backtop

Print E-mail del.icio.us 26 comment(s) - last by Visual.. on Mar 24 at 9:19 AM
AMD reveals next-generation desktop chipset details

AMD's latest roadmap details much of its next-generation chipset architecture, codenamed RD890 and RD880.  The pinnacle feature of this new architecture is the inclusion of an input/output memory management unit, or IOMMU.

An IOMMU is a device that supports mapping memory addresses. Since an IOMMU holds many benefits in server applications, it can sometimes be found in high end server hardware, but at the moment no desktop chipsets support IOMMU.  

The IOMMU connects a DMA-capable I/O bus to the primary storage memory of a PC. The memory management unit maps virtual addresses to physical addresses, similar to what a CPU memory management performs.  Although presently there are few desktop applications that can benefit from IOMMU setups, the technology can greatly benefit virtualization applications.

A complete description of AMD's IOMMU is available on AMD's website.

AMD RD890 will offer few other features over the existing AMD 790 chipset, with the exception of a new of a cryptic "DirectX 10 accelerator."

To keep up with the new NVIDIA chipsets, AMD will launch an intermediate step between RX880 and the existing RX780 (AMD 770) chipset, dubbed RX780H.  This new chipset is pin-compatible with AMD 770 and AMD 790X motherboards, and is actually the first to feature the DirectX 10 accelerator.  

AMD declined to let DailyTech know if this means integration of a low-end GPU for "Hybrid" Crossfire, or something totally different.  RX780H is scheduled for a Fall 2008 launch.

The company is expected to launch its AMD 790G chipset later this spring, which incorporates the elements of the AMD 790 chipset with a low-end DirectX 10 integrated graphics core.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
Good to see
By eye smite on 3/21/2008 1:02:03 PM , Rating: 3
None of this may be earth shattering, but it's nice to see the AMD steamroller finally moving at a steady pace with it all. Here's to more tidbits here and there from AMD. This will keep intel being innovative to stay ahead which will benefit us all.




RE: Good to see
By Visual on 3/21/08, Rating: 0
RE: Good to see
By FITCamaro on 3/21/2008 1:31:31 PM , Rating: 2
quote:
so for the first time we will have un-gimped 3d gaming in virtual machines


Is this correct?


RE: Good to see
By amanojaku on 3/21/2008 1:52:01 PM , Rating: 5
No. I work at VMware and that's a question that comes up fairly often. Everyone wants to run Linux or MacOS on their laptop and Crysis in a Windows VM.

The problem is latency: the IOMMU must serve as a go-between for memory requests. When you consider how much data is sent to and from the video card and how it's processed you begin to understand why virtualization is unlikely. The IOMMU will allow you to use the video card, but the performance will be terrible.


RE: Good to see
By FITCamaro on 3/21/2008 3:46:07 PM , Rating: 4
Well then get off this site and get your @$$ working so I can do that then!

Just kidding. You guys make a great product. Have used it a few times before.


RE: Good to see
By sheh on 3/21/2008 5:45:14 PM , Rating: 2
What makes this so much different than CPU translation, (where there's much more traffic)? And if this is critical enough, perhaps it could be integrated closer with the memory controller.

Also, aren't graphic cards anyway doing most of their processing locally, hence the minimal impact of going from AGP x4 to 8 or PCIe?


RE: Good to see
By amanojaku on 3/22/2008 7:42:08 PM , Rating: 4
quote:
What makes this so much different than CPU translation, (where there's much more traffic)?

It's not really any different. It's just external AND a conversion, thereby creating latency. Imagine the number of commands required to display an exploding grenade and its smoke-trailed shrapnel!

quote:
Also, aren't graphic cards anyway doing most of their processing locally, hence the minimal impact of going from AGP x4 to 8 or PCIe?

I'm not in engineering so I'm no expert, but here's a few things I've learned:

1) There is no business justification for high end graphics (OpenGL, DirectX, etc...) in a VM. VMware fulfills business needs; home users are not the target market.

2) Most people have it backwards. Why run your games in a VM from your office machine? You should run your office VM from your game machine!

3) Several factors point to a far off date for graphics virtualization. GPUs are stressed as it is; imagine suddenly needing two or three times the graphics power because of some power hungry apps. Then there's the reality that by virtualizing the card you are making its resources shared. Sharing in a CPU or GPU means context switching and its resulting loss in performace. Unless some way can be found to partition a card's resources the processes must be context switched. And we all know how gamers like stuttering displays...


RE: Good to see
By sheh on 3/22/2008 11:28:12 PM , Rating: 2
quote:
It's just external AND a conversion, thereby creating latency. Imagine the number of commands required to display an exploding grenade and its smoke-trailed shrapnel!
Wasn't one of the features of D3D10 virtualization? Of course, I expect the local graphics processing to stay local. Only the external bus communication should be translated. And even with context switching there should be an exclusive mode (at least for fullscreen).

quote:
1) There is no business justification for high end graphics (OpenGL, DirectX, etc...) in a VM. VMware fulfills business needs; home users are not the target market.
Of course there is; running your Windows games while using *nix as your primary OS. :)


RE: Good to see
By Jovec on 3/21/2008 6:33:49 PM , Rating: 2
Something like: Host OS -> on-board video, VM -> direct/dedicated access to video card. How feasible would this be with IOMMU?


