.0 x16 link between the NB and x48 bridge.

 

[seronx]

In that case it's a supermassive marketing fail. Just one 2.0 x16 link between the NB and x48 bridge.

It is the same for Intel, supermassive marketing fail. How dare ASUS use Intel tech on AMD boards.

Sandy Bridge, Ivy Bridge = 1 x PCI-E 2.0 x16 lane to a 48-lane PCI-E 2.0/3.0 switch.

 

[Maban]

It is the same for Intel, supermassive marketing fail. How dare ASUS use Intel tech on AMD boards.

Sandy Bridge, Ivy Bridge = 1 x PCI-E 2.0 x16 lane to a 48-lane PCI-E 2.0/3.0 switch.

Ivy Bridge is native 3.0, so 3.0 x16 to 48 lane switch.

 

[seronx]

Ivy Bridge is native 3.0, so 3.0 x16 to 48 lane switch.

Actually, if you read into the fine print the PCI-E lanes in Ivy Bridge are 2.0 lanes.

There is a reason why Ivy Bridge is only 200 MB/s faster than 990FX with a 680 GTX in bandwidth.

 

[Maban]

Actually, if you read into the fine print the PCI-E lanes in Ivy Bridge are 2.0 lanes.

Actually if you read the datasheet, any review, any Z77 motherboard manual or spec sheet, any news site, and any Internet site anywhere with any info on Ivy Bridge, yes, they are indeed 3.0.

 

[seronx]

Actually if you read the datasheet, any review, any Z77 motherboard manual or spec sheet, any news site, and any Internet site anywhere with any info on Ivy Bridge, yes, they are indeed 3.0.

Nope.

The IC on the motherboard is PCI-E 3.0 but Ivy Bridge is indeed PCI-E 2.0 connected to PCI-E 3.0. How you can tell actual native PCI-E 3.0 from fake PCI-E 3.0, PLX Switches.

 

[Maban]

Nope.
http://img442.imageshack.us/img442/3081/hardwareheaven.jpg

The IC on the motherboard is PCI-E 3.0 but Ivy Bridge is indeed PCI-E 2.0 connected to PCI-E 3.0. How you can tell actual native PCI-E 3.0 from fake PCI-E 3.0, PLX Switches.

That chart doesn't mean anything in the context of this discussion. That simply shows that the Revodrive 3 X2 they were using for that specific test maxes out at about 1640MB/s with the parameters they tested it with. That card is PCIe 2.0. It is in no way going to show any dramatic difference when connected to a PCIe 3.0 host.

 

[seronx]

That chart doesn't mean anything in the context of this discussion. That simply shows that the Revodrive 3 X2 they were using for that specific test maxes out at about 1640MB/s with the parameters they tested it with. That card is PCIe 2.0. It is in no way going to show any dramatic difference when connected to a PCIe 3.0 host.

That is a 680 GTX. You have the right to believe anything you want...though. It doesn't mean sound will move faster than light anytime soon.

 

[Maban]

That is a 680 GTX. You have the right to believe anything you want...though. It doesn't mean sound will move faster than light anytime soon.

No it's not. If you go back one page to the Test System and Methodology, it quite clearly states that a Revodrive 3 X2 was used for that test.

 

[cadaveca]

That is a 680 GTX. You have the right to believe anything you want...though. It doesn't mean sound will move faster than light anytime soon.

Intel SAYS:

BTW, it's not just the switches, SLI support in BIOS also plays a role in PCIe 3.0 support. Boards without SLI, all support PCIe 3.0; P67, H67, Z68, Z75, H77, Z77 all inclusive.

 

[Disparia]

Asus, this is just silly and wasteful. Go sit in the corner and think about what you've done.

 

[neliz]

BTW, it's not just the switches, SLI support in BIOS also plays a role in PCIe 3.0 support. Boards without SLI, all support PCIe 3.0; P67, H67, Z68, Z75, H77, Z77 all inclusive.

Actually, SLI doesn't have too much to do with this.

If there are no PCIe switches between the CPU and the card, the card will run at it's maximum speed, in that case GEN3.

