Nvidia reportedly talking with VIA about acquisition

[Idontcare]

Nvidia reportedly talking with VIA about acquisition
Monica Chen, Taipei; Joseph Tsai, DIGITIMES [Tuesday 18 March 2008]

Nvidia reportedly was in talks with VIA Technologies about a possible acquisition, but no deal was made due to the high price named by VIA, according to sources at motherboard makers.

Nvidia and VIA discussed three possible scenarios, including a strategic alliance, acquisition of only VIA's processor division, or the acquisition of the entire company, according to the sources. Though no deal was agreed upon, since VIA continues to see losses, the company is expected to open up negotiations again soon with a lower price, noted the sources.

VIA has denied the speculation about an acquisition by Nvidia and pointed out that the company currently does not have a plan to reduce its capital as recent reports in Chinese-language media have suggested. The company said that its main business is still focused on x86 CPUs and chipsets and therefore will not give up these segments.

http://www.digitimes.com/mobos/a20080318PD212.html

Now wouldn't that be intriguing.

 

[Tim]

Nvidia could put VIA tech to use and make horrible performing video cards!
... Of course the amazingly low power requirements would blow the competition away.


Kidding aside, it is an interesting bit of news.

 

[AmberClad]

Interesting. So would that possibly lead to essentially three different camps for CPU/GPU combos. DAAMIT, Intel CPU+Intel Larrabee, VIA CPU+Nvidia GPU? Course, with the exception of DAAMIT, the other two are just hypotheticals at this point.

 

[Idontcare]

Originally posted by: AmberClad
Interesting. So would that possibly lead to essentially three different camps for CPU/GPU combos. DAAMIT, Intel CPU+Intel Larrabee, VIA CPU+Nvidia GPU? Course, with the exception of DAAMIT, the other two are just hypotheticals at this point.

That's the way I read it.

Intel competes fresh with little experience in the discreet GPU market, Nvidia competes fresh with little experience in the discreet CPU market (but with no hands-on fab experience, so still limited to foundry technology), and AMD competes in both with experience in both and with fab experience.

Wouldn't that make for an intriguing 2010 or 2011 when consumer products from such a venture ought to start hitting the shelves?

 

[Modelworks]

Its not going to allow nvidia to make x86 cpu.
Via has a license from intel, but like amd it does not transfer if the company is sold.
There is currently no way for nvidia to get a x86 license unless they purchase it directly from intel.

I think if nvidia did buy via it would be for making mobile devices, embedded designs.
via has some really low power cpu technology, teamed with the nvidia chipset could make some really nice low power devices.

 

[Idontcare]

Originally posted by: Modelworks
Its not going to allow nvidia to make x86 cpu.
Via has a license from intel, but like amd it does not transfer if the company is sold.
There is currently no way for nvidia to get a x86 license unless they purchase it directly from intel.

I think if nvidia did by via it would be for making mobile devices, embedded designs.
via has some really low power cpu, teamed with the nvidia chipset could make some really nice low power devices.

My understanding is this is technically correct, no one disagrees with the letter of the contract.

But somehow Via managed to "retain" the x86 licenses that Cyrix and Centaur had when Via purchased them...soooo whatever leverage made that possible (be it outright $$$, unofficial DoJ pressure, whatever) just a few short years ago I am willing to openly speculate it will probably be brought to bear again.

Nvidia has been holding out on SLi for good reason, who knows what kind of "cross licensing" deal they are holding out for. Maybe not probable, but it is plausible.

 

[Modelworks]

Cyrix was able to get out of it because they didn't license the technology from intel at the time.
They reverse engineered an intel x86 cpu, and tried to claim that certain functions discovered were not proprietary to intel.
Intel rather than make it all public in court, settled with cyrix, and allowed them to make their cpus at fabs that had intel licenses. Via is still under that agreement.

Thats not going to happen again.
Intel is way to big now and would be more than willing to tie it up in court for years.

I think people need to focus on the other tech via has besides the x86 cpu.

 

[Idontcare]

Originally posted by: Modelworks
Cyrix was able to get out of it because they didn't license the technology from intel at the time.
They reverse engineered an intel x86 cpu, and tried to claim that certain functions discovered were not proprietary to intel.
Intel rather than make it all public in court, settled with cyrix, and allowed them to make their cpus at fabs that had intel licenses. Via is still under that agreement.

These are facts? or opinions? I can't tell which. I was not aware Intel x86 licenses were tied to fabs...that is an "odd" licensing model to say the least and is certainly "news" to me.

