AMD Presents New Horizon

[kac77]

Heh...

I'm just going to tell you up front you have no idea what you are talking about. IPC (or MIPS) are absolute.

Let's review your question exactly as you put it.

Ok.. so how many instructions per clock cycle does bulldozer do?

That's so damn general you could drive a bus through it.

It was a rhetorical question because people throw around IPC without knowing what it is.

And I'm just going to tell you that you are trying to be facetious but don't know how to do it correctly, or were you being dishonest when you knew an answer? Essentially you are being too cute by half and hoping no one realizes it. What you are attempting to argue is that IPC is a strictly defined definition to mean only one thing and one thing only (max throughput and in your case any instruction type). This is not the case from a programmatic point of view, or real world workloads (this right here is why we test with them to begin with).

But in cause you are wondering, Bulldozer's integer instructions per clock cycle is 8.

Sure in a perfect world ( I seem to remember it being 4 though). If you were correct with your assertion with regards to software then no matter what compiler tuning was given upon compile we would always get the same result. We don't not even close. You are attempting to conflate IPC from max throughput of an architecture, and then go on to claim that it can be universally applied to every workload without taking into account, branch mispredicts, pipeline stalls, latency, bitness, or cache size. Sure it's 8 instructions per cycle but it isn't with shitty code, or code meant for another architecture. This is just not going to fly with anyone. I mean it's kind of cute, but demonstrably ignorant. The whole point of providing you with that link is to show that actual software that people use can affect any architecture's capability to achieve max throughput, especially when it came to BD.

What software is involved is meaningless.

No one believes that except you...like literally. Any chip without software is a paper weight. It's not processing a damn thing without it, nor will it understand how to process it without it.

This is a hardware specification which has software *implications*.

This much is very true. But that's not what you are arguing here. You also fail to realize that shitty code can have implications regardless of what should be possible.

"Ricin" on the other hand has 16 instructions per clock cycle. Like the Intel chip used in the AMD demo.

Vague, but I thought you didn't know at all? Whoops. Ulterior motive alert!

All you had to do was know what you were talking about- no Google needed

Ignorance is bliss as they say. However, you might want to read some in depth analysis of any CPU architecture because all of it reviews max throughput against all of the different scenarios I listed above. It would do you some good that way when you figure you are going to be too cute by half with anyone maybe just maybe you will think twice.

 

[billabong132]

The performance scores seem to be scaling almost linearly between the different sampling settings. This means that you get a fairly accurate 150 samples score if you multiply a 200 score by 0.75, or a 100 score by 1.5. Here's a post from another thread:



64.26 x 0.75 = 48.195
32.75 x 1.5 = 49.125

< 1 second is within margin of error. Hope this helps you compare the scores already posted.

I'll need to see some residual analysis before I accept your claim of a linear relationship

 

[aztekk]

I'll need to see some residual analysis before I accept your claim of a linear relationship

Haha I was anticipating a smart-ass response, thus I used words "seem" and "almost". If you want to read through 100 results amongst random bs posted in various threads, you are more than welcome to disprove my "claim" with your own analysis sir

 

[noko]

Other factors also affects the outcome of this render - as in render resolution. AMD should just be very clear on what settings (all of them) used. Such as default but samples 150 etc. It could have been samples 100 and resolution of 1024x768 vice 800x600.

Now by the way, am I the only one here that gets turned on by Su??? I think she did a fairly good job, not perfect but much better then anything before.

 

[ikevi]

I'll need to see some residual analysis before I accept your claim of a linear relationship

It seems to be really close to linear.

Just running it on my computer for 100,150 and 200. (No OC.) Windows 10 pro

 

[aztekk]

It seems to be really close to linear.

Just running it on my computer for 100,150 and 200. (No OC.) Windows 10 pro

View attachment 12666

He will complain about low sample size next. (he would be right, technically, we only have test results from 2 machines here)

 

[ikevi]

He will complain about low sample size next. (he would be right, technically, we only have test results from 2 machines here)

True, I could bother to test this out on 5 or so more computers, but meh there is really no benefit from any of this, besides having a bit of fun seeing what our computers can do.



A few side notes to folks.

