How come we only have 3 CPU/GPU brands?

[mr moose]

6 minutes ago, Bhav said:

No, Voodoo 4 + 5 were turd:

 

https://www.anandtech.com/show/535/8

that really depends on the CPU and the resolution.  The 5500 seemed to perform a lot better or at least withing margin of error in the higher resolution runs. 

 

[Bhav]

1 minute ago, mr moose said:

that really depends on the CPU and the resolution.  The 5500 seemed to perform a lot better or at least withing margin of error in the higher resolution runs. 

 

The 5500 was also highly expensive, and not compared in that review were the Geforce 2's which were the actual competing products due to how delayed then Voodoo 4 + 5 were.

[Apepa]

44 minutes ago, TopHatProductions115 said:

@Almostbauws

Last I checked, Intel holds the x86 ownership. AMD holds the ownership for the extension, x86_64 (AMD64). Thus, Intel holds the OG while AMD has the better extension that we all use to this day. Correct me if I'm wrong...

Basically. That's why I'm curious about how Via can produce modern x86 chips when all it's licences date back to Cyrix's settlements in the 90s, well before AMD64 came about.

[Apepa]

Just now, Apepa said:

Basically. That's why I'm curious about how Via can produce a modern x86 chip when all it's licences date back to Cyrix's settlements in the 90s, well before AMD64 came about.

 

[Apepa]

40 minutes ago, dalekphalm said:

There used to be many CPU manufacturers. I assume you're referring to x86 CPU's specifically? Because there are more than just AMD and Intel making CPU's right now.

 

You have:

1. IBM

2. Qualcomm

3. Samsung

4. Apple

5. NVIDIA (I know you listed them above, but I was unsure if you knew they made CPU's)

 

There are others, but I can't think of them off the top of my head. The difference, of course, is none of these are x86 CPU's

Since we're being smartarses, I should point out that 3 of those companies base their CPUs on the same licensed designs.

 

Since we're being smartarses.

[dalekphalm]

6 minutes ago, Apepa said:

Since we're being smartarses, I should point out that 3 of those companies base their CPUs on the same licensed designs.

 

Since we're being smartarses.

Not entirely true.

 

While Samsung manufactures Qualcomm SoC's, they also design their own SoC's - both are ARM designs, but the architecture is unique between them. That'd be like saying "AMD and Intel use the same licensed designs" because they're both x86.

 

Apple also heavily designs their own architecture, which ls also ARM.

 

NVIDIA does all kinds of shit, including more typical ARM designs, as well as crazy stuff like the Denver SoC.

 

Since we're being smart"arses", and all.

[mr moose]

22 minutes ago, Bhav said:

The 5500 was also highly expensive, and not compared in that review were the Geforce 2's which were the actual competing products due to how delayed then Voodoo 4 + 5 were.

I thought the geforce 2s were released later

 

[Apepa]

6 minutes ago, dalekphalm said:

Not entirely true.

 

That'd be like saying "AMD and Intel use the same licensed designs" because they're both x86

 

Since we're being smart"arses", and all.

Not quite, since ARM basically runs the opposite business model to Intel there's a lot more scope for custom designs to borrow from existing ARM ones than there is for x86 development. Whereas Intel is already threatening to sue Qualcomm just for developing x86 emulation in its ARM CPUs.

[dalekphalm]

Just now, Apepa said:

Not quite, since ARM basically runs the opposite business model to Intel there's a lot more scope for custom designs to borrow from existing ARM ones than there is for x86 development. Whereas Intel is already threatening to sue Qualcomm just for developing x86 emulation in its ARM CPUs.

Sure, but saying that an Exynos SoC is the "same design" as an A11, or the same as a SnapDragon 835, is disingenuous. They are all quite different from one another.

 

An Exynos M2 Core vs a Cortex A73 Core vs an Apple Monsoon Core are all very different.

 

So, we can agree that ARM is much more flexible with the way they allow licensing, but to say they're all based on the same licensed designs is missing the point.

[Apepa]

6 minutes ago, dalekphalm said:

to say they're all based on the same licensed designs is missing the point.

Missing the point to the same degree as listing ARM/POWER etc. manufacturers, when the OP was clearly talking about mainstream Wintel desktop chip makers? 

 

[dalekphalm]

Just now, Apepa said:

Missing the point to the same degree as listing ARM/POWER etc. manufacturers, when the OP was clearly talking about mainstream Wintel desktop chip makers? 

 

It might entirely be possible that the OP had not considered other CPU's. Hell, they probably didn't even know about the PowerPC CPU.

 

Plus they didn't seem to know that VIA still makes x86.

 

So I cannot make any assumptions about the OP's intentions.

 

But I think we've derailed long enough

[Apepa]

6 minutes ago, dalekphalm said:

Plus they didn't seem to know that VIA still makes x86.

I didn't even know VIA still makes x86 (until I read it in this thread), which makes me sad because I used to own one of those Cyrix CPUs from the 90s that were faster and cheaper than the Intel chips they were socket-compatible with.