Regardless, you still didn't address Centaur, or how NexGen got their license before AMD bought them. Why then but not now?

I don't see business sense for Intel to goad the DoJ into an inquiry...it won't be good for Intel and it desparately won't be good for Intel's shareholders. It's not about corporate might, it is about making money and ensuring you don't do silly political stuff which jeopardizes your ability to continue making gobs of money. Inviting a DoJ inquiry is unprecedented, can't think of a single example of a company doing this by choice. Microsoft in Europe perhaps, but that isn't a USA DoJ thing.

Originally posted by: Modelworks
I think people need to focus on the other tech via has besides the x86 cpu.

The intriguing part of these x86 discussions is not because Via has existing x86 chips...its because such a scenario gives Nvidia a plausible crowbar to angle themselves into the x86 CPU market by way arguing "precendence" with the x86 license transfers that have happened in Via's history.

Presumably Nvidia's own CPU design would be rolled out, not merely extensions centaur designs.

There's really nothing preventing Nvidia from designing an x86 compatible CPU, they just can't sell it until they have the license rights. So who knows how much time they have been biding, no reason to show your hand if no one has called the table.

Meh, I really don't enjoy these types of debates...they are usually too light on facts and too heavy on opinion. This whole thread is just about speculation, even the article cited is just a rumor. But there is no harm in running out "what if" scenarios just to see how such worlds would operate, is there?

 

[Modelworks]

Originally posted by: Idontcare

Originally posted by: Modelworks
Cyrix was able to get out of it because they didn't license the technology from intel at the time.
They reverse engineered an intel x86 cpu, and tried to claim that certain functions discovered were not proprietary to intel.
Intel rather than make it all public in court, settled with cyrix, and allowed them to make their cpus at fabs that had intel licenses. Via is still under that agreement.

These are facts? or opinions? I can't tell which. I was not aware Intel x86 licenses were tied to fabs...that is an "odd" licensing model to say the least and is certainly "news" to me.

They are facts.
http://www.ll.georgetown.edu/f...5opinions/95-1246.html
http://www.solarnavigator.net/sponsorship/cyrix.htm

Unlike AMD, Cyrix had never manufactured or sold Intel designs under a negotiated license. Cyrix's designs were the result of meticulous in-house reverse-engineering. So while AMD's 386s and even 486s had some Intel-written microcode software, Cyrix's designs were completely independent. Focused on removing potential competitors, Intel spent many years in legal battles with Cyrix, claiming that the Cyrix 486 violated Intel's patents. (Just as Intel did with every other x86 CPU manufacturer right up until 1998.)



By and large, Intel lost the Cyrix case. But the final settlement was out of court: Intel agreed that Cyrix had the right to produce their own x86 designs in any foundry that happened to already hold an Intel license. Both firms gained out of this: Cyrix could carry on having their CPUs made by Texas Instruments, SGS Thomson, or IBM (as it happened, all holders of Intel cross-licenses); Intel avoided a potentially embarrassing loss.



The follow-on 1997 Cyrix-Intel litigation was the reverse: instead of Intel claiming that Cyrix 486 chips violated their patents, now Cyrix claimed that Intel's Pentium Pro and Pentium II violated Cyrix patents?in particular, power management and register renaming techniques. The case was expected to drag on for years but eventually was to be settled out of court. In fact it was settled quite promptly, by another mutual cross license agreement. Intel and Cyrix now had full and free access to each others patents. The settlement didn't say whether the Pentium Pro violated Cyrix patents or not, it simply allowed Intel to carry on making them either way?exactly as the previous settlement side-stepped Intel's claim that the Cyrix 486 violated Intel patents.

 

[Idontcare]

Originally posted by: Modelworks

Originally posted by: Idontcare

Originally posted by: Modelworks
Cyrix was able to get out of it because they didn't license the technology from intel at the time.
They reverse engineered an intel x86 cpu, and tried to claim that certain functions discovered were not proprietary to intel.
Intel rather than make it all public in court, settled with cyrix, and allowed them to make their cpus at fabs that had intel licenses. Via is still under that agreement.

These are facts? or opinions? I can't tell which. I was not aware Intel x86 licenses were tied to fabs...that is an "odd" licensing model to say the least and is certainly "news" to me.