If you think Intel is holding something back you are living in a pipedream. You know that if they they had a process in their back pocket that was so good it would convince all their customers to upgrade on a 2 year cadence again they would have been using it. (Not saying they aren't working on more things, just saying it makes no financial sense to not release a better fully developed process if it is ready...) Everyone has to realize that the current products out there are still completely focused on the mobile space. (And looking at their stock/earnings they seem to be doing ok with that...)

As already said by folks. The demos at least show that AMD has a potentially good chip on their hand. How good, we don't know yet. BUT they have managed to make it so that everyone in the know is at least going to be keeping 1 eye out for move info/full reviews, and that likely is what they would consider a success for this announcement.

 

[blkt]

I other words...

"Shit, kyles asking more questions, he may write something we don't like! Quick don't respond!"

The irony is they should be more worried about a lack of response.

 

[Shintai]

First it depends on workload. No two programs will perform the same. You give BD a streaming workload it's competitive. If you don't well... lets just say it's underwhelming and leave it at that.

However, many people have done tests. All ya had to do was use Google.

Here ya go.

People forget how far behind AMD is on IPC, even with all the numbers adding up. You could add 100% to PD for example and still be way behind.

 

[Shintai]

Found the link to above mentioned reddit post:

The Blender file is still the same and unchanged. And we all know AMD used 2.77, most likely with custom libs.

So its a flat out lie.

150 samples wont do either to fit the results.

 

[Shintai]

True, I could bother to test this out on 5 or so more computers, but meh there is really no benefit from any of this, besides having a bit of fun seeing what our computers can do.



A few side notes to folks.

If you think Intel is holding something back you are living in a pipedream. You know that if they they had a process in their back pocket that was so good it would convince all their customers to upgrade on a 2 year cadence again they would have been using it. (Not saying they aren't working on more things, just saying it makes no financial sense to not release a better fully developed process if it is ready...) Everyone has to realize that the current products out there are still completely focused on the mobile space. (And looking at their stock/earnings they seem to be doing ok with that...)

As already said by folks. The demos at least show that AMD has a potentially good chip on their hand. How good, we don't know yet. BUT they have managed to make it so that everyone in the know is at least going to be keeping 1 eye out for move info/full reviews, and that likely is what they would consider a success for this announcement.

Its not long ago people was dreaming of Polaris being a Titan Killer after...their careful crafted event to show just that.

Whatever you see demonstrated will be the absolute best cases. The BF1 60 FPS 4K could have been done with an FX for example. The Blender bench is a very light IPC load too. Also why you see the 6900K use to little power. Or if you test it yourself with a 6700K you see it being a 65W chip running it etc.

 

[blkt]

While better than last, AMD really need to find a person that can do great presentations. Lisa Su will never be one that does.

JHH or Panos Panay level is something they need.

So...put Lisa on a weight training program, plaster a tattoo of AMD on her shoulder and top it off with jeans/spiffy leather jacket?

 

[Shintai]

So...put Lisa on a weight training program, plaster a tattoo of AMD on her shoulder and top it off with jeans/spiffy leather jacket?

She doesn't have the talent or voice. Panos for example I think is one of the absolute best.

It would help them a great deal if they had a single skilled person with the charisma doing these presentations. Instead of one PR disaster after the other like Roy, Raja, Lisa etc doing it.

 

[blkt]

She doesn't have the talent and voice. Panos for example I think is one of the absolute best.

It would help them a great deal if they had a single skilled person with the charisma doing these presentations. Instead of one PR disaster after the other like Roy, Raja, Lisa etc doing it.

Where is your sense of humor? I sort of agree, but I still think an engineer is best for presentation. First to blame, for me, would be what she isn't allowed to say or, more importantly, the presentation planning which puts her in such an awkward position. They can't keep Jim Keller around for long and I doubt he would want to be on stage anyway. This architecture will have numerous applications for AMD and the fact that Jim was involved in the design of Zen gives me all the confirmation I need to know AMD is serious this time around. Their marketing department however...sigh.

 

[NKD]

Its not long ago people was dreaming of Polaris being a Titan Killer after...their careful crafted event to show just that.

