Does NVIDIA know something about Bulldozer?

I doubt it. I believe it simply means they no longer will have a chipset that supports AM3+, and therefore their only option to get AMD users to buy an SLI setup is to license SLI on an AMD chipset, whereas AM3 still had the M4N98TD and NF980-G65.
 
The comments are pretty funny though. Props to NVIDIA for not deleting them immediately.
 
Last edited:
AMD_Gamer said:
What do you guys think this move means?
Click to expand...
It means that jen-hsun huang took a chill pill and finally decided to let AMD board makers pay nvidia money to enable SLI compatibility.

I'd guess that nvidia is giving up on its AMD chipset business, as SLI was the only draw on that platform.
 
Tsumi said:
I doubt it. I believe it simply means they no longer will have a chipset that supports AM3+, and therefore their only option to get AMD users to buy an SLI setup is to license SLI on an AMD chipset, whereas AM3 still had the M4N98TD and NF980-G65.
Click to expand...

This.
 
Are you gonna post this in every thread that includes Nvidia or Bulldozer?

No they just wanna make more money. If Kepler has omgwtfpwning sli scaling than they obviously also want people going AMD to get it.
 
LOL, the comments are openly bashing nVidia without even being nasty! :p Not even a single nVidia defendor of note on the entire comments roll, so far!!
 
Yet manufacturers have to pay them a royalty to get their boards certified so whose to gain in the end?
 
AMD’s stature as the preferred gaming CPU fell by the wayside and Intel CPUs have been the gamers’ choice.+ For this reason, we’ve only licensed SLI for motherboards with Intel chipsets.

^this lol what bullshit excuse looool

They worse than ****** politicians
 
are they going to enable it on older chipsets?

there were rumors for years that SLI worked fine on amd boards, but the board makers never enabled it because they didn't want to piss Nvidia off.
 
Glad they had a change of heart. Though, I thought this was confirmed weeks ago. They've been robbing themselves of sales for the past 1.5-2 years. I know plenty of people who went with an AMD GPU because the only way to get SLI on AMD was to use nvidia's re-hashed 780/980 chipsets.
 
TheBlueChanell said:
Glad they had a change of heart. Though, I thought this was confirmed weeks ago. They've been robbing themselves of sales for the past 1.5-2 years. I know plenty of people who went with an AMD GPU because the only way to get SLI on AMD was to use nvidia's re-hashed 780/980 chipsets.
Click to expand...

This. I changed my socket just to have the choice of what I want, since AMD has great CFX boards but their SLI boards(barely in stock) are pure garbage from what I read. I think this helps both companies.
 
This is GREAT news for me! I'm sticking with Nvidia for another Gen or 2, because I have a 3-D Vision projector and it works great. I know most people don't care about PhysX or 3-D gaiming but I do, there is a couple great games that make use of it. I do want to get a new Nvidia card when the time is right and rock Metro 2033 with nicer eye candy by using my current 260GTX for Physx and a 500 series as main. ;) Then SLI that card when it shows age or whatever.

Now I can get a newer MOBO with SLI/CFX support in a few months with a BD CPU, and have a ton of versatility.

If this deal works out well, you just might see Nvidia licensing Physx on down the road with AMD cards...which in the end will make PC games more appealing than console, even though they always been more appealing to me from a technological standpoint.
 
I doubt you will ever see nVidia allow PhysX on AMD GPU's. Like I said, this decision was most likely made from a purely business standpoint, in the fact that nVidia won't have any options for SLI motherboards for AM3+. Trust me, if they still had SLI motherboards compatible with bulldozer, you would not see SLI on 9xx chipsets. Whatever nVidia can keep to themselves, they will keep to themselves. They tried that last time with Intel and forcing NF200 chips... didn't work too well. Guess they learned their lesson from Intel and decided to just allow it, instead of trying to force NF200 chips on AMD boards. If they had been successful with Intel and was able to force NF200 chips to allow SLI, then we would see AMD boards requiring NF200 chips for SLI.
 
The last I heard it is actually AMD that chooses not to support CUDA (and by extension Phyx), and not that Nvidia does not allow them too. The reason they have this stance is likely due to the long term implications of pushing CUDA to be a defacto standard that is owned by Nvidia.
 
