AMD: Zen will offer 40% faster performance per clock than Carrizo

Zarathustra[H];1041924340 said:
JF-AMD is John Fruehe. He was a director of product marketing at AMD.

Sure, I know what you mean, investor disclosures are sacred and all that, but its not like he was just some Joe shmoe leaker who claimed he worked for AMD and leaked shit on forums.

He was pretty much the public face of AMD for the year proceeding the bulldozer launch.

That was a fun time. I've been on several forums over the years when you get a zealot like figure and then their predictions turn out wrong or their core argument disappears...and then they do too surprisingly.;)
 
Been thinking about my next machine build. All the PCs I've ever built myself (and 90% for customers) over the past 20+ years have been AMD based but I don't see that happening next year.

The trend will be truly bucked. All I see in the AMD section is what looks like 2010 technology.
 
Yeah exactly. Notice how he didn't say anything about IPC lol.

And we didn't even ask when he was here because no one even though that single core performance can drop below Phenom II :D

I'm still laughing at my own stupidity when i remember how I bough X6 with mobo in hope of getting kickass 8 core cpu that will be compatible with that nice freshly released 890GX mobo :p
 
Been thinking about my next machine build. All the PCs I've ever built myself (and 90% for customers) over the past 20+ years have been AMD based but I don't see that happening next year.

The trend will be truly bucked. All I see in the AMD section is what looks like 2010 technology.

FM2+ is updated. Just not an option for many of us.
 
Bulldozer was faster in that it could clock higher. I went from a Phenom II X6 at 4.0 GHz to a 8150 and had to get it to 4.5 GHz before I saw an increase in performance. So I guess you could say it was "better" in a way. :D

I remember that JF-AMD. He ran around touting how IPC didnt matter and all kinds of other crap and then always had the caveat that he was server side and wasnt an expert on the consumer side so couldnt answer any questions really.....but had "I know Bulldozer" as his sig. Then when it came out and Bulldozer was slower than Phenom II he posted a big rant on OCN and basically blames all of US! My favorite quote was something like, "all my bosses at AMD said I was wasting my time talking to a bunch of knuckle draggers on these forums. Guess they were right". I just tried to find that post on OCN but its either been deleted of my search skills suck.
 
Bulldozer was faster in that it could clock higher. I went from a Phenom II X6 at 4.0 GHz to a 8150 and had to get it to 4.5 GHz before I saw an increase in performance. So I guess you could say it was "better" in a way. :D

I remember that JF-AMD. He ran around touting how IPC didnt matter and all kinds of other crap and then always had the caveat that he was server side and wasnt an expert on the consumer side so couldnt answer any questions really.....but had "I know Bulldozer" as his sig. Then when it came out and Bulldozer was slower than Phenom II he posted a big rant on OCN and basically blames all of US! My favorite quote was something like, "all my bosses at AMD said I was wasting my time talking to a bunch of knuckle draggers on these forums. Guess they were right". I just tried to find that post on OCN but its either been deleted of my search skills suck.
Hard to say who is worse. Roy@AMD or JF-AMD.
 
That was a fun time. I've been on several forums over the years when you get a zealot like figure and then their predictions turn out wrong or their core argument disappears...and then they do too surprisingly.;)

Well the reason he disappeared was because he got fired lol, but he did ragequit the tech community after the chip came out and everybody saw how it performed. But I never understood why so many people were acting surprised because the performance was leaked well before the chip even released, so it wasn't anything particularly new. Plus the actual engineers said the uarch would lose some performance on the architectural side (which obviously refers to ipc/perf per core/what have you) but was designed to clock high. So they basically told everybody from the get-go what they were getting. AMD's terrible marketing department decided that wouldn't sound good, so they just focused on the octocore stuff and multi-threaded performance and other cherry-picked benches instead.

Now it's not anything like that. AMD are being bullish about the actual IPC increases, even going so far to say that the 40% figure is independent of process node which means it's purely uarch based. Process node on top of that would determine how high parts can clock. Keller did say in 2014 that "we know how to do high-frequency design" but I think Zen will launch with clocks ranging from 3.3ghz to 4ghz for highest bins.
 
Now it's not anything like that. AMD are being bullish about the actual IPC increases, even going so far to say that the 40% figure is independent of process node which means it's purely uarch based. Process node on top of that would determine how high parts can clock. Keller did say in 2014 that "we know how to do high-frequency design" but I think Zen will launch with clocks ranging from 3.3ghz to 4ghz for highest bins.

Safe guess as that is what Intel did in their first 14nm FinFET release :)
 
Well the reason he disappeared was because he got fired lol, but he did ragequit the tech community after the chip came out and everybody saw how it performed.