RE: Good to see
By Xajel on 3/22/2008 3:36:10 AM , Rating: 2
As you're working in VMWare, I have a question...

when we will be able to switch between OS on the fly ?

I mean, when the hardware will be able to run every OS on Hardware VM, and then we will be able to run several OS's in the same time ( or if we have not enough RAM then hibernate, ofcourse for those guys, adding RAM will be much better and faster )

I always wondered what stop us from making this happened ?

loading every OS in the RAM each with Virtual RAM space, and switching between them on the fly, no need to do hard coding in the OS as every thing will be handled by Hardware, each OS will see this computer as normal computer with normal hardware


RE: Good to see
By winterspan on 3/22/2008 4:40:42 AM , Rating: 2
What about Intel's new x86 extensions aka "virtualization for directed I/O" aka VT-d ? That involves having an IOMMU, right?
Will that allow good GPU use in a VM ?


RE: Good to see
By Visual on 3/24/2008 9:19:14 AM , Rating: 2
even your own product, vmware workstation, following in the steps of parallels desktop, already has some support for direct3d acceleration through api-level translation:
http://www.vmware.com/support/ws55/doc/ws_vidsound...

the IOMMU approach will certainly provide much better compatibility and suppost than your current attempts to fix it at the software api level.

and how can you say its latency will be slower even before you've seen the actual IOMMU implementation?

besides, i didn't think that running crysis does all that much dma transfers during actual gameplay (we're talking about a real graphic card, not onboard video that has its framebuffer in system ram), and i can deal with somewhat slower initial load times...


RE: Good to see
By FITCamaro on 3/21/2008 1:30:06 PM , Rating: 2
One place AMD is definitely ahead of Nvidia in is chipsets. Nvidia is still using 90nm chipsets which cause them to run HOT while AMD is using 55nm chipsets.


RE: Good to see
By DigitalFreak on 3/21/08, Rating: -1
RE: Good to see
By teldar on 3/21/2008 3:08:11 PM , Rating: 3
I believe the comment was a response to the THREAD. Not to the article. That would be the way a thread typically works....


RE: Good to see
By FITCamaro on 3/21/2008 3:44:18 PM , Rating: 2
quote:
You want a cookie or something?


Yes. I would love to have a cookie.


RE: Good to see
By orphen193 on 3/21/2008 4:03:57 PM , Rating: 2
hey I would love a cookie too :D


RE: Good to see
By FITCamaro on 3/21/2008 4:39:36 PM , Rating: 2
No. They're all for me.


RE: Good to see
By StormEffect on 3/21/2008 7:13:37 PM , Rating: 2
You made me LOL.

The cookie is a lie!


so how long before...
By kattanna on 3/21/2008 5:02:14 PM , Rating: 2
the case simply contains the PSU, storage/drives, a bank of expansion cards, and a big black box that connects everything.

the big black box is the all in one CPU/memory..etc combo unit

when you want to upgrade, you simply change out the black box.




RE: so how long before...
By DragonMaster0 on 3/21/2008 9:57:15 PM , Rating: 2
Simple... Do like some old XTs, the motherboard was simply a lot of ISA slots and a PSU connector. The CPU, memory controller and RAM chips were on their own card, the I/O ports another one, etc.


RE: so how long before...
By bigboxes on 3/21/2008 10:10:47 PM , Rating: 6
I am not black, although I have been described as big. :p


AMD should keep news to themselves this early....
By wingless on 3/21/2008 6:32:15 PM , Rating: 2
AMD should not come out with new hardware ideas this early because they will only be copied and 1-upped by Intel. I'm sure we'll hear news that Intel is planning the same technology for desktops in a month. Also this news makes us not want to buy AMD now in anticipation of some nicer tech that will come out in a few months. The new RX780H may be worth waiting for. Also I won't even consider buying anything AMD until the SB700 or 750 is paired with a 790FX chipset. We're all also waiting for B3 or 45nm Phenoms too. AMD won't sell a damn thing until consumers expect NO new revisions out in the short-term.




By sweetsauce on 3/21/2008 6:52:41 PM , Rating: 2
Intel doesn't copy, they decide not to implement things like on-chip controllers until they feel it will be a benefit. Virtualization on desktops is great for us geeks, but the common pc user will never have use for this for ages if ever.


By winterspan on 3/22/2008 4:22:59 PM , Rating: 2
This is by no means an AMD-exclusive idea. Intel
has had plans for isolating virtualized hardware with an IOMMU
for years now. Just like Intel's "Vanderpool" AKA VT-x instructions, they have VT-d instructions coming out in the future which require an IOMMU to make sure that virtualized operating systems can't break through isolation by using DMA commands on the memory pool.
I really doubt Intel came up with this either.. The concept of an IOMMU has most likely been well known by other companies and academics for decades...


By wingless on 3/22/2008 11:32:35 AM , Rating: 2
I can't wait to run F@H in Ubuntu under VMWare with this extra performance boost. Maybe AMD will match or exceed the PPD of Intel Quads running the same software setup. F@H runs much faster under Linux to begin with and any help to speed up virtualization is welcomed with open arms.




"Nowadays, security guys break the Mac every single day. Every single day, they come out with a total exploit, your machine can be taken over totally. I dare anybody to do that once a month on the Windows machine." -- Bill Gates











botimage
Copyright 2010 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki