PDA

View Full Version : AMD 8-series chipset



sink257
16th January 2010, 03:11 AM
Anyone has details about the 8-series chipset, and when it will be releasing?

So far there's only been talk of an improved SB, so the NB is the same as the 7-series one? :confused:

TakeANumber
16th January 2010, 05:17 AM
Here are a couple of pages that may shed some light in NB changes (with the IGP that resides in the NB chip):

http://www.semiaccurate.com/2010/01/08/amd-shows-integrated-dx11-and-mobile-quads/

(Charlie was told it's running Dirt 2 in DX11)

http://downloads.guru3d.com/ATI-Catalyst-10.1-Beta--Windows-7-%7C-Vista-(32%7C64)-download-2455.html

(note the line: "AMD Dual Graphics: Possibility of combining performances of the IGP of the RS880 with the forthcoming entry level GPUs from the Cedar and Park ranges." Both the RS880 chipset and Cedar/Park discrete GPUs are still unreleased)


What both of these articles suggests to me is that the new IGP supports DX11 and you can Hybrid Crossfire it with the lower-end Evergreen GPU's, further bolstering the DX11-compliant IGP assumption/speculation.

It also further suggests that the release will happen very soon, or why else put the next gen Hybrid Crossfire in the January 2010 beta driver? February/March release, maybe?

Game_boy
16th January 2010, 08:02 AM
Try not to think of it terms of 'must have a new NB'. Think what features a new AMD chipset should have that customers would benefit from.

Apart from USB3 (which an add-on board can achieve and even Intel won't have) and the SATA6 it is confirmed to have, I can't say lack of chipset features are holding back the AMD platform.

kalizec
16th January 2010, 10:28 AM
First off, 8-series will be 55nm (7-series was 65nm), so it'll be more power efficient.

Second, the amount of PCIe-lanes in the NB is still unknown. (Afaik).
But it'll have to be at least 42 lanes given the boards I've seen.
If they upped the amount though, my guess is that the new amount will be 50 in the NB. (3x16 + 2 for NIC).

Thirdly SATA 6.0 Gbps is very nice for SSD's, especially considering they must've fixed AHCI. (Which was sometimes problematic in the 6- & 7-series.)

Finally, I hope, but don't know yet, whether they implemented IOMMU. Basically virtualization support for IO.

But I've also got some criticism. Imho they should've ditched the ATA133 and gone with 8x SATA. USB3.0 not being implemented can be blaimed on Intel holding tight onto the specs during development.

rpmrg
16th January 2010, 10:53 AM
Is the new IOMMU only for IO? The docs at AMD say about many other things, like GPU Virtualization.

265586888
16th January 2010, 11:08 AM
Members of the series should include:
RD890,
RS880 (which is 785G, it's 8-series if you consider the original codename ;)),
RS880C (cost-down version, should be 785V if it exist),
RS880D (with x16-x16 CrossFireX, i.e. dual PCI-E 2.0 x16 slots),
RS880M (mobile, M880G), and
RS880MC (mobile cost-down version, should be M880V if exist)

Southbridges*:
SB850
SB820

*They should be the same difference as SB750 and SB710: crappy RAID 5

265586888
16th January 2010, 11:10 AM
Is the new IOMMU only for IO? The docs at AMD say about many other things, like GPU Virtualization.

Consider this will do the same stuff as Intel's VT-d technology.

kalizec
16th January 2010, 11:12 AM
Is the new IOMMU only for IO? The docs at AMD say about many other things, like GPU Virtualization.

From the viewpoint of a software-developer GPU's are also IO. Reason being that all modern IO is communicated using IO-adresses. These addresses are used for anything that's hooked up via PCI, including PCIe.

rpmrg
16th January 2010, 11:24 AM
Thanks for the clarification. I wasnt clear enough, should had stated about disk subsystems.

http://support.amd.com/us/Processor_TechDocs/34434-IOMMU-Rev_1.26_2-11-09.pdf

"
The IOMMU is designed to support three main usage models:
• Direct user process access to a single device like a graphics controller,
• Direct virtual machine guest access to a collection of devices that have been dedicated to that guest, and
• A single non-virtualized OS using the IOMMU to enforce device to system memory access controls. "

265586888
16th January 2010, 11:36 AM
Read this:
click here (http://semiaccurate.com/forums/showpost.php?p=11878&postcount=59)

Something about IOMMU which I've found in old discussions. :)

EDIT: And an article from AMD: http ://developer.amd.com/documentation/articles/pages/892006101.aspx (http://developer.amd.com/documentation/articles/pages/892006101.aspx)

rpmrg
16th January 2010, 11:45 AM
Thanks. It would be great if we had access also to the GPU from VM's. Unfortunately, i cant see that happening anytime soon. pity.

imported_charlie
16th January 2010, 12:12 PM
Evergreen + 785 has the hardware, the software infrastructure needs to catch up. By that, I mean VMWare et al, not ATI drivers. The new parts are 'fully' virtualizable from what I understand, but given that it is the first implementation, I would guess that it will be a little bumpy until Evergreen refresh or Northern Islands + 880.

-Charlie

265586888
16th January 2010, 01:49 PM
You mean the DMA feature on R600/RV770/Evergreen already supports IOMMU? :eek:

rpmrg
17th January 2010, 12:31 AM
Fantastic info. Thanks Charlie.

imported_charlie
17th January 2010, 09:58 AM
You mean the DMA feature on R600/RV770/Evergreen already supports IOMMU? :eek:

No, just Evergreen. I asked about it at the Evergreen launch.

-Charlie

265586888
17th January 2010, 10:05 AM
No, just Evergreen. I asked about it at the Evergreen launch.

-Charlie
Coinciding the launch of SR56x0 chipsets...
Now GPGPU variants (FireStream) should get interesting...

sink257
17th January 2010, 11:21 AM
Thanks for your replies guys, cleared the air on the NB thing a little.

Now about the release date...

265586888
17th January 2010, 11:32 AM
Thanks for your replies guys, cleared the air on the NB thing a little.

Now about the release date...
As the "SB8xx" string appeared in Catalyst 10.2 early beta drivers, you could say RD890+SB850 combo should launch at least according to schedule, April 2010 or even earlier, maybe coinciding the launch of Thuban CPUs in Q2 2010, as this combo promises good overclocking capabilities.

IOMMU already in SR6x0 series (RD890S, RD870S) chipsets and 785G (RS880), which has already been launched, but you'll have to wait for Socket G34 CPUs...

RS880 is already released as 785G.

imported_charlie
17th January 2010, 07:50 PM
I would not expect anything really interesting or usable with respect to virtualization of GPUs until late this year. There is too much that needs to be hammered out and tried, retried, and finally gotten working for things to be, well, happy. Expect the first attempt or two to suck.

If anything happens, I would expect the big bang to be in the cat 10.3, possibly 10.4 drivers. Keep an eye on those, ATI has some good stuff coming.

-Charlie

rpmrg
18th January 2010, 06:12 AM
Good news nevertheless. Now, could you provide some info if Bulldozer could plug into the 890/AM3 socket; :D

265586888
18th January 2010, 07:13 AM
News say that the new processors based on Bulldozer architecture will use AM3r2... Of course it will be compatible to Socket AM3, maybe with some electrical pinout differences...

So you can say it's a "yes"...

sink257
18th January 2010, 07:31 AM
Good news nevertheless. Now, could you provide some info if Bulldozer could plug into the 890/AM3 socket; :D

Given AMDs track record, I don't see why that should not be possible.

However, if that hinders performance, then I'd rather they not have backwards compatibility.

toughluck
18th January 2010, 12:01 PM
Good news nevertheless. Now, could you provide some info if Bulldozer could plug into the 890/AM3 socket; :D

It will depend how mainstream DDR4 is by the time Bulldozer hits the market.

rpmrg
18th January 2010, 12:09 PM
Indeed. But they could always work out a dual memory controller in 2 SKU's for the AM3 crowd and remove the DDR3 support in the AM3r2 platform. Failed speculation but it could work, i guess.

265586888
18th January 2010, 05:26 PM
Changing the memory controller will surely change the pinout, no matter for number of pins and it's layout... But then AMD can always change the socket for backward compatibility, consider Socket AM3 on motherboard is 941-pins, while Socket AM3 on the CPU is 938-pins.

But considering AMD won't follow closely to memory type transitions as Intel, and DDR4 will appear at least in 2012, so there shouldn't be a major change in socket in 2011 for AMD. But you can expect AMD will also support low-power DDR3 though... As for DDR4 transition, unless there's some big changes, you can expect AMD will do the transition a year after Intel does, just like they did with DDR3 support.

WordPress Appliance - Powered by TurnKey Linux