They are facts.
http://www.ll.georgetown.edu/f...5opinions/95-1246.html
http://www.solarnavigator.net/sponsorship/cyrix.htm

Ah, I see the disconnect now, you used the word "fabs" in your post when you meant to write "any foundry with existing licenses". I was wondering why it would be tied to a fab...I see it is not. So it was a fact but only not the way you wrote it, the way you wrote it was false, ergo my confusion. Fabs don't have licenses, the company's that own fabs have licenses. Company's close fabs and open new fabs, has nothing to do with their license.

Any who, this still doesn't explain why Via gets to manufacture x86 compatible licenses. They were not a foundry with x86 license in the 90's, nor does TSMC (current foundry of Via's chips) have such a license now.

Don't you think this is the slightest bit relevant to determining whether or not Nvidia has leverage to wriggle their way into an x86 license were they to procure Via?

 

[Lonyo]

http://www.via.com.tw/en/resou...pr030408patentcase.jsp

VIA and Intel Settle Patent Infringement Cases
VIA Technologies, Inc, a leading innovator and developer of silicon chip technologies and PC platform solutions, today announced that VIA and Intel Corporation have reached a settlement agreement in a series of pending patent lawsuits related to chipsets and microprocessors. The agreement encompasses 11 pending cases in five countries involving 27 patents.

Under terms of the settlement, both companies will dismiss all pending legal claims in all jurisdictions. The companies also entered into a ten-year patent cross license agreement covering each company's products. As part of the agreement Intel granted VIA a license to sell microprocessors that are compatible with the x86 instruction set but not pin compatible or bus compatible with Intel microprocessors.

Intel further agreed for a period of three years, not to assert its patents on VIA bus or pin compatible microprocessors. Intel also granted VIA a four year license to design and sell chipsets that are compatible with the Intel microprocessor bus and agreed not to assert its patents on VIA or its customers or distributors on such chipsets for a fifth year. The agreement will be royalty bearing to Intel for some products. The license agreements do not apply to S3 Graphics, a company partially owned by VIA.

Each company is responsible for its own attorney's fees. Specific financial and other terms of the agreement were not disclosed.

Intel initiated the litigations in September 2001, claiming that VIA's microprocessor and chipset products infringed Intel patents. VIA counter sued, claiming that Intel microprocessors infringed three patents VIA acquired in connection with its acquisition of IDT's Centaur subsidiary.

Runs from 2003-2013.

 

[Idontcare]

Well I'm definitely getting the trend here...so whomever buy's AMD or whomever Nvidia buy's they need to get themselves snarled in a patent lawsuit with Intel so they can "settle out of court" for a x86 license. Wonder how long that takes. Years?

 

[Modelworks]

Originally posted by: Idontcare
w fabs, has nothing to do with their license.

Any who, this still doesn't explain why Via gets to manufacture x86 compatible licenses. They were not a foundry with x86 license in the 90's, nor does TSMC (current foundry of Via's chips) have such a license now.

Don't you think this is the slightest bit relevant to determining whether or not Nvidia has leverage to wriggle their way into an x86 license were they to procure Via?

Via does have a x86 license.
Their chipset license though expired april 07, they have until april 08 to sell products .

Nvidia could be considering a new chipset totally different from intels designs and sockets.
The license does say they could make cpu's of their own form and pins.
That agreement is valid till 2011.


Its a big leap for nvidia and I don't know they have the capital for that.
It would mean releasing a motherboard with its own chipset, cpu socket style and buses.

Under terms of the settlement, both companies will dismiss all pending legal claims in all jurisdictions. The companies also entered into a ten-year patent cross license agreement covering each company's products. As part of the agreement Intel granted VIA a license to sell microprocessors that are compatible with the x86 instruction set but not pin compatible or bus compatible with Intel microprocessors.

Intel further agreed for a period of three years, not to assert its patents on VIA bus or pin compatible microprocessors. Intel also granted VIA a four year license to design and sell chipsets that are compatible with the Intel microprocessor bus and agreed not to assert its patents on VIA or its customers or distributors on such chipsets for a fifth year.The agreement will be royalty bearing to Intel for some products. The license agreements do not apply to S3 Graphics, a company partially owned by VIA.

Each company is responsible for its own attorney's fees. Specific financial and other terms of the agreement were not disclosed.

Intel initiated the litigations in September 2001, claiming that VIA's microprocessor and chipset products infringed Intel patents. VIA counter sued, claiming that Intel microprocessors infringed three patents VIA acquired in connection with its acquisition of IDT's Centaur subsidiary.

 

[Idontcare]

Thanks Modelworks and Lonyo for interjecting the facts about Via/Centaur/Cyrix into this thread. Very much a value-add.

