Dr Su will present Zen 2 and Navi at Computex 2019 (May 27th)

[Trixanity]

[Sparviero]

Good to see. I had a hard time avoiding a stop at Microcenter today. 

[Stefan Payne]

9 hours ago, Master Disaster said:

There was actually 4 CPU vendors at one point.

more like A Dozend.

Who remembers Siemens x86 CPUs or Texas Instruments, IBM??
 

According to CPU World:

 

For 8086:
AMD, Fujitsu, Harris, Intel, Oki, MHS, Mitsubishi, NEC, Siemens

80286
AMD, Fujitsu, Harris, IBM, Intersil, Siemens

80386:

AMD, Chips, Cyrix, IBM, Intel, MC, Texas Instruments

80486:

AMD, Cryrix, IBM, INtel, ST, Texas Instruments, UMC

 

Later there were a couple of Startups like Transmeta, Rise Technology, NexGen, IDT.

 

They get less because Intel sued most of them out of x86 Cloning business back in the mid 80s or so.

 

But there is a third one:
Zhaoxin

And against them, Intel can't do shit...

[Sparviero]

On 4/3/2019 at 6:53 PM, Stefan Payne said:

 

 

But there is a third one:
Zhaoxin

And against them, Intel can't do shit...

Just need a good social score and party membership to get high end variants and overclock. Plus install an app so they can verify you aren’t using a vpn. And a tencent camera to monitor you while you sit at the computer. 

[Ithanul]

On 4/3/2019 at 7:37 AM, RejZoR said:

It's gonna be interesting to have 3 graphic vendors (AMD, NVIDIA and Intel) after what, 15 years of having only 2? Would be cool if we got a 3rd CPU vendor as well, like we used to back in the day. That goes a bit further back...

Do folks just don't pay attention to the fact there are still more than two processor vendors around?

 

They just don't compete in consumer market like Intel and AMD do.

The other vendors that are still around:

IBM

ARM

Ampere Computing

Fujitsu

Marvell Technology Group

 

Nice little read:  https://www.eweek.com/pc-hardware/eweek-s-top-vendors-server-processor-makers

[RejZoR]

7 minutes ago, Ithanul said:

Do folks just don't pay attention to the fact there are still more than two processor vendors around?

 

They just don't compete in consumer market like Intel and AMD do.

The other vendors that are still around:

IBM

ARM

Ampere Computing

Fujitsu

Marvell Technology Group

 

Nice little read:  https://www.eweek.com/pc-hardware/eweek-s-top-vendors-server-processor-makers

No they aren't. If we're gonna go that route there are hundreds of CPU companies. They don't make high performance enterprise or even consumer CPU's. Marvell for example makes CPU's for low power stuff like routers, you won't see Marvell chips powering laptops or desktop PC's. I even doubt they have anything for any kind of servers. Would you really call that a CPU vendor? Besides, it's kinda obvious we're talking x86 compatible CPU's here, not all CPU's in general.

[Ithanul]

20 minutes ago, RejZoR said:

No they aren't. If we're gonna go that route there are hundreds of CPU companies. They don't make high performance enterprise or even consumer CPU's. Marvell for example makes CPU's for low power stuff like routers, you won't see Marvell chips powering laptops or desktop PC's. I even doubt they have anything for any kind of servers. Would you really call that a CPU vendor? Besides, it's kinda obvious we're talking x86 compatible CPU's here, not all CPU's in general.

They sitll make processors.

And, like I said, they just don't make them for the consumer market like Intel and AMD.

 

Routers are specilized computers that sole job is routing.  Those chips are still processors.

 

And yes, I consider them CPU vendors as they make processors for computers.  Not for laptops or desktops, but those are not the only computers out there.  

 

Also, Marvell came into the server space during 2018:  

• Marvell propelled itself into the server chip space last year when it completed a $6 billion acquisition of Cavium, a key player in the Arm server chip market. Cavium had developed the ThunderX line of processors, coming out with ThunderX2 in May 2018.

 

And if sticking to x86, IBM and Fujitsu still make x86 processors for ZF Micro and VIA. 

[Drak3]

34 minutes ago, Ithanul said:

ARM

Does ARM really constitute a legitimate vendor?

 

BTW, you forgot Samsung, Apple, and Huawei.

[Ithanul]

8 minutes ago, Drak3 said:

Does ARM really constitute a legitimate vendor?

 

BTW, you forgot Samsung, Apple, and Huawei.

Ah yes, those three as well. 

 

ARM is technically not in the same sense as the others.  They design chips and license them out.

But, they did unveil their own edge and cloud computing platform last year.  As they are making the 16nm "Cosmos" platform and later the N1 built on their 7nm "Ares" platform.  The Ares is interesting as it scales up to 128 cores.

 

There is a lot of shake up in server and cloud market at the moment.

 

Hence, I'm curious what AMD plans with EYPC. Hopefully they will address the SEV vulnerability.

