AMD going 20nm this year?

[mjw21a]

Doesn't AMD have MMX, SSE, SSE2, SSE3, SSE4a,3DNOW?

Yes, however any software compiled using the Intel compiler will not use those extensions when running on a non Intel CPU. Thats what this is about, it's old news really.

The down side is that most dev's use the Intel compiler.

 

[Bubba-Hotepp]

Yes, however any software compiled using the Intel compiler will not use those extensions when running on a non Intel CPU. Thats what this is about, it's old news really.

The down side is that most dev's use the Intel compiler.

I don't know if I would say "most" (don't have enough info to declare that) but it's at the very least a significant amount and definately enough to skew results and CPU performance impressions.

Currently I'm trying to figure out how to edit a .vmx file so that I can change the CPUID in a VMWare Virtual machine to show up as intel so that I can test all the popular benchmarks. If anyone has any info let me know. What I've tried isn't working so far.

 

[Frakk]

I don't know if I would say "most" (don't have enough info to declare that) but it's at the very least a significant amount and definately enough to skew results and CPU performance impressions.

Currently I'm trying to figure out how to edit a .vmx file so that I can change the CPUID in a VMWare Virtual machine to show up as intel so that I can test all the popular benchmarks. If anyone has any info let me know. What I've tried isn't working so far.

Yeah, me too.... when i get back home in a day or so. the machine i'm using right now is an Intel so its not going to work

If you get it working before i do let me know, we can compare results, see how it affects K10 and K11, i think i may have an old 65nm K8 stuffed in a cupboard somewhere too...

 

[Bubba-Hotepp]

Yeah, me too.... when i get back home in a day or so. the machine i'm using right now is an Intel so its not going to work

If you get it working before i do let me know, we can compare results, see how it affects K10 and K11, i think i may have an old 65nm K8 stuffed in a cupboard somewhere too...

With yours if you're able to change the CPUID for the virtual machine to AMD you could see if the performance drops (opposite of what I'm trying to do )

 

[Frakk]

With yours if you're able to change the CPUID for the virtual machine to AMD you could see if the performance drops (opposite of what I'm trying to do )

 

[Bubba-Hotepp]

Haven't figured it out yet. I might just have to pick up a via nano board and use the program that agner wrote that changes the CPUID to "genuineIntel" or "AuthenticAMD".

 

[Frakk]

Haven't figured it out yet. I might just have to pick up a via nano board and use the program that agner wrote that changes the CPUID to "genuineIntel" or "AuthenticAMD".

Yeah, i don't think it will work without some serious interference with the OS to make 'it' think its an Intel.

I'm also going to look again at agner progy, but not on my beloved Thuban.

I have been into the cupboards and up turned a few boxes that have been accumulating over the years... i pleasantly surprised myself, here is what i manged to throw together...

MOBO: ASRock K10N78, CPU: AMD Athlon x2 5200+ / Cooler: Some unknown top down 4 Heat pipe HS with a red xilence PWM fan
RAM: Hynix DDR2-800, 2x 1GB
HDD: Seagate 250GB ATA
Case: CM Elite 330
An IDE- LG DVD RW.
And a not at all shabby OCZ600SXS2 PSU which looks brand new. it has 4x 18A 12v rails, which is probably why i never used it

I also managed to get 2 GF 8500GT's going but i can't find an SLI bridge
Also, i never kept any IDE cables so i don't have any, so i can't install a Windows OS on it, tho i did find some broken IDE to SATA converters, i might get one going if i'm lucky

Then i can have some fun with this beast

 

[Bubba-Hotepp]

Yeah, i don't think it will work without some serious interference with the OS to make 'it' think its an Intel.

I'm also going to look again at agner progy, but not on my beloved Thuban.

I have been into the cupboards and up turned a few boxes that have been accumulating over the years... i pleasantly surprised myself, here is what i manged to throw together...

MOBO: ASRock K10N78, CPU: AMD Athlon x2 5200+ / Cooler: Some unknown top down 4 Heat pipe HS with a red xilence PWM fan
RAM: Hynix DDR2-800, 2x 1GB
HDD: Seagate 250GB ATA
Case: CM Elite 330
An IDE- LG DVD RW.
And a not at all shabby OCZ600SXS2 PSU which looks brand new. it has 4x 18A 12v rails, which is probably why i never used it

I also managed to get 2 GF 8500GT's going but i can't find an SLI bridge
Also, i never kept any IDE cables so i don't have any, so i can't install a Windows OS on it, tho i did find some broken IDE to SATA converters, i might get one going if i'm lucky

Then i can have some fun with this beast

His program only works on Via CPU's. Since WinXP only fully supported Intel and AMD, Via worked in a "cheat" which allows the chips CPUID to be manipulated.