Perhaps, who knows, but I'm glad NVIDIA finally decided to allow SLI on AMD chipsets, which took them long enough. I just hope Bulldozer does actually perform because I want one.
 
limitedaccess said:
The last I heard it is actually AMD that chooses not to support CUDA (and by extension Phyx), and not that Nvidia does not allow them too. The reason they have this stance is likely due to the long term implications of pushing CUDA to be a defacto standard that is owned by Nvidia.
Click to expand...

Go install an Nvidia card for Physx using an AMD card as primary, see what happens.

xIronCrossx said:
SLI boards(barely in stock) are pure garbage from what I read.
Click to expand...
Yep, all Nvidia chipsets for AMD are garbage.
 
Last edited:
Either way I think this is a good move for nvidia and consumers, given that they're not doing chipsets anymore. This way they make the amount off of partners who include SLI in their motherboards as well as possibly increase sales of GPUs because SLI is possible on both Intel and AMD platforms (and the flexibility is nice for consumers)
 
Lorien said:
Go install an Nvidia card for Physx using an AMD card as primary, see what happens.
Click to expand...

You are describing something different. The implications of AMD cards themselves supporting CUDA and allowing an Nvidia card to run PhysX with an AMD card present is different.

The current landscape might be different, but Nvidia in the past had offered AMD (and even Intel) the ability to license CUDA, and I believe at no or essentially no cost. Again both declined to do so for obvious reasons if you think about it. AMD when asked even cited the possible issue of CUDA not being optimal for their hardware (intentionally or otherwise).
 
limitedaccess said:
You are describing something different. The implications of AMD cards themselves supporting CUDA and allowing an Nvidia card to run PhysX with an AMD card present is different.

The current landscape might be different, but Nvidia in the past had offered AMD (and even Intel) the ability to license CUDA, and I believe at no or essentially no cost. Again both declined to do so for obvious reasons if you think about it. AMD when asked even cited the possible issue of CUDA not being optimal for their hardware (intentionally or otherwise).
Click to expand...

AMD doesn't use CUDA because Nvidia will not let them have access to the source codes so AMD can compile it for themselves. Nvidia wants to compile a CUDA runtime for AMD cards, because they are afraid that AMD will "cheat" somehow. Why would anyone expect a major company to compile a runtime that may run faster on thier competitor's hardware? Intel compiler, anyone?
 
If you read what I've said, I have that AMD does not support CUDA, and by extension PhysX for valid reasons. Neither do I feel AMD should support it, and yes it is in AMDs best long term interests to pursue the other alternatives out there and help push those standards, and not one which is owned by Nvidia. But they are not technically blocked from doing so.
 
JMccovery said:
AMD doesn't use CUDA because Nvidia will not let them have access to the source codes so AMD can compile it for themselves. Nvidia wants to compile a CUDA runtime for AMD cards
Click to expand...
That is not the problem, but since you brought it up...

It's not a matter of "compiling" but how CUDA is tuned for nvidia architecture that would make a high performance port difficult. The difference in the organization of the ALUs and hardware dependent techniques will always make CUDA run more efficiently on nvidia's hardware. (Apple's work with nvidia on creating OpenCL for standardization submission, which is very much like CUDA that it is based on, gets rid of or modifies the nvidia dependencies.) nvidia has tuned CUDA performance for years and it even works with the lowly 8400GS. AMD didn't have CUDA required hardware support, and not even complete, until the HD 4000 series.

AMD has no desire to support CUDA for the same reasons nvidia has no desire to support Stream, sort of. Each is the others' "native" GPGPU programming API. It may not run well and in the case of Stream, no large base of application code exists as with CUDA.

It's probably a matter of time before nvidia moves PhysX to OpenCL. Don't expect any miracles like equal performance. OpenCL performance can often be dependent on how the code is written or tuned for a particular architecture. nvidia has no incentive to make it perform well on AMD cards.

Anyways, nvidia made a half-hearted offer towards AMD a couple of years ago to help it support CUDA. AMD wasn't interested.
 
Tsumi said:
I doubt you will ever see nVidia allow PhysX on AMD GPU's. Like I said, this decision was most likely made from a purely business standpoint, in the fact that nVidia won't have any options for SLI motherboards for AM3+. Trust me, if they still had SLI motherboards compatible with bulldozer, you would not see SLI on 9xx chipsets. Whatever nVidia can keep to themselves, they will keep to themselves. They tried that last time with Intel and forcing NF200 chips... didn't work too well. Guess they learned their lesson from Intel and decided to just allow it, instead of trying to force NF200 chips on AMD boards. If they had been successful with Intel and was able to force NF200 chips to allow SLI, then we would see AMD boards requiring NF200 chips for SLI.
Click to expand...

If you look at it the nf200 chip basically create a qpi/dmi path whatever intel wanna call it. They work the same way. Both network based, packets wrapped in a overhead, point to point link, same encoding etc etc. Now if efficiency and latency been the achilles heel of Pci-e surely all that bandwidth said the qpi is capable of must be far fetch as they have the same limitations.
 
Spooony said:
If you look at it the nf200 chip basically create a qpi/dmi path whatever intel wanna call it. They work the same way. Both network based, packets wrapped in a overhead, point to point link, same encoding etc etc. Now if efficiency and latency been the achilles heel of Pci-e surely all that bandwidth said the qpi is capable of must be far fetch as they have the same limitations.
Click to expand...

Which is completely unrelated to the point I was making. Why don't you go troll somewhere else.
 
Spooony said:
If you look at it the nf200 chip basically create a qpi/dmi path whatever intel wanna call it. They work the same way. Both network based, packets wrapped in a overhead, point to point link, same encoding etc etc. Now if efficiency and latency been the achilles heel of Pci-e surely all that bandwidth said the qpi is capable of must be far fetch as they have the same limitations.
Click to expand...

Umm... If I remember correctly, the NF200 is the same chip that has been used on all dual gpu cards from Nvidia since the 9800GX2, in that all it is, is a PCI-Express switch (Nvidia says that it is more than just a PCI-E switch). It can be used on any motherboard containing any PCI-Express chipset two split either 8 or 16 lanes into two 16-lane connections.
 
Tsumi said:
Which is completely unrelated to the point I was making. Why don't you go troll somewhere else.
Click to expand...

WTF are you on about? Nf200 is a duplexer. there are numerous board manufacturers not just intel. They can't force a thing on intel which intel created that's Pci-e and its connection protocol.
The nf200 is used when you use 3 or 4 gpus to make the scaling better. It can't be forced onto anything coz its not needed to run 2 gpus.
 
JMccovery said:
Umm... If I remember correctly, the NF200 is the same chip that has been used on all dual gpu cards from Nvidia since the 9800GX2, in that all it is, is a PCI-Express switch (Nvidia says that it is more than just a PCI-E switch). It can be used on any motherboard containing any PCI-Express chipset two split either 8 or 16 lanes into two 16-lane connections.
Click to expand...

The NF200 is just a dynamic PCI-e bandwidth allocation balancer. It doesn't do anything to improvethe efficiency of the rendering pipeline, it just tries to make sure that every card gets the PCI-e bandwidth it needs.
Its a duplexer basically. It creates more qpi/dma/Pci-e lanes but it still follows the connection protocol of the mentioned above. Remember Pci-e is a point to point link same as qpi and dmi.
 
Last edited:
Spooony said:
WTF are you on about? Nf200 is a duplexer. there are numerous board manufacturers not just intel. They can't force a thing on intel which intel created that's Pci-e and its connection protocol.
The nf200 is used when you use 3 or 4 gpus to make the scaling better. It can't be forced onto anything coz its not needed to run 2 gpus.
Click to expand...

Exactly...it's on tons of SB mobos, actually, last I looked - due to SB not having enough lanes. It's on dual GPU cards to allow them to both work over the same PCIe interface.

Tsumi just wants to rail agianst nvidia, really...
 
Spooony said:
WTF are you on about? Nf200 is a duplexer. there are numerous board manufacturers not just intel. They can't force a thing on intel which intel created that's Pci-e and its connection protocol.
The nf200 is used when you use 3 or 4 gpus to make the scaling better. It can't be forced onto anything coz its not needed to run 2 gpus.
Click to expand...

Why don't you actually read my post, and go read up on your history.

nVidia tried to do a thing where they would only license Intel chipset boards for SLI only if they contain a NF200 chip. I did not mention at any point number of PCI-E lanes, what the NF200 actually does, or anything like that. The point simply is, nVidia initially refused to license initial P55 and x58 boards unless they contained a NF200 chip in their build. Most manufacturers decided not to, and most consumers decided that buying a NF200 board was not worth the additional cost (approximately $30 over regular version), and simply did not buy SLI setups. That, and pressure from Intel, finally got nVidia to relent and license SLI for regular P55 and x58 boards.

Wait, I guess you wouldn't know this because you are very ignorant of most things computer related. Of course you wouldn't know the history of SLI on P55 and X58 chipsets. How could I have thought that highly of you?

rampantandroid said:
Exactly...it's on tons of SB mobos, actually, last I looked - due to SB not having enough lanes. It's on dual GPU cards to allow them to both work over the same PCIe interface.

Tsumi just wants to rail agianst nvidia, really...
Click to expand...

At what point did I rail against nVidia? I'm just saying that was their business strategy in the past, and it backfired on them. That's why they're just going ahead with allowing SLI licensing on AMD chipsets, instead of trying to do something similar with x58 and P55. And I've got nothing against nVidia, look at my signature. I use two GTX 470's.

The whole point of this thread is about nVidia allowing SLI on AMD chipsets, and speculation as to why they're doing that/discussions about it. I have laid out my reasons, and some people aren't logical enough to understand the entire post instead of just nitpicking at a few sentences that stand out for them.
 
Last edited:
Tsumi said:
Why don't you actually read my post, and go read up on your history.

nVidia tried to do a thing where they would only license Intel chipset boards for SLI only if they contain a NF200 chip. I did not mention at any point number of PCI-E lanes, what the NF200 actually does, or anything like that. The point simply is, nVidia initially refused to license initial P55 and x58 boards unless they contained a NF200 chip in their build. Most manufacturers decided not to, and most consumers decided that buying a NF200 board was not worth the additional cost (approximately $30 over regular version), and simply did not buy SLI setups. That, and pressure from Intel, finally got nVidia to relent and license SLI for regular P55 and x58 boards.

Wait, I guess you wouldn't know this because you are very ignorant of most things computer related. Of course you wouldn't know the history of SLI on P55 and X58 chipsets. How could I have thought that highly of you?



At what point did I rail against nVidia? I'm just saying that was their business strategy in the past, and it backfired on them. That's why they're just going ahead with allowing SLI licensing on AMD chipsets, instead of trying to do something similar with x58 and P55. And I've got nothing against nVidia, look at my signature. I use two GTX 470's.

The whole point of this thread is about nVidia allowing SLI on AMD chipsets, and speculation as to why they're doing that/discussions about it. I have laid out my reasons, and some people aren't logical enough to understand the entire post instead of just nitpicking at a few sentences that stand out for them.
Click to expand...

no nvidia wanted 5usd per mobo to remove their driver blocks simple. They won't certify a board compatible with 3-way sli without the nf200 chip because the scaling would be bad and it would come back to nvidia in the end. Same reason back when a 3 Crossfire setup was awful. It pointed back to Amd in the end which actually is just a board that were struggling for bandwidth. If Amd did the same no way they would certify a board as capable for 3-way when it doesn't have the bandwidth.
Those were back in the days aswell when boards didn't have the bandwidth that they got today. So no one was forced they were not just going to certify something which a other manufacturer will use as a marketing tool which will not work or perform well where the bad performance in the end will be pointed towards them.
 
Spooony said:
no nvidia wanted 5usd per mobo to remove their driver blocks simple. They won't certify a board compatible with 3-way sli without the nf200 chip because the scaling would be bad and it would come back to nvidia in the end. Same reason back when a 3 Crossfire setup was awful. It pointed back to Amd in the end which actually is just a board that were struggling for bandwidth. If Amd did the same no way they would certify a board as capable for 3-way when it doesn't have the bandwidth.
Those were back in the days aswell when boards didn't have the bandwidth that they got today. So no one was forced they were not just going to certify something which a other manufacturer will use as a marketing tool which will not work or perform well where the bad performance in the end will be pointed towards them.
Click to expand...

No, they initially wanted "rigorous and extensive" testing involving sending sample boards to them to "ensure" compatibility with SLI, which drives up the cost of the boards. Boards with NF200 chips were automatically given SLI, but those that weren't, had to be individually sent in. Then nVidia would give them a special code to put into the bios that would allow the drivers to allow SLI. That was before the 5 USD per board thing.
 
Plenty of sad stories with certain hardware setups, IDE transfer corruption (if i recall this right). There was always something going on which was troublesome for some.
 
Acer_Sheep said:
Could you be more exact why? I didn't hear about it.
Click to expand...

Let's see:
Data Corruption in AHCI/RAID mode
MediaShield corrupting downloads
Ran extremely hot (let that little fan on the chipset die, and it's over)
Some nForce 4/5 implementations suffered from the same PCI issues as the nForce 1/2
 
You must log in or register to reply here.
Back
Top