Whatever you see demonstrated will be the absolute best cases. The BF1 60 FPS 4K could have been done with an FX for example. The Blender bench is a very light IPC load too. Also why you see the 6900K use to little power. Or if you test it yourself with a 6700K you see it being a 65W chip running it etc.

I crown you the biggest AMD hater I have ever seriously known. There is absolutely no reply or nothing that will meet your unreasonable demands. I said this before its just sample size. What conspiracy theory are you cooking now? 100 sample size was in the 20s and people have been getting similar scores there. So only makes sense sample size it at play here. There is no frickin conspiracy theory here, you really think Lisa is that fuckin stupid that she tells you to go home and try it and knows that she could get away with it and people won't try to match those results. Unreal man!

Now could it be a mystery that the dude that ran the bench on the stage probably ran his own sample size? yea very likely he clicked a few buttons lol. I think you are pulling at straws just about now.

 

[Shintai]

Where is your sense of humor? I sort of agree, but I still think an engineer is best for presentation. First to blame, for me, would be what she isn't allowed to say or, more importantly, the presentation planning which puts her in such an awkward position. They can't keep Jim Keller around for long and I doubt he would want to be on stage anyway. This architecture will have numerous applications for AMD and the fact that Jim was involved in the design of Zen gives me all the confirmation I need to know AMD is serious this time around. Their marketing department however...sigh.

You put way too much into Jim Keller to put it mildly. That guy is more glamour than substance. If Zen is a success, Keller is the last one you should credit that to.

 

[Shintai]

I crown you the biggest AMD hater I have ever seriously known. There is absolutely no reply or nothing that will meet your unreasonable demands. I said this before its just sample size. What conspiracy theory are you cooking now? 100 sample size was in the 20s and people have been getting similar scores there. So only makes sense sample size it at play here. There is no frickin conspiracy theory here, you really think Lisa is that fuckin stupid that she tells you to go home and try it and knows that she could get away with it and people won't try to match those results. Unreal man!

Now could it be a mystery that the dude that ran the bench on the stage probably ran his own sample size? yea very likely. I think you are pulling at straws just about now.

The problem came when AMD invited us all to do the bench. Had they never done that we wouldn't have had this discussion. AMD uses a different Blender version and unknown settings we already determined isn't the default.

So wrong one to blame your anger on. You have to do that to the person inviting all of us to do it.

 

[NKD]

You put way too much into Jim Keller to put it mildly. That guy is more glamour than substance. If Zen is a success, Keller is the last one you should credit that to.

LoL why the hate on keller, so you are telling me he has accomplished nothing? I doubt Apple pulled him over to their side for no reason, so you are going to tell me APPLE didn't have kick ass processor designs after he went there. There A series processors are nothing to sneeze at now I know I can't give all the credit to him, but the guy has credibility or he wouldn't be in so much demand.

 

[Shintai]

LoL why the hate on keller, so you are telling me he has accomplished nothing? I doubt Apple pulled him over to their side for no reason, so you are going to tell me APPLE didn't have kick ass processor designs after he went there. There A series processors are nothing to sneeze at now I know I can't give all the credit to him, but the guy has credibility or he wouldn't be in so much demand.

There are 1000 of people on a CPU project. But without Keller they are just all useless people? Keller got more credit than he deserves. He was at Apple with what, A4? He didn't do K8 as credited. It was Fred Weber. He ragequit AMD long before Zen was done.

You have to thank all these unknown people instead. They are the ones doing the heavy lifting.

Can you even tell me who is in charge of Zen at AMD?

 

[NKD]

The problem came when AMD invited us all to do the bench. Had they never done that we wouldn't have had this discussion. AMD uses a different Blender version and unknown settings we already determined isn't the default.

So wrong one to blame your anger on. You have to do that to the person inviting all of us to do it.

They just uploaded one of the sample images, and they invited people to render it. Obviously they didn't load the same one as the benches. I don;t think its anything to say oh they are trying to hide something. I am sure we will see people testing this at 150 samples, confident it will get close to the benches they ran, just like the 100 sample results were similar.

 

[Shintai]

They just uploaded one of the sample images, and they invited people to render it. Obviously they didn't load the same one as the benches. I don;t think its anything to say oh they are trying to hide something. I am sure we will see people testing this at 150 samples, confident it will get close to the benches they ran, just like the 100 sample image were similar.

The 100 samples wasn't similar at all and neither is 150. People also tried 128 for that matter.

 

[Shintai]

It seems to be really close to linear.

Just running it on my computer for 100,150 and 200. (No OC.) Windows 10 pro

View attachment 12666

Note it runs at 3.3Ghz with 6-8 cores, specially in a light load as Blender.
http://www.intel.com/content/www/us/en/support/processors/000006652.html

 

[ikevi]

Note it runs at 3.3Ghz with 6-8 cores, specially in a light load as Blender.
http://www.intel.com/content/www/us/en/support/processors/000006652.html

Yeah I know. I was just saying I had it at default settings. I haven't played with this computer at all, even if it is water cooled ;-)

 

[NKD]

There are 1000 of people on a CPU project. But without Keller they are just all useless people? Keller got more credit than he deserves. He was at Apple with what, A4? He didn't do K8 as credited. It was Fred Weber. He ragequit AMD long before Zen was done.

The 100 samples wasn't similar at all and neither is 150. People also tried 128 for that matter.

Not at all? I saw people getting scores in the 20s. Isn't that what amd got?

 

[blkt]

You put way too much into Jim Keller to put it mildly. That guy is more glamour than substance. If Zen is a success, Keller is the last one you should credit that to.

So, it is simply a coincidence the Athlon K7 and later the A8 (hello AMD64) were so great for AMD and within a few years of Keller's departure, on each, the rapid progress turned into a trickle?

There are 1000 of people on a CPU project. But without Keller they are just all useless people? Keller got more credit than he deserves. He was at Apple with what, A4? He didn't do K8 as credited. It was Fred Weber. He ragequit AMD long before Zen was done.

You have to thank all these unknown people instead. They are the ones doing the heavy lifting.

Can you even tell me who is in charge of Zen at AMD?

A4 and A5. Keller was often noted for crediting his team. Maybe he was a team player, a leader and knew how to get the ball rolling?. What is this ragequit you speak of?

 

[Shintai]

So, it is simply a coincidence the Athlon K7 and later the A8 (hello AMD64) were so great for AMD and within a few years of Keller's departure, on each, the rapid progress turned into a trickle?A4 and A5. Keller was often noted for crediting his team. Maybe he was a team player, a leader and knew how to get the ball rolling?. What is this ragequit you speak of?

K7 he was just part of the team.
K8 was Fred Weber, not Jim Keller.
Zen is Suzanne Plummer, not Jim Keller.
Apples most successful chips are not designed by Jim Keller. And even with the A4 he was just part of the team.

Also to glorify 1 manger type of person in a sea of 1000+ PhDs. is...incredible degrading for those people.

 

[Shintai]

Not at all? I saw people getting scores in the 20s. Isn't that what amd got?

You should check both the press and live demo again. Then look up scores and you want something that is rather precise not with a 10-30% variance at best.

 

[blkt]

K8 was Fred Weber, not Jim Keller.
Zen is Suzanne Plummer, not Jim Keller.
Apples most successful chips are not designed by Jim Keller. And even with the A4 he was just part of the team.

Also to glorify 1 manger type of person in a sea of 1000+ PhDs. is...incredible degrading for those people.

You didn't answer my question. I cannot find any information about Keller leaving the company on bad terms, or leaving his part in the Zen project unfinished (or anything regarding the project being offset or delayed for that matter). However, you are right in the sense that Jim left AMD before K8 was finalized. Fred Weber took over the K8 program and led the final design. Jim is notorious for switching companies as soon as he can to work on new projects.

 

[Shintai]

You didn't answer my question. I cannot find any information about Keller leaving the company on bad terms, or leaving his part in the Zen project unfinished (or anything regarding the project being offset or delayed for that matter).

When you quit with no warning and there is no replacement planned at all. Its not because everything was cherry fine. Not to mention the date, the 18th in a month. CTO Mark Papermaster had to step in.

There has been no K12 news since Keller left as a note.