Originally posted by: Modelworks
Its a big leap for nvidia and I don't know they have the capital for that.
It would mean releasing a motherboard with its own chipset, cpu socket style and buses.

I agree on that, not that Intel gets a free lunch trying to dive into the discreet GPU biz either, but Nvidia has an uphill climb no matter which way they attempt to expand if they intend to diversify their business model beyond the current market segments they compete in.

But wouldn't you agree that albeit a high-cost high-effort pathway to diversify (getting into x86)...purchasing Via is probably the lowest-cost easiest-effort pathway available to Nvidia to enter the discreet x86 compatible marketspace?

Not saying they would...but can you contemplate a more compelling alternate pathway to buying Via (for the licensing angle) IF Nvidia were intent on getting into the CPU business?

 

[Auric]

From fuzzy memory...

NexGen x86 compatible CPUs were not pin compatible with Intel so that in combination with being independently designed, no license was required. AMD aquired NexGen and they were then made pin compatible.

S3 outbid Intel during a blind auction for Exponential's IP and this was later used when VIA aquired S3 (Graphics) as the basis for a cross-licensing agreement.

 

[Modelworks]

I don't think there is any easy way for nvidia to do it , but via is probably the easiest way.
They would have until 2011 to make products, which is only 3 years.
Thats not a lot of time to develop a whole new socket, chipset and hardware.

What I would really really love to see is nvidia develop an entirely new processor that doesn't use x86 at all.
Basically a new platform.
Its been a good architecture and there is life left in it, but there are other technologies nvidia could pursue that wouldn't cause them the patent problems.

 

[Lonyo]

Originally posted by: Modelworks
I don't think there is any easy way for nvidia to do it , but via is probably the easiest way.
They would have until 2011 to make products, which is only 3 years.
Thats not a lot of time to develop a whole new socket, chipset and hardware.

What I would really really love to see is nvidia develop an entirely new processor that doesn't use x86 at all.
Basically a new platform.
Its been a good architecture and there is life left in it, but there are other technologies nvidia could pursue that wouldn't cause them the patent problems.

It would cause serious other problems, especially when you consider people like Intel are trying to go fully x86 (that's what they're driving towards with Silverthorne's successor, an x86 chip which can be used in mobiles).
Creating a non-x86 processor would be pretty pointless except for the embedded/mobile market which they already have a graphics presence in, and which Intel is planning to break into currently with x86 chips.
They'd have to try and compete with ARM, who already have a strong hold, and the might of Intel + x86.

I'm really not sure what they would be able to do with a non-x86 processor.

 

[v8envy]

Been tried -- remember Transmeta? I can't blame you if you don't. The ole Fish-n-chips wasn't all that interesting to enthusiasts.

It's not that the IA instruction set is awesome and great. The problem is the enormous amount of application code which only runs on the x86 code museum. Any desktop cpu in this day and age has to be instruction level compatible with Intel. If not approximately 0 existing applications will run well (including the Microsoft Office hairball, Flash nastiness, proprietary media players, games, etc) and your market penetration will likewise be approximately 0.

Power is a better architecture. Hell, the old motorola 68xxx was a better (cleaner) architecture. Itanic is debatable. This has nothing to do with superiority and everything to do with application availability.

 

[Modelworks]

I was thinking of Sparc or Mips.

x86 is starting to have issues.
Multiprocessing is really starting to be a problem for developers to implement on x86 and Sparc and Mips both handle multiprocessing well.
And there are plenty of programmers out there who know both.

Granted it would take more money than nvidia has.

 

[v8envy]

That's not CPU architecture at work there. Sequent made machines with dozens of Intel 386 and higher processors that scaled very, very well. The problem is the backplane. Desktops are designed for a completely different purchasing mentality than big, heavy servers.

The 'desktop' segment Sparc and mips boxes performed like utter, unbridled ass -- even with more CPUs. I remember doing some benchmarks back in 2004 or so on a Sunfire 480 (could be wrong here) and even with 4 ~1.3 ghz CPUs it couldn't hold a candle to a single 2 ghz opteron for ~50ish concurrent threads. Sun sales guys were pushing opterons, they knew how poorly the sparc chips did outside of a very narrow niche.

Give each CPU a nice fast dedicated path to RAM. Create a low latency, high bandwidth interconnect between them all and the Intel instruction set won't be holding you back. I guarantee it. Do the opposite and neither Sparc nor MIPS instruction sets will help.

