Discussion: AMD Ryzen

[juanrga]

Like most people interested in technology I love the info leaks as much as the next person, but you also have to know where those leaks are coming from. Often times it is an unscrupulous technician who is perhaps unhappy with his current pay. An engineer worth his salt will not risk losing their job, and perhaps millions of dollars over a lifetime career, by breaking NDAs or other confidentiality agreements.

I typically go with who has been more accurate over time so far things seem to be adding up and conclusions are starting to be made after Zen releases or they at least do a paper launch we will know. The fact that they underclock the Intel sample is concerning why not just use the same clock speed? Why not tell us the TDP? I personally never trust a company when they are to quiet about a launch. I still remember when GCN came out Amd was in a major hurry to show it off(cause it was great). Now i don't see it. I just hope they don't delay Zen until this time next year waiting for Global foundry to create some magical die.

I smell that "magical die" is not coming... AMD just announced a re-negotiation of the WSA

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=2198716

Interesting that people scream AMD is dying, but the Dubai firm that owns GF has included an option to purchase AMD stock up to 19.99% of outstanding shares in the Wafer agreement.

Not to put too fine a point on it...but that does not seem like a firm concerned about the future of that company.

The post was about Globalfoundries issues, and everyone knows how well Mubadala runs the semiconductor business: multiple delays, cancellation of nodes, process never matching expectations/promises,.. Not only AMD suffers, recall what happened to Apple.

Globalfoundries is so good in the business, that their fabs are almost empty and have to force AMD (via the WSA) to fabricate chips with them; otherwise AMD would choose other foundries. If you read the official announcement. AMD has accepted to pay hundred millions to Globalfoundries for fabricating chips on other foundries like TSMC and Samsung. AMD wouldn't do that if Globalfoundries process were so good. And having to fabricate in other foundries will generate problems for AMD. The first? Paying twice. The second? Healthy companies as Apple will pay a plus to TSMC and Samsung get early access to new process nodes leaving AMD in the queue.

You must not have read what I said...

More specifically:

Mubadala included an option for Mubadala to purchase up to 19.99% of outstanding shares of AMD stock.

Not AMD purchasing GF stock.

I will say it once again: Mubadala is known by not taking wise decisions.

 

[juanrga]

Yeah. And at this point with the current slowing it looks like that 10nm and 7nm will be hit at the same time for everyone. It's actually plausible that GloFo hits 7nm before Intel because they're skipping 10nm.

But Globalfoundries '7nm' will be as Intel 10nm or inferior in reality. And it is difficult to believe that Glofo 7nm is comming first, considering that Glofo 14nm is comming with many delays, even when they only had to license the process from Samsung.

 

[-Fran-]

There is no evidence that either way was better, again it is my opinion that it has not benefitted them and they would have been better off.

And I don't think Intel will stop. I think they will open up their FABs to foundry like work and they already have, again LG has signed to FAB 10nm ARM chips already. If Intel starts to pull more business that could pay for the R&D for process nodes then their server and HPC profits will become freed up for other uses.

I agree with you that the current situation with GloFo is not helping AMD at all. Also- money taken out of the equation, AMD would be better fabbing their own chips as you say (Intel is proof of that, and back in the day the main advantage AMD had was *better fab and 'to the metal' design*- the AMD 286, AMD 386 and AMD 486 chips were all faster than the Intel parts they were based on thanks to AMD's tweaks to the layout and fab process 😛

The issue was that AMD *were* doing well- and pushing forward with their fabs, but then they were late with Phenom I, the process wasn't good enough, the chip itself had problems and their revenue fell off a cliff in the face of a resurgent Intel with the superior Core 2 (even if I personally think Phenom was a much more elegant design- the benchmarks heavily favored Core). At that point they had to cut back on something- and as the process / manufacturing tech is by far the most expensive that was the obvious choice (if you want to get a comparison to costs of running fab vs fabless just look at ARM). They *had* to sell it off, there was no choice. I honestly believe had they attempted to keep Glofo as part of AMD, they would have gone bankrupt.

Indeed. Without furthering the point of GF sucking, AMD was doing fine at some point fabricating their own CPUs and they just blew the opportunity to keep the momentum when they actually had the resources for it. So many missed opportunities in AMD's history... Oh well.

Going forward, at least, we can agree AMD being free of GFs' incompetence is a good thing; even if it's expensive to get away from them. If they (GF) ever get their manure together, I wonder how the relationship would go. At this point, I wonder if that is more like a pipe-dream than something feasible...

Cheers!

 

[8350rocks]

I never said that they did. Please re-read my posts, and you'll see that I was speaking hypothetically.

My point was that if Intel could ship an 8 core 16 thread chip @ 4.0 GHz base clock, they would already be doing so, and charging an exorbitant amount of money for it.

 

[-Fran-]

I never said that they did. Please re-read my posts, and you'll see that I was speaking hypothetically.