 

[razor1]

Yeah K12 ARM was a telling sign, something went very wrong with it, and AMD stopped talking about the the moment Keller left. Don't spend millions on a CPU and just stop talking about it right after the lead designer leaves.

 

[razor1]

Not at all? I saw people getting scores in the 20s. Isn't that what amd got?

There has been a lot of people trying to replicate the scores shown in the live demo, its not double with just sample size reduction.

 

[ecmaster76]

Yeah K12 ARM was a telling sign, something went very wrong with it, and AMD stopped talking about the the moment Keller left. Don't spend millions on a CPU and just stop talking about it right after the lead designer leaves.

Perhaps they realized there wasn't a strong enough market demand for a high end ARM server part and canned the project. Perhaps Keller left in frustration or because he already had the high level design for Zen finalized and wanted to move on.

 

[Shintai]

Perhaps they realized there wasn't a strong enough market demand for a high end ARM server part and canned the project. Perhaps Keller left in frustration or because he already had the high level design for Zen finalized and wanted to move on.

Last part would imply a planned scheduled leave.

 

[razor1]

Perhaps they realized there wasn't a strong enough market demand for a high end ARM server part and canned the project. Perhaps Keller left in frustration or because he already had the high level design for Zen finalized and wanted to move on.

Since the AM4 platform would be the basis of such a market, they were expecting it to help them take on Intel in the server market where linux is used more than Windows. And ARM servers are supposed to be growing business.

I don't want to derail the thread of keller, but as good as Keller is he is only one man, and Keller seems like alot of hype for AMD, at least the way they portrayed him, like the second coming of Christ.

 

[MrCaffeineX]

Yeah, unfortunately the tech sites you can actually trust just keep dwindling.

For me its HardOCP first, and Anandtech a distant second. Even with Anand no longer at the helm, they seem to be maintaining good editorial standards.

As long as you're not expecting a timely GPU review from them...

 

[Gideon]

That is what Jim Keller does, takes on a big new project and then he leaves when it's done or his part of the chip is done. Nothing new about that. Also certain people on this forum will always spin things for AMD in a negative light. For some of us that have been around for awhile it's like a big broken record on this forum when it comes to AMD, tho it will be nice to see them being at a competitive level with Intel again. If they surpass Intel again like they did with the Athlon, then expect even more hate, unlikely with Zen but possible with Zen+.

 

[ManofGod]

So, it is simply a coincidence the Athlon K7 and later the A8 (hello AMD64) were so great for AMD and within a few years of Keller's departure, on each, the rapid progress turned into a trickle?A4 and A5. Keller was often noted for crediting his team. Maybe he was a team player, a leader and knew how to get the ball rolling?. What is this ragequit you speak of?

The rage quit is entirely in his mind. One person really can make a significant difference, especially when there was no one else to fill those shoes.

 

[serious]

Eventually, reviewers will get their hands on Ryzen, and if they can't replicate the result (not referring to the actual time those setups take on the test, but the difference in performance between 2 hardware setup), then AMD will not need to worry about product promotion as every single tech site is going to have their spin on it. Su said:

• Cyzen's performance is on par with Intel 6900k (at least with some selected tasks).
• Cyzen use less electricity (more efficient).

I don't think there are gray areas here, either true or false. I believe reviewers have already planned how to test this exact point Su tried to demonstrate already, so this presentation simply set up the stage for the actual reviews that will be done by 3rd parties, which is really good. The courage of doing something like this alone deserve my vote from my wallet, unless this isn't courage but pure stupidity.

 

[Shintai]

That is what Jim Keller does, takes on a big new project and then he leaves when it's done or his part of the chip is done. Nothing new about that.

Please document this.

Comments like that is directly respect less to everyone who works on Zen. What would all those peasants be without the glorious Keller? Anyway, wrong person. The person in charge of Zen is Suzanne Plummer if you missed it.

 

[Gideon]

Please document this.

Comments like that is directly respect less to everyone who works on Zen. What would all those peasants be without the glorious Keller? Anyway, wrong person. The person in charge of Zen is Suzanne Plummer if you missed it.

http://venturebeat.com/2012/08/01/apple-cpu-jim-keller-amd/

http://www.anandtech.com/show/9643/jim-keller-leaves-amd