Before the first benchmarks leaked that showed it having lower IPC JF-AMD kept saying "IPC increases!" When the details of the cache design (write-through has NEVER been high performance) it was clear higher IPC was simply not possible. I argued that point with him for quite a while and as more and more details came out I kept telling him that there was simply no way a 2ALU+2AGU processor with shared front AND back-ends could ever have increased IPC when coming from a 3ALU+3AGU design like k10 which also had a superior cache system.

He just kept arguing that k10 didn't use those extra resources effectively, but Bulldozer would. But that was obviously not going to happen, and is obviously not what happened (though Bulldozer does, certainly, use its few executions resources MUCH more effectively than k10 used its resources, even 100% efficiency would not result in parity with Nehalem). Excavator is the first instance where k15 has better IPC than k10 in most benchmarks (accounting for the lack of L3 in Steamroller (A8-7600) then the IPC increased in the given tests for Excavator estimation).

Zen, though, stands a really good chance at matching Haswell - which is only 37~39% faster than Penryn, k10's direct competition. If it beats Sandy Bridge, and can hit or exceed 4.5GHz, can offer more cores, and possibly also DDR3 compatibility (they already have the flexible memory controller), then it will be a shoe-in upgrade for many people with existing systems and lots of expensive high-end DDR3 (like myself) who could use more cores and is not really all that interested in a few extra percent in IPC - but are unwilling to give up any (like myself :D).
 
Well the reason he disappeared was because he got fired lol, but he did ragequit the tech community after the chip came out and everybody saw how it performed. But I never understood why so many people were acting surprised because the performance was leaked well before the chip even released, so it wasn't anything particularly new. Plus the actual engineers said the uarch would lose some performance on the architectural side (which obviously refers to ipc/perf per core/what have you) but was designed to clock high. So they basically told everybody from the get-go what they were getting. AMD's terrible marketing department decided that wouldn't sound good, so they just focused on the octocore stuff and multi-threaded performance and other cherry-picked benches instead.

Now it's not anything like that. AMD are being bullish about the actual IPC increases, even going so far to say that the 40% figure is independent of process node which means it's purely uarch based. Process node on top of that would determine how high parts can clock. Keller did say in 2014 that "we know how to do high-frequency design" but I think Zen will launch with clocks ranging from 3.3ghz to 4ghz for highest bins.

I remember he still argued Bulldozer will be great even after first leaks telling us to ignore those results ;)
 
I'm still laughing at my own stupidity when i remember how I bough X6 with mobo in hope of getting kickass 8 core cpu that will be compatible with that nice freshly released 890GX mobo :p

I did something similar.

As soon as 990FX launched I bought a 990FX motherboard and a Phenoml II x6 1090T to serve as my wait for bulldozer build, with the expectation of picking up bulldozer on launch and dropping it in.

When bulldozer turned out to be the major disappointment it was, I went out and bought my x79 motherboard and 3930k which is still going strong 5 years later.
 
It's a shame that they canned the three-module Steamroller variant. I understand why they did it, but it would've been a top seller I think.

indeed, fm2+ is a perfectly good platform. much better than am3+.
 
indeed, fm2+ is a perfectly good platform. much better than am3+.

I have to agree with this. Also, if that had not canned the 3 module steamroller part, perhaps they would not have had to do a $65 right off of APUs? Basically, I think they made some really bad short and long term positions because of that. Instead, they produce the 9590 and 9370 which was not really better. Basically, I think they burned a lot of customer bridges by these decisions.
 
I have to agree with this. Also, if that had not canned the 3 module steamroller part, perhaps they would not have had to do a $65 right off of APUs? Basically, I think they made some really bad short and long term positions because of that. Instead, they produce the 9590 and 9370 which was not really better. Basically, I think they burned a lot of customer bridges by these decisions.

The thing is, now that people don't have to upgrade their PCs every 6 months to a year like we did 10 years ago why would you compromise and buy an AMD based rig?

If I know I'm only spending $1000 say every 5-6 years for the core parts (instead of say $500 every 18 months) then why not get the fastest and best supported for that outlay? Taking a risk on a CPU didn't matter years ago as you knew you'd be swapping it out for another before the end of the year. I remember one year going through about five AMD CPUs before I settled on the Opteron 180.

Buying AMD now just seems too much a backward step for such a long term purchase.
 
Buying AMD now just seems too much a backward step for such a long term purchase.

It depends for gaming on Mantle/DX12/Vulkan it might last longer then you think. And for desktop programs I'm not to sure what the problem would be. The only "nasty" thing is that you need more power for 8 core.

If a stock FX-8350 can push 100K batches then it is simple it can last a hell of a long time ...
 