 

1996-2006 was such a great time to be a PC enthusiast.

[DocSwag]

11 hours ago, RorzNZ said:

35 watts for 8c8t @ 2GHz isn't too bad tbh. If its a cheap price, which it might well be it would be pretty nice!

Still not much better than the jaguar cores in stuff like the Xbox one and ps4. And it sounds like they aren't capable of clocking much higher so...

 

Pretty much sounds like jaguar level cores that can't clock high at all.

[Mr.Meerkat]

On 2/9/2018 at 2:12 PM, Apepa said:

They both do, amd64 is an extension of the architecture Intel first developed in the 1970s and later licenced to AMD.

Not sure if anyone has corrected you yet but...

 

Correction: AMD64 (aka. X86-64,. 64 bit X86) is the 64-bit extension of Intel's X86 ISA and is created (and owned) by AMD. This results in a cross-license agreement between AMD and intel where AMD gets to use Intel's X86 while Intel gets to use AMD's X86-64. 

[Apepa]

8 minutes ago, Mr.Meerkat said:

Not sure if anyone has corrected you yet but...

 

Correction: AMD64 (aka. X86-64,. 64 bit X86) is the 64-bit extension of Intel's X86 ISA and is created (and owned) by AMD. This results in a cross-license agreement between AMD and intel where AMD gets to use Intel's X86 while Intel gets to use AMD's X86-64. 

That's what I said.

[Mr.Meerkat]

1 minute ago, Apepa said:

That's what I said.

Actually re-reading your comment, I guess? I find the way you've structured and worded it a bit confusing . 

[Apepa]

3 minutes ago, Mr.Meerkat said:

Actually re-reading your comment, I guess? I find the way you've structured and worded it a bit confusing . 

Context. Someone said it was owned by Intel, someone else said it was owned by AMD, I was pointing out that it's essentially owned by both. Since AMD64 is derived from Intel IP, a third party would need licenses from both Intel and AMD in order to use it.

 

Which is incidentally why I keep asking how VIA is still making x86 chips. I know they have rights to x86 that they got from acquiring Cyrix, but those date back to 32-bit designs from the 90s, and I can't find anything online about them licensing the extensions from AMD.

[Guest]

1 hour ago, DocSwag said:

Still not much better than the jaguar cores in stuff like the Xbox one and ps4. And it sounds like they aren't capable of clocking much higher so...

 

Pretty much sounds like jaguar level cores that can't clock high at all.

well if its a cheap price is what i'm saying

[AMDNeko]

1 hour ago, Apepa said:

Context. Someone said it was owned by Intel, someone else said it was owned by AMD, I was pointing out that it's essentially owned by both. Since AMD64 is derived from Intel IP, a third party would need licenses from both Intel and AMD in order to use it.

 

Which is incidentally why I keep asking how VIA is still making x86 chips. I know they have rights to x86 that they got from acquiring Cyrix, but those date back to 32-bit designs from the 90s, and I can't find anything online about them licensing the extensions from AMD.

 

  Ya...for those unawares basically the first "true" X86 (32bit as we know it) chip was the 80386 from intel which amd also made..then Intel tried making 64bit arc with IA-64 which was a horrible mess while AMD made x86-64 possible with AMD64 and therefor locked market in for 64bit processors..VIA has x86 rights from Cyrix as stated and AMD & Intel have cross licensing agreement as stated..this alone is why AMD "cannot die" and will never be purchased either as if AMD changes hands the rights to AMD64 are basically invalidated. Considering Via just got back into x86-64 game though its safe to assume we can maybe see something soon.

 

While any manufacturer can go to AMD for x86-64 licensing most see x86 as long in the tooth and prefer licensing ARM as its more lucrative where it matters and x86-64 on mobile is still somewhat of a mess.

[BirdyTheMighty]

Isn't the x86/x64 is copyrighted by Intel and then x64 by amd and then licenced it to Intel. Soo if there were to be a new cpu company that isn't a power cpu aka non x86 consoles. They would pay a hefty price

 

Ps: lol should of read the post in the thread people already mentioned this

[2FA]

The original x86 patent has long expired so there isn't an actual license on it anymore, what hasn't expired though are the much newer additions to the architecture such as the instruction set extensions (think SSE and AVX). The x86_64 patent is still valid though.

[asus killer]

for gpus there is something that no one mentioned, besides the complexity of the hardware there is also the problem of keeping the software side up to date (drivers) especially in games.

 

for cpus like many mentioned the cross licensing between amd and intel for x68 and 64 makes it impossible to have any more players on the consumer PC market.

[dalekphalm]

6 minutes ago, asus killer said:

for gpus there is something that no one mentioned, besides the complexity of the hardware there is also the problem of keeping the software side up to date (drivers) especially in games.

 

for cpus like many mentioned the cross licensing between amd and intel for x68 and 64 makes it impossible to have any more players on the consumer PC market.

Not impossible, but very impractical.