[Drak3]

1 minute ago, Ithanul said:

But, they did unveil their own edge and cloud computing platform last year.  As they are making the 16nm "Cosmos" platform and later the N1 built on their 7nm "Ares" platform.  The Ares is interesting as it scales up to 128 cores.

Ah, cool.

 

Hope it's not a flop like Operton A.

[Ithanul]

1 minute ago, Drak3 said:

Ah, cool.

 

Hope it's not a flop like Operton A.

Same.  I really hope they do well.  As I'm interested what they can manage with the that platform.

They even manage to get OEMs like HP, Lenovo, and Cray interested in using ARM-based SoCs in their systems.

 

So much stuff going on in the server, datacenter, and cloud market currently.

[Drak3]

1 minute ago, Ithanul said:

Same.  I really hope they do well.  As I'm interested what they can manage with the that platform.

They even manage to get OEMs like HP, Lenovo, and Cray interested in using ARM-based SoCs in their systems.

 

So much stuff going on in the server, datacenter, and cloud market currently.

What I'd really like to see is a new ARM design that was more like Apple's A series, put into systems like Synology's NASes.

[CarlBar]

1 hour ago, Ithanul said:

Do folks just don't pay attention to the fact there are still more than two processor vendors around?

 

They just don't compete in consumer market like Intel and AMD do.

The other vendors that are still around:

IBM

ARM

Ampere Computing

Fujitsu

Marvell Technology Group

 

Nice little read:  https://www.eweek.com/pc-hardware/eweek-s-top-vendors-server-processor-makers

 

Some of us are to varying degree's aware. But mobile phone CPU/APU's aside they're largely or entirely irrelevant to our use cases, (and mobile phone chips are only relevant to some of our use cases), so mentioning them is about as relevant as tits on a fish. 

[Stefan Payne]

15 hours ago, RejZoR said:

Marvell for example makes CPU's for low power stuff like routers, you won't see Marvell chips powering laptops or desktop PC's. I even doubt they have anything for any kind of servers. Would you really call that a CPU vendor?

Well, the most powerful Device that uses a Marvell CPU/SOC is the Playstation 4.

No, don't ask what Sony did and why they connected a CPU to a CPU to connect to IO...

15 hours ago, RejZoR said:

Besides, it's kinda obvious we're talking x86 compatible CPU's here, not all CPU's in general.

Yeah and there are not even a hand full companys that are allowed to make them...

 

In this case I'm totally for a 3rd party licensing that allows competition...

[RejZoR]

I wonder when x86 licensing thing even happened, given that Cyrix reverse engineered x86 CPU's from Intel and made their own. They even got patents for some of the stuff and later cross patents with Intel so they both could use all the stuff between each other. Was this super restrictive x86 licensing done later?

[Master Disaster]

2 hours ago, RejZoR said:

I wonder when x86 licensing thing even happened, given that Cyrix reverse engineered x86 CPU's from Intel and made their own. They even got patents for some of the stuff and later cross patents with Intel so they both could use all the stuff between each other. Was this super restrictive x86 licensing done later?

Intel invented the architecture with the 8086 in the 70s. As with all inventions you make it you own it. The modern X86 instruction set didn't exist until 80386 however, before that X86 was more of an embedded processor, ironically.

 

There's an interesting story here, basically even though they own they rights to the architecture they cannot stop people from making "compatibles" which is what happened with later revisions.

 

Intel renamed to Pentium and invented MMX Technology because names can be trademarked where as numbers cannot.

Because of this it caused AMD to start developing their own extension to X86 which became X86-64. Essentially because Intel tried to stop others from copying them it led to them losing their effective monopoly on the tech. Itanium flopped hard and X86-64 became the defacto architecture.

 

Edit

 

Another interesting fact that I just learned while making sure the above was factually correct.

 

Turns out it was Via who invented speculative execution so we can blame them for Spectre and Meltdown.

 

TIL

[Totally Average Gameplay]

 

Time to party early #AMDwins
(way too early)

[Jito463]

On 4/2/2019 at 12:52 PM, Master Disaster said:

AMD have also confirmed they'll be presenting Zen 2, both desktop and server variants as well as Navi

AMD needs to play "Hey, LISTEN!" when they show off Navi at Computex.

[mr moose]

On 4/6/2019 at 8:04 PM, Master Disaster said:

Turns out it was Via who invented speculative execution so we can blame them for Spectre and Meltdown.

 

TIL

Shooosh you,  what are all the Intel haters gonna say if they find out it wasn't entirely Intel's fault?

 

A. they'll probably just dismiss it by claiming they should have know better...

[Stefan Payne]

On 4/6/2019 at 11:04 AM, Master Disaster said:

Turns out it was Via who invented speculative execution so we can blame them for Spectre and Meltdown.

Nope, that is misinformation...

It was Intel with the first x86 microprocessor in 1995 with the Pentium PRO.

The AMD K5 was the first x86 compatible processor that also implemented it as well as Cyrix 6x86 a couple of Months later.