I think you're referring to the fact that SLI NEEDS switches on 1155 to operate (it requires x8 traffic, for both GEN2 and GEN3), otherwise you'd be stuck with a fixed configuration 8x8 board. The switches are there to redirect the traffic to the first card in case the clockgen in the PCH doesn't detect a card in the second slot.

 

[cadaveca]

Actually, SLI doesn't have too much to do with this.

If there are no PCIe switches between the CPU and the card, the card will run at it's maximum speed, in that case GEN3.

I think you're referring to the fact that SLI NEEDS switches on 1155 to operate, otherwise you'd be stuck with a fixed configuration 8x8. The switches are there to redirect the traffic to the first card in case the clockgen in the PCH doesn't detect a card in the second slot.

Seems to be some Crossfire-only boards that run 3.0 on 2.0 bridges(Biostar). WTF is going on or why, whatever, I am not sure.

It's like when adding IVB support to BIOS, something weird happened.

I almost want to say that AMD allows PCIe 3.0 for VGA on PCIe 2.0 bus, regardless of bridges or what have you. If it's just encoding or whatever, I don't know. X79 is why I want to say that..you get PCIe 3.0 with AMD on any X79, NVidia on X79..not so much.

Don't forget x16/x4 boards, too.(support Crossfire, but not SLi, no bridges)

Not MSi products, but as you know, I've used very few MSi boards.

Thanks very much for the input though, when it comes to MSi, of course your word is king.

 

[neliz]

Seems to be some Crossfire-only boards that run 3.0 on 2.0 bridges(Biostar). WTF is going on or why, whatever, I am not sure.

It's like when adding IVB support to BIOS, something weird happened.

I almost want to say that AMD allows PCIe 3.0 for VGA on PCIe 2.0 bus, regardless of bridges or what have you. If it's just encoding or whatever, I don't know.

Are they running Full x16? in case of Gen2 switches, the card can run GEN3 x8 (bypassing the switches) or GEN2 x16. In that case there's a minimal difference in latency and bandwidth that would prefer GEN3 x8 over GEN2 x16 when initializing.

Thanks very much for the input though, when it comes to MSI, of course your word is king.

I can't bend Intel's rules

Don't forget x16/x4 boards, too.(support Crossfire, but not SLi, no bridges)

In this case the first slot is routed straight to the CPU and it will always init in x16, the x4 traffic will come from the PCH.

 

[cadaveca]

I can't bend Intel's rules

I know, but as you know already, this has been something that has played out very weird on P67/Z68.

None of you OEMs really agree as to what's what, so I can only hold you to MSi products.

 

[neliz]

I know, but as you know already, this has been something that has played out very weird on P67/Z68.

the straight facts became too cluttered with marketing FUD with everyone falling over another to support "GEN3".

I know, but as you know already, this has been something that has played out very weird on P67/Z68.

None of you OEMs really agree as to what's what

Just some brands wanted you to believe that x8 GEN3 is the same as X16 GEN3 (saves a lot of money on the PCI Express switches!)
That's why I said, Intel is leading in this. Follow Intel's guidelines and you're good, skimp on them and you're bad. I just wish there was a mandatory certification for GEN3

 

[cadaveca]

the straight facts became too cluttered with marketing FUD with everyone falling over another to support "GEN3".
Just some brands wanted you to believe that x8 GEN3 is the same as X16 GEN3 (saves a lot of money on the PCI Express switches!)

Ah, but otherwise, I got this PCIe 3.0 thing down right?

 

[neliz]

Ah, but otherwise, I got this PCIe 3.0 thing down right?

Yes.

RE: The X79 part, It has everything to do with time spent validating and certificating, but after a while it was fixed: http://www.techpowerup.com/168027/N...g-Patch-for-Sandy-Bridge-E-HEDT-Platform.html

 

[cadaveca]

We've had a few users now wondering about PCIe 3.0 on P67/Z68, and sometimes when using SB CPUs. This has not helped with the confusion.


And yes, a real standard would be great.