I have to agree with this. Also, if that had not canned the 3 module steamroller part, perhaps they would not have had to do a $65 right off of APUs? Basically, I think they made some really bad short and long term positions because of that. Instead, they produce the 9590 and 9370 which was not really better. Basically, I think they burned a lot of customer bridges by these decisions.

Well the reason the 3 module SR part got canned is because they couldn't get a good enough process node to put it on. I mean theoretically they could've made it, but it would've clocked even worse than the SR we got and consumed much more power as well. Would've been pointless for both desktop and mobile.
 
The thing is, now that people don't have to upgrade their PCs every 6 months to a year like we did 10 years ago why would you compromise and buy an AMD based rig?

If I know I'm only spending $1000 say every 5-6 years for the core parts (instead of say $500 every 18 months) then why not get the fastest and best supported for that outlay? Taking a risk on a CPU didn't matter years ago as you knew you'd be swapping it out for another before the end of the year. I remember one year going through about five AMD CPUs before I settled on the Opteron 180.

Buying AMD now just seems too much a backward step for such a long term purchase.

With AMD current offer You would have to own Core 2 system or older to even consider their parts as upgrade.
 
Yeah no point in buying AMD CPUs at this point. I haven't browsed the AMD CPU section on sites like newegg in years.
 
The thing is, now that people don't have to upgrade their PCs every 6 months to a year like we did 10 years ago why would you compromise and buy an AMD based rig?

If I know I'm only spending $1000 say every 5-6 years for the core parts (instead of say $500 every 18 months) then why not get the fastest and best supported for that outlay? Taking a risk on a CPU didn't matter years ago as you knew you'd be swapping it out for another before the end of the year. I remember one year going through about five AMD CPUs before I settled on the Opteron 180.

Buying AMD now just seems too much a backward step for such a long term purchase.

Which simply does not address the comments I made. ;) They released a 9370 and 9590 in 2013 well keeping their mouths shut about Steamroller CPU releases. They are lots of folks who were willing to upgrade but, when you give them nothing to upgrade to and tell them nothing, you burn the bridges you built over the years.

Therefore AMD screwed up big time by not releasing Steamroller based CPU's on FM2+ chipsets. The fact that they are losing money and had to write down a 65 Million Dollar APU lose speaks volumes to that fact.
 
FM2+ is updated. Just not an option for many of us.

Which would have been had their released new 6 and 8 core cpu's on that platform. However, they simply ignored the customers and went on losing money.

Well the reason the 3 module SR part got canned is because they couldn't get a good enough process node to put it on. I mean theoretically they could've made it, but it would've clocked even worse than the SR we got and consumed much more power as well. Would've been pointless for both desktop and mobile.

Which tells me they were clueless with what they were doing. The APU's are nice but, they are nothing to upgrade too but, just building a second computer for or something cheap.
 
They weren't clueless for the decision, it was the smart thing to do. Why build something that performs worse than its predecessor? That's the same reason they didn't bother making SR and XV-based Opteron parts, hence no AM3+ successors past Piledriver. Jim Keller rejoined AMD in mid-2012 and that was the same year erroneous reports came out saying that everything past Piledriver was canceled. It was partially true, but the reality was that they were taking that R&D and refocusing it on Zen and K12. The obvious downside being them having to retreat from the HEDT market and having to rely on Piledriver for years on end for AM3+ and mainstream APU's for everything else.

Of course enthusiasts ended up having to wait for Zen or jump ship to Intel, but that's just a necessary evil. GloFo dropped the ball yet again and they had no way to do a 3-module SR part. The 28nm BULK process just wasn't suitable for such a thing.
 
They weren't clueless for the decision, it was the smart thing to do. Why build something that performs worse than its predecessor? That's the same reason they didn't bother making SR and XV-based Opteron parts, hence no AM3+ successors past Piledriver. Jim Keller rejoined AMD in mid-2012 and that was the same year erroneous reports came out saying that everything past Piledriver was canceled. It was partially true, but the reality was that they were taking that R&D and refocusing it on Zen and K12. The obvious downside being them having to retreat from the HEDT market and having to rely on Piledriver for years on end for AM3+ and mainstream APU's for everything else.

Of course enthusiasts ended up having to wait for Zen or jump ship to Intel, but that's just a necessary evil. GloFo dropped the ball yet again and they had no way to do a 3-module SR part. The 28nm BULK process just wasn't suitable for such a thing.

The best way to put it then is, in my opinion, they were foolish to stick with Global Foundries.
 
Which would have been had their released new 6 and 8 core cpu's on that platform. However, they simply ignored the customers and went on losing money.

It wouldn't have worked it is a platform which does not go above 100 Watt.
 
The best way to put it then is, in my opinion, they were foolish to stick with Global Foundries.