My point was that if Intel could ship an 8 core 16 thread chip @ 4.0 GHz base clock, they would already be doing so, and charging an exorbitant amount of money for it.

Uhm... I don't think so.

If you look at Intel's line up, it's not like in the P4 days where they had to play the "MOAR HURTZ" card; pretty much like AMD has to do now.

Intel won't come up with a new model just because they "want" or there's a tiny market to grab. It's not Intel's style. They know they own HEDT with the -E CPUs and the mainstream is happy with 4C/8T CPUs, so they're in no rush or pressure to design or market such a beastly chip. Remember, such a CPU would cost money and not money they would get back with 100% certainty, so since they don't need a shinny flagship with AMD lagging as it is now, they don't even need to bother.

Maybe, and just maybe, if Zen puts a good show they might think about it.

Cheers!

 

[jdwii]

Nicely said Yuka! People who are old enough or at least researched it when Amd came out with a amazing chip Intel ran and made a new extreme P4! Overall tho Amd was still better.

 

[gamerk316]

In Intels defense, their Engineers thought they could take Netburst to 10 GHz plus. They (badly) underestimated how much processor leakage was going to kill them. No one would have cared if AMD had 1.5x the IPC if Intel ended up with 5x the clockspeed advantage.

 

[jdwii]

In Intels defense, their Engineers thought they could take Netburst to 10 GHz plus. They (badly) underestimated how much processor leakage was going to kill them. No one would have cared if AMD had 1.5x the IPC if Intel ended up with 5x the clockspeed advantage.

Something like that sounds more like a statement coming from management then the engineering team, Intel has the best engineers in the world i doubt they would ever make such a claim, just tells me when it comes to tech companies i'd much rather have the engineers control things.

 

[ComputerSecurityGuy]

I don't know. I'm not sure how much the engineers particularly the wafer designers as opposed to the fab people, knew about sub 100nm processes. Previously every fab increase allowed a significant clockspeed jump.

 

[-Fran-]

Well, leakage has been a known physics effect when the tunneling effect was discovered. That was in the 70s? It was theorized about at the beginning of the century even. Plus, there are more factors that come into play other than shrinking. I can think on the machinery itself being stupid hard to make and the process chain of getting the waffers must be stupid complex as well.

I would imagine Intel engies knew about it, but the marketing dept extrapolated a bit too much from what the engineers expected. Plus, when unknown, most projections are linear with big bags of error associated, but they never say that explicitly. They used base assumptions that didn't turn out to be true in the long run.

In any case, one thing I know for sure; no matter the amount of conjecture we make, we will never be at the same level of the engies on either side can make (in terms of "predictions", as some people like to call). Educated guesses at best? 😛

Cheers!

 

[simon12]

Don't think I ever heard the 10GHz story before. Just found this if anyone else is interested http://www.anandtech.com/show/680/6
Basically in 2000 they thought 10GHz on 70nm by 2005

 

[turkey3_scratch]

Don't think I ever heard the 10GHz story before. Just found this if anyone else is interested http://www.anandtech.com/show/680/6
Basically in 2000 they thought 10GHz on 70nm by 2005

Ha that's hilarious.

 

[jimmysmitty]

Don't think I ever heard the 10GHz story before. Just found this if anyone else is interested http://www.anandtech.com/show/680/6
Basically in 2000 they thought 10GHz on 70nm by 2005

To be fair, no scientist or company has ever been correct about predicting the future. AMD thought that Core 2 was going to be the same joke as Netburst and wasn't worried.

In the 50s and 60s we were supposed to have jet packs, nuclear driven cars that also flew and tons of wondrous technologies.

I am sure people who were kids then are disappointed and bitter.

 

[TechyInAZ]

Don't think I ever heard the 10GHz story before. Just found this if anyone else is interested http://www.anandtech.com/show/680/6
Basically in 2000 they thought 10GHz on 70nm by 2005

To be fair, no scientist or company has ever been correct about predicting the future. AMD thought that Core 2 was going to be the same joke as Netburst and wasn't worried.

In the 50s and 60s we were supposed to have jet packs, nuclear driven cars that also flew and tons of wondrous technologies.

I am sure people who were kids then are disappointed and bitter.

Also, flying cars btw.

 

[Turb0Yoda]

Don't think I ever heard the 10GHz story before. Just found this if anyone else is interested http://www.anandtech.com/show/680/6
Basically in 2000 they thought 10GHz on 70nm by 2005

To be fair, no scientist or company has ever been correct about predicting the future. AMD thought that Core 2 was going to be the same joke as Netburst and wasn't worried.

In the 50s and 60s we were supposed to have jet packs, nuclear driven cars that also flew and tons of wondrous technologies.

I am sure people who were kids then are disappointed and bitter.

Also, flying cars btw.

Yeah, Howard Stark even demo'd his... (langauge) lied to us

Language Edit. Remember, this is a family friendly forum.