 

[Idontcare]

Originally posted by: v8envy
That's not CPU architecture at work there. Sequent made machines with dozens of Intel 386 and higher processors that scaled very, very well. The problem is the backplane. Desktops are designed for a completely different purchasing mentality than big, heavy servers.

The 'desktop' segment Sparc and mips boxes performed like utter, unbridled ass -- even with more CPUs. I remember doing some benchmarks back in 2004 or so on a Sunfire 480 (could be wrong here) and even with 4 ~1.3 ghz CPUs it couldn't hold a candle to a single 2 ghz opteron for ~50ish concurrent threads. Sun sales guys were pushing opterons, they knew how poorly the sparc chips did outside of a very narrow niche.

Give each CPU a nice fast dedicated path to RAM. Create a low latency, high bandwidth interconnect between them all and the Intel instruction set won't be holding you back. I guarantee it. Do the opposite and neither Sparc nor MIPS instruction sets will help.

You should check out a Niagara2 system, if you are still into the stuff you were doing in 2004.

 

[soonerproud]

Originally posted by: Modelworks
Its not going to allow nvidia to make x86 cpu.
Via has a license from intel, but like amd it does not transfer if the company is sold.
There is currently no way for nvidia to get a x86 license unless they purchase it directly from intel.

I think if nvidia did buy via it would be for making mobile devices, embedded designs.
via has some really low power cpu technology, teamed with the nvidia chipset could make some really nice low power devices.

Are you so certain that Nvidia doesn't have their own X86 license? Nvidia has been hinting recently of doing their own x86 processor tyo combine with it's gpu, so it makes sense they already have a license.

 

[Modelworks]

The problem with x86 multiprocessing has nothing to do with the ram or the interfaces.
Its the architecture thats the problem.
IA is very serial in nature and that doesn't do well with multiple processors.

IA multiprocessing is great for things like number crunching but when it comes to real time application acceleration its very hard to program and use all cores.
x86 was not designed with multiple processors in mind, where as things like risc were.

 

[Modelworks]

Originally posted by: soonerproud

Originally posted by: Modelworks
Its not going to allow nvidia to make x86 cpu.
Via has a license from intel, but like amd it does not transfer if the company is sold.
There is currently no way for nvidia to get a x86 license unless they purchase it directly from intel.

I think if nvidia did buy via it would be for making mobile devices, embedded designs.
via has some really low power cpu technology, teamed with the nvidia chipset could make some really nice low power devices.

Are you so certain that Nvidia doesn't have their own X86 license? Nvidia has been hinting recently of doing their own x86 processor tyo combine with it's gpu, so it makes sense they already have a license.

They have a license to produce chipsets, but not cpu.

 

[v8envy]

Originally posted by: Modelworks
The problem with x86 multiprocessing has nothing to do with the ram or the interfaces.
Its the architecture thats the problem.
IA is very serial in nature and that doesn't do well with multiple processors.

IA multiprocessing is great for things like number crunching but when it comes to real time application acceleration its very hard to program and use all cores.
x86 was not designed with multiple processors in mind, where as things like risc were.

This is a bit of a red herring. Latency has nothing to do with the total aggregate instructions/time you're going to get out of your massively paralellized system. If your system calls for every process to migrate between cores for each instruction you've got problems no architecture will address. No process is that fine-grained.

Under the covers all the recent IA CPUs are running a very RISC-like processor. They're just decoding and executing a nasty CISC code museum instruction set in a superscalar way. This implies long pipelines, yes. This has 0 to do with being designed for multiple CPUs working on tasks concurrently -- four concurrently executing CISC instructions that do the same work as 16 RISC instructions and execute in the same amount of time are equivalent for all practical purposes.

MIPS and Sparc chips were designed to run UNIX and thusly were designed to optimize for context switching. That's why they scale so well on apps using the POSIX APIs. A less primitive operating system designed not to copy data from user space to kernel space at the drop of a hat would not show such a massive scaling advantage with those architectures. The OSes have as more to do with scaling to > 4 processors than instruction sets do -- it's not fair to compare the scaling of a CPU running a desktop lineage OS (windows) to a server heritage OS (Solaris). Run the same rev of Linux on both with the same apps and then let's compare scaling notes.

If we're talking about on-cpu primitives to implement concurrency control (mutexes etc) then you may have something. That hasn't been a priority for Intel & friends until recently. But that's not an instruction set or underlying architecture flaw -- it just means it hasn't been a top priority to optimize for.