 

But the Pentium PRO was released at the end of 1995, K6 was march 1996.

https://en.wikipedia.org/wiki/AMD_K5
https://en.wikipedia.org/wiki/Cyrix_6x86

 

 

I though the Nx586 might have been the first but I couldn't find any proof of that. On the contrary, German Wikipedia lists speculative and out of order execution as features for the Nx686 (wich later became the K6)...

 

 

Anyway, I  doubt that the Pentium PRO was the first implementation of speculative execution, while it is possible, I'd say it was something used by a RISC Chip or something long forgotten. I really doubt that Intel had the first implementation of that. But maybe @leadeater knows more about that. I couldn't find much Information about speculative execution of such early processors but I also haven't looked into it too much...

And I exepct someone else to have done it before...

 

The only thing I could find was out of order execution. ANd whatever you think was the first OoO CPU, you're wrong. Its either the CDC6600 or the IBM System/360 Model 91, depending on how you look at it.
The first Out of Order Microprocessor was the Power1 from IBM as well...

 

 

But since they all came at roughly the same time, they could all be based on the same research papers from that time. And been developed from there...

 

Here Wikipedia on Out of Order Execution:

https://en.wikipedia.org/wiki/Out-of-order_execution

 

So from the looks of it, it looks like it was actually IBM with the first out of order processors, after that probably Fujitsu/HAL with the SPARC64 in 1995 as well as Intel with the Pentium PRO.

[leadeater]

12 minutes ago, Stefan Payne said:

-snip-

Informational sake here's an explanation of the difference between Out of Order Execution and Speculative Execution.

https://stackoverflow.com/questions/49601910/out-of-order-execution-vs-speculative-execution

 

Also x86 instruction set doesn't have to implement speculative execution and didn't start with it either. Where we can officially say x86 instruction set started is a bit of opinion but it's based off the 8086 Intel processor and every subsequent processor after that used it and developed it further, along with the other off shoot companies like Cyrix.

 

So the first x86 processor is either than 8086 or 80186. You can have a processor with the x86 instruction set that is also In Order and no Speculative Execution, we just don't do that today.

 

As for who did Speculative Execution first, no idea and not really the fussed either. Branch prediction is the area you'll want to be looking at to figure out who did it first but it's likely SPARC or MIPS (RISC).

[Guest]

On ‎4‎/‎3‎/‎2019 at 2:31 PM, VegetableStu said:

Vega 64/56 and Vega V2 are technically high end cards ._. considering they don't squeeze that many CUs for the middle stack

Vega cards have dissapeared from shelfs everywhere  these last couple of months

 

[Stefan Payne]

4 minutes ago, leadeater said:

So the first x86 processor is either than 8086 or 80186. You can have a processor with the x86 instruction set that is also In Order and no Speculative Execution, we just don't do that today.

Just for reference/Information:
The last in order x86 CPU (Series) was the Atom 2000 ones.

For example Intel Atom Z2760

4 minutes ago, leadeater said:

As for who did Speculative Execution first, no idea and not really the fussed either. Branch prediction is the area you'll want to be looking at to figure out who did it first but it's likely SPARC or MIPS (RISC).

Yeah, that's what I think as well, though I'd not rule out IBM with their Power Architecture and DEC's Alpha Architecture...

 

 

[Stefan Payne]

18 minutes ago, Christiaan21-03 said:

Vega cards have dissapeared from shelfs everywhere  these last couple of months

Meaning that they get ready for NAVI. That is not an uncommon thing to do to get rid of old stock before the new one arrives....

 

And I'd expect them to clear the shelves of "old stock"...

 

Also Mindfactory hadn't had a bigger amount of AMD Cards (besides Radeon 7) in their Mindstar in days.
HOWEVER, the regular Price of a Powercolor RX VEGA 56 Red Dragon (with the small PCB) is less than 300€

MSI Blower Card (wich sucks) is 286. An ASUS RX VEGA 64 is in stock for 388€ Gigabyte 360, MSI VEGA64 Reference 389€...

 

So not entirely true...

But Prices look good.

And especially the Powercolor RX VEGA56 seems to be more interesting than an RX590.

 

 

[Guest]

2 minutes ago, Stefan Payne said:

Meaning that they get ready for NAVI. That is not an uncommon thing to do to get rid of old stock before the new one arrives.

 

 

That might mean that NAVI is rather close.

 

Also Mindfactory hadn't had a bigger amount of AMD Cards (besides Radeon 7) in their Mindstar in days.
HOWEVER, the regular Price of a Powercolor RX VEGA 56 Red Dragon (with the small PCB) is less than 300€

MSI Blower Card (wich sucks) is 286. An ASUS RX VEGA 64 is in stock for 388€ Gigabyte 360, MSI VEGA64 Reference 389€...

 

So not entirely true...

But Prices look good.

And especially the Powercolor RX VEGA56 seems to be more interesting than an RX590.

 

 

can't wait how the new navi cards will deliver in both price range and tpd or tbp …. have saved meself already 500 quid.. got the manufacterer down to either aorus or rog strix… what can go wrong in these next months.. finally i will able to see for myself if developers these days really only do a.i. stuff correct with enough gb's on the card