 

[turkey3_scratch]

Yeah but did they predict computers at all? At least, computers as they became in the 80s and 90s?

 

[gamerk316]

Don't think I ever heard the 10GHz story before. Just found this if anyone else is interested http://www.anandtech.com/show/680/6
Basically in 2000 they thought 10GHz on 70nm by 2005

Ha that's hilarious.

In hindsight? Sure. But that was the plan, and the entire architecture was designed around high clocks. Goes to show how much physics remains a limiting factor, even after additional process shrinks and power optimizations.

 

[gamerk316]

http://www.pcper.com/news/Processors/Leaked-Geekbench-Results-AMD-Zen-Performance-Extrapolated?destination=node/66172#comment-308881

Specifically:

AES numbers make sense, given AMD has dedicated resources on the CPU die. For all other tasks, you see performance 10-30% slower then IB at the same clocks.

If true (And that's a BIG if), Zen likely doesn't even match Sandy Bridge IPC levels.

EDIT

I want to make this very clear, if these results are accurate, Zen won't even beat Sandy Bridge head to head. This would be significantly below even my worst-case estimations for Zen's performance.

There's a side of me that says the results are so bad, they can't possibly be real. Then again, I said the same thing in regards to Bulldozer when the first real numbers started to leak.

 

[-Fran-]

His numbers are skewed badly:

1.- "The results from Geekbench show performance from a two physical processor system and a total of 64 cores running at 1.44 GHz" -> "If we assume perfect clock speed scaling, we can guess at performance levels that AMD Zen might see at various clocks."

Everything else is a waste of time IMO. If you're going to write an article with that *fat* assumption, might as well don't. Given the information that was released by AMD and how their designs have been in the past, they don't have linear scaling. Far from it. Plus, does geekbench even scale linearly with speed?

If there is salt to be used, this is when.

Cheers!

 

[scuzzycard]

There's this too - looks like it could be a Zen mobile processor. It says 4 cores, but it has to be 2 cores / 4 threads @ 2.3GHz:

http://browser.primatelabs.com/v4/cpu/199148

Compare that to an i3 6100u @ 2.3GHz which does about 40% better on a single and multi-threaded.

http://browser.primatelabs.com/v4/cpu/220450

Lastly, 4 "48 core" engineering samples @2.1GHz

http://browser.primatelabs.com/v4/cpu/135457

The multicore score is only 10X the single-core, which is odd.

If these results are real, it would suggest that Zen is not particularly good at Geekbench. What any of this actually means is anyone's guess.

 

[gamerk316]

His numbers are skewed badly:

1.- "The results from Geekbench show performance from a two physical processor system and a total of 64 cores running at 1.44 GHz" -> "If we assume perfect clock speed scaling, we can guess at performance levels that AMD Zen might see at various clocks."

Everything else is a waste of time IMO. If you're going to write an article with that *fat* assumption, might as well don't. Given the information that was released by AMD and how their designs have been in the past, they don't have linear scaling. Far from it. Plus, does geekbench even scale linearly with speed?

If there is salt to be used, this is when.

Cheers!

You can solve for clockspeed in this case easily enough; you solve for IPC, then replace the clockspeed in the equation and solve for performance.

Secondly, Intel doesn't get linear gains with clock either. No one does, due to interchip communication issues. Should affect both roughly equally.

Easy way to confirm any bias: Just underclock an i7 and run the Geekbench suite, and compare against Zen at the same clock. Easy enough to solve for IPC, factoring in the differing core counts.

 

[cdrkf]

There's this too - looks like it could be a Zen mobile processor. It says 4 cores, but it has to be 2 cores / 4 threads @ 2.3GHz:

http://browser.primatelabs.com/v4/cpu/199148

Compare that to an i3 6100u @ 2.3GHz which does about 40% better on a single and multi-threaded.

http://browser.primatelabs.com/v4/cpu/220450

Lastly, 4 "48 core" engineering samples @2.1GHz

http://browser.primatelabs.com/v4/cpu/135457

The multicore score is only 10X the single-core, which is odd.

If these results are real, it would suggest that Zen is not particularly good at Geekbench. What any of this actually means is anyone's guess.

The mobile part is not Zen- it's listed as 'Family 21' which is Bulldozer 😛

Also look at the caches... 2mb l2 and no l3. That is not Zen.

Also the '48 core' leak- That is listed as Family 16 (which is K10) so that's probably one of the old 'magny cores' processors from ages ago.

 

[jdwii]

Man even i'm expecting sandy-ivy IPC with integer based operations. Something is probably off.

 

[gamerk316]

Well, I plan to solve for IPC at the clocks the processors were ran at tomorrow morning. Should remove the frequency scaling issues from the results, and give an idea where IPC actually is.

 

[-Fran-]

Thanks gamerk, looking forward to your numbers.

Cheers!