 

[mjw21a]

I'd not worry about it.... I'd have looked at playing arounnd with such things when I was in my mid 20's...... Right now, I've not even turned on my main (signiture) rig in over 2 weeks. My HTPC gets heavy use though

I just can't be bothered playing around with my computer at the moment.... I'm guessing that will pass in time..... I'm looking forward to playing with Piledriver though I suspect that will only be a week or so of tinkering till I'm bored with it.....

 

[Neuromancer]

Thought 2014 was the earliest for Amd to get to 20nm

Prototype 2012 production 2013.

14nm prototype 2013 production 2014.

Thats their roadmap at least. Based on Bulldozer being about 3 years late? I would suspect that means we will see 20nm in 2016. and 14nm in 2018.

 

[mjw21a]

Depends on what you call late. I was reading one of Ed Stroligo's old articles the other day (nostalgia) from 2008 where he references Bulldozer won't be out untill 2011 at the earliest. Well you know, they did actually manage that..... It wasn't so much late as that it didn't meet peoples expectations.

Kind of like going to a movie and expecting it to be awesome, but then you're disappointed. Whereas you can sometimes go into a movie expecting it to be crap, and end up really enjoying it where the next person hated it because they expected too much. Whatever guys, it's the first gen of a new architecture from a company on a limited budget. Keep it real.

 

[Frakk]

That ^^

And they seem to be managing Trinity on time.

 

[mjw21a]

Agreed, I for one will be replacing my FX8120 once Piledriver chips are available. I also hope they add another module, despite the fact that absolutely nothing will use it..... Yea, I like seeing all those cores on my desktop widget and I want 10 cores dammit! 12 would be even cooler again

I'd get absolutely zero noticeable performance boost in anything I do getting this, but I still want it

 

[Neuromancer]

Depends on what you call late. I was reading one of Ed Stroligo's old articles the other day (nostalgia) from 2008 where he references Bulldozer won't be out untill 2011 at the earliest. Well you know, they did actually manage that..... It wasn't so much late as that it didn't meet peoples expectations.

2008 they were talking about 2009 release. I was reading through eds articles yesterday too when I saw the thread bumped

Later in 2008 they might have revised that or Ed just had a better view point than other news organizations did (which is highly plausible).

I looked that time frame up before I posted the 3 years late because I remember hearing about BD and 128bit FPUs and quad channel memory WAY WAY back. And it seemed like it was constantly wait for it, wait for it... maybe next year, wait for it ... maybe the following the year.

It is possible Ed had more insight then was expected but what I found was Bulldozer samples on 45nm to be produced in 2009. Maybe 3 years was a stretch but still it was late. I will not argue the performance aspect of BD, I am not sold it is not a better platform than Thuban. Once I get my BD into 24/7 mode I will know more


EDIT Looks like ARM will be doing 20nm 2H this year

 

[mjw21a]

Lol, Ed did have more insight than other people in the media did back then and unfortunately nobody else in the media has that insight today. I suspect it will be a combination of factors. One would be someone heard that it would be earlier and the rumor spread due to people wanting to believe (the usual basically). It's quite possible due to that rumour that the then current head of AMD (Hector Ruiz) would have pandered to the press to try and make AMD look good.

You have to remember that Hector was all pretty much hot air and overall was very bad for AMD. Lots of talk and no action under his reign. The only things that went right were at the begining of his reign, basically work done when Jerry Sanders was running things still. Every CEO they've had since then has slowly improved things from the mess he left behind.

Likewise, getting rid of their manufacturing capabillity appears to be the best thing they could have done. It injects more money into manufacturing plants, who also work harder to gain the business of clients. The move to smaller process nodes with AMD is now happening faster than they have ever done so before, and almost entirely because they can now change between manufacturers.

 

[Bubba-Hotepp]

Lol, Ed did have more insight than other people in the media did back then and unfortunately nobody else in the media has that insight today. I suspect it will be a combination of factors. One would be someone heard that it would be earlier and the rumor spread due to people wanting to believe (the usual basically). It's quite possible due to that rumour that the then current head of AMD (Hector Ruiz) would have pandered to the press to try and make AMD look good.