 

Any new CPU manufacturers would have to do one of the following:

1. Convince Intel and AMD to license out x86 and x86-64 (Incredibly unlikely)

2. Design a new CPU architecture, and then either manually integrate support into an Open Source OS (eg: Linux), or convince Microsoft to add support into Windows

2.a. Furthermore, they would have to then ensure that their CPU is supported by API's such as DX11, DX12, Vulkan, etc

2.b. For gaming, additional support at the driver level by AMD/NVIDIA may be required

 

It's definitely not impossible, but it would require a HUGE time and money investment. Probably multiple Billions of USD, and enough power to convince or force the various other companies to implement support for the new CPU arch - whether this is willing, done with a bucket of money, or other tactics such as buyouts/mergers.

 

New GPU architecture has less roadblocks. All you need really, is LOTS of money, and good engineers. You don't need the underlying OS to support it, since you can just write drivers, same with API's for gaming. The only problem is that very few companies have the required money, and little or none of them have the desire to put that much of it on the line for potentially small monetary gains.

[NinjaQuick]

2 hours ago, PPCs-Kat said:

 

While any manufacturer can go to AMD for x86-64 licensing most see x86 as long in the tooth and prefer licensing ARM as its more lucrative where it matters and x86-64 on mobile is still somewhat of a mess.

The last bit basically touches on an important subject. Other comments have duly pointed out x86 is not actually proprietary anymore, sse is. Avx (and newer is). However, why is it we don't see ARM on desktops is an important question, and the answer also answers why x86 doesn't really exist on mobile. 

 

The desktop, or more specifically windows, depends on x86 for legacy reasons. Much of what x86 ISA accomplishes (here as the sum of all extensions) is doable in a few more clocks on ARM, while some commands simply have no comparison and will take much much longer using ARM ISAs. X86 is not efficient by any means, at least not in the way ARM is, but trades that off for very long instruction sets that can be executed very quickly at some energy cost. Microsoft has been making an effort to bring arm into the fold, but compilers are all very highly tuned for the needs of x86 processors, and very few developers are putting our software that natively can run on ARM. 

 

Now, can a newcomer disrupt x86? Yes. It would take 4 or 5 years to get to market, and while licensing would be a challenge, it is not something impossible to achieve. Both Intel and amd would actually benefit from a third real competitor, and can't just deny licensing for no reason other than to have no competition. A very large part of the cost of making cpus is just the fab facilities themselves. If manufactured as a customer of an existing fab, the majority cost then becomes RnD. Which is a actually more of a smoking gun here. Any talent needed to make a new arch and that can design something that can compete with amd or Intel is already tied up by nda and contract with the very same companies. It takes hundreds of thousands if not millions of man hours to really nail a micro architecture of the scale we are talking, and to achieve a real change you need expert talent. This is why Apple suddenly surged ahead to the front of the ARM race. They have the money to develop cpus. They could have invested into x86, but there's already two giants there, and the products Intel has been making has been very much to apples request at times. Nobody new shows up because there's simply no need for someone else. Amd and Intel own a majority of the talent, and what little talent slips away goes on to make radical changes across the industry, especially in mobile. This is again, not because arm is cheaper to develop, in fact more money is spent on arm each year than on x86, in terms of raw RnD budgets combined. 

 

 

GPUs are a different matter and I won't fit into the ten minutes I have to talk.

[NinjaQuick]

Alright, so GPUs.

 

GPUs are not as prohibitively expensive to develop, since they are very simple instruction pipelines. They can be fully designed in software, and are very scalable, allowing for easy scaling of the complexity to allow a more complete usage of a silicone wafer. And while Driver development is expensive, for sure, there's no problem supporting DX12/OpenGL/OpenCL, etc if the hardware has native considerations for the math needed, and the driver simply has to point the call in the right direction. What makes entry difficult is that Nvidia and AMD already have the channels somewhat clogged with their portfolios, getting graphics DRAM is hard as is, and finding time at a fab with a competitive process only makes it all even harder. Like, we may see processors as impossibly complex things, and they are - extremely complex. But they are fundamentally just a series of tradeoffs. Every CPU (and GPUs, too) takes ones and zeros in, and pushes those signals through transistors, the chain of ones and zeroes ultimately being what determines what the processor does at the next junction. This could happen a million times to a single signal before it is done going through the gates and comes out the other end as a result. Since there are only so many ways to make an addition instruction work, the real differences are ultimately how far common instructions have to go (farther = more time, maybe even more clocks, more time is more heat), and in turn, which requests get prime real estate, and which are relegated to being multi-clock tasks. Generally speaking, this can be fairly automated and straight forward to devise in rendering-related instructions. Creating compute is a little more complicated, but with designers knowing already what limited scope they will have to design for (Windows or Open Source APIs, very little else), even this isn't an insurmountable challenge.

 

The hardest part for any newcomer is getting test parts, getting experienced designers, engineers and architects - and breaking into a market where most users assume the only cards available to buy are GeForce.