You have to remember that Hector was all pretty much hot air and overall was very bad for AMD. Lots of talk and no action under his reign. The only things that went right were at the begining of his reign, basically work done when Jerry Sanders was running things still. Every CEO they've had since then has slowly improved things from the mess he left behind.

Likewise, getting rid of their manufacturing capabillity appears to be the best thing they could have done. It injects more money into manufacturing plants, who also work harder to gain the business of clients. The move to smaller process nodes with AMD is now happening faster than they have ever done so before, and almost entirely because they can now change between manufacturers.

Oh, don't get us started on H. Ruiz. He squandered a tech lead (athlon64) and gutted AMD while running away with his pockets lined with cash *cough* global foundries *cough*.

 

[mjw21a]

Yeah, he single handledly destroyed the future of a great up and coming tech company. Not to mention insider trading and all that other crap he was involved in. It's going to take AMD years just to get to performance partiy with Intel again now.

 

[Neuromancer]

I agree with Ed's insight comment, I actually said the same thing but that line got deleted when I edited my post down from novella format


So 20nm next year you still think a possibility then? AMD has been plagued with rumours/hopes/suppositions in the past. If anything, I would think 20nm GPU followed by 20nm APUs and then MAYBE assuming both of those are in the works by the end of '13, we MIGHT see a, uh... "steamroller" based on 20nm BD cores.

In all honesty there are a few things I would like to see changed before worrying about improving CPU performance anymore at this stage, like someone creating a successful alternative to ATX design mobos and cases.

Laptops and handhelds and such use connections on either side of a motherboard.. Would like to see that happen with desktops too. think of the possibilities!!! It would require mobo makers AND case makers to get on the same page though.

Don't think we will see if for a while though, BTX never took off (why would it since it offered no advantage over rATX?)

Sorry got OT there.


EDIT: Oh yeah... back on failed promises from AMD in the HR days.. I wanted reverse hyperthreading! (taking a single execution thread and instead of pounding away at it for say 6 minutes on a single core, break it up across 4 cores and do it in half (or less!) of the time...)

 

[Bubba-Hotepp]

I agree with Ed's insight comment, I actually said the same thing but that line got deleted when I edited my post down from novella format


So 20nm next year you still think a possibility then? AMD has been plagued with rumours/hopes/suppositions in the past. If anything, I would think 20nm GPU followed by 20nm APUs and then MAYBE assuming both of those are in the works by the end of '13, we MIGHT see a, uh... "steamroller" based on 20nm BD cores.

In all honesty there are a few things I would like to see changed before worrying about improving CPU performance anymore at this stage, like someone creating a successful alternative to ATX design mobos and cases.

Laptops and handhelds and such use connections on either side of a motherboard.. Would like to see that happen with desktops too. think of the possibilities!!! It would require mobo makers AND case makers to get on the same page though.

Don't think we will see if for a while though, BTX never took off (why would it since it offered no advantage over rATX?)

Sorry got OT there.


EDIT: Oh yeah... back on failed promises from AMD in the HR days.. I wanted reverse hyperthreading! (taking a single execution thread and instead of pounding away at it for say 6 minutes on a single core, break it up across 4 cores and do it in half (or less!) of the time...)

Here's a little food for thought. Process shrink IMHO is not the "fix all" that everyone thinks it is. There's this presumption that by shrinking down to 20nm that performance will go up. That's not necessarily true. There needs to be a simultaneous design change as well. Now add in the fact that as we hit 20nm and lower, without changes to the materials used (i.e. silicon) it becomes less and less profitable to manufacture chips and CPU prices will go higher and higher.

 

[mjw21a]

Yeah, I think we'll see 20nm next year. The thing to remember though is that APU's are AMD's priority and everything else comes after that. I actually agree that APU's should come first, as once code is written to make use of the GPU portion I believe we'll see performance parity with Intel in many things. Things will only slant in AMD's favour if this is indeed the way things are going. I expect Intel to remain ahead in integer performance though.

No, the main advantage of process shrink now is power use and also cost of manufacturing a chip. AMD presently has to manage competitive designs using the lowest number of transistors possible due to being behind in process tech. If they can come close to parity on manufacturing I think we'll see them quite a lot closer to Intels performance levels, actually ahead on those tasks using GPU