There aren't exactly a whole lot of options: Global Foundries, TSMC, and Samsung. It also takes a lot of work to adapt a processor to a different manufacturing process. So they couldn't have just easily switched over to TSMC because GloFo dropped the ball.

The biggest problem for AMD is that they simply don't have a lot of resources. So mistakes and things out of their control hurt them a lot more than a company like Intel.
 
With AMD current offer You would have to own Core 2 system or older to even consider their parts as upgrade.

For a gaming PC or higher end, I agree. If the system was C2Q or older, an AMD FX system would be a good upgrade. I was actually using a Phenom II 980 BE and was somewhat torn between going with a new FX build or an Intel. Based off my budge, I was able to go with the X58 setup you see in my sig for just a couple hundred dollars (got a couple steals) and its a healthy increase in performance over the Phenom II and outperforms any of the other AMD offerings.

I have really high hopes for Zen but like its been stated already in this thread its going to be a make or break scenario for AMD.
 
after the bulldozer launch I can see why they are going to be quiet even if Zen is the killer app a lot of us hope for.

As stated above, i'd just be happy with an AMD product that can best the 2600k i've had no reason to leave (lets be honest, the intel chips after sb/ib are marginal upgrades if you already have one). i've always loved the fact that i could treat my amd setups like a hooker from gta and it ran fine, whereas my Intel setups were always like a needy gf in terms of setup and tweaking.

also as a side comment. I do still build fx boxes for my work on occasion. for office work the price point is excellent and being a smaller firm it allows for excellent upgrade paths for work pc's. for higher end stuff i can definitely see it not being worth it (though my portable gaming cube is AMD again due to price and i'm only really running things like League of Legends, counterstrike and WoW on it)
 
I am a hard core AMD fan ,but even i have moved on. When you can get a Xeon processor for $75 that outperforms amd's best at less watts...
 
I am a hard core AMD fan ,but even i have moved on. When you can get a Xeon processor for $75 that outperforms amd's best at less watts...

Xeon may well be my next build too. Next year when it's due should be some good second hand bargains out there. But even the new stuff is pretty reasonable. Xeon kit is pretty bulletproof too.
 
Last edited:
Xeon may well be my next build too. Next year when it's due should be some good second hand bargains out there. But even the new stuff is pretty reasonable. Xeon kit is pretty bulletproof too.

You guys keep saying Xeon but you which one? The old Socket 1156? What socket and type please?
 
Xeon may well be my next build too. Next year when it's due should be some good second hand bargains out there. But even the new stuff is pretty reasonable. Xeon kit is pretty bulletproof too.

If it's one thing I love about server and workstation class kit is the reliability of it. Definitely beats the pants of most anything consumer grade.

You guys keep saying Xeon but you which one? The old Socket 1156? What socket and type please?

I'm assuming either 1366, or 2011. 771 stuff is a bit long in the tooth these days. Not sure about 115x, though.
 
If it's one thing I love about server and workstation class kit is the reliability of it. Definitely beats the pants of most anything consumer grade.



I'm assuming either 1366, or 2011. 771 stuff is a bit long in the tooth these days. Not sure about 115x, though.

Ok, thanks, it was the Socket 1366 I was thinking of. :)
 
You know, Nehalem doesn't really have an IPC advantage over Piledriver... source, I'm running Nehalem right now... see my sig.

That's a highly subjective claim. There will be some instances where they trade blows, but largely Nehalem will outperform Vishera, especially considering multithreaded scenarios where a 6 core Xeon is used.
 
You know, Nehalem doesn't really have an IPC advantage over Piledriver... source, I'm running Nehalem right now... see my sig.

My ESXi server followed the following progression over the last few years: (all stock speeds)

AMD FX-8120 (8 cores, 8 logical, 3.1Ghz, 4.0ghz turbo) -> AMD FX-8350 (8 cores, 8 logical, 4.0ghz, 4.2 turbo) -> 2x Xeon L5640 (12 cores, 24 logical 2.26ghz, 2.8ghz turbo)

I tend to disagree with the statement above.

Even in single threaded stuff, with the huge clock speed deficit, the nehalem beats the FX chips in most tasks. Then factor in that I can have 12 cores (24 logical) while using less power than AMD's 8 cores (8 logical) and the leap is even larger.
 
There are some very tasty rumors floating about out there--I'm holding off for AM4, may get a new Radeon in the meantime...next-gen, mind. Should have enough saved up to get a good one new by the time anything comes out. :D
Edit: Worst case, I'm sorely disappointed, and opt for a z170 board with one of AMD's new GPUs.
 
If it's one thing I love about server and workstation class kit is the reliability of it. Definitely beats the pants of most anything consumer grade.



I'm assuming either 1366, or 2011. 771 stuff is a bit long in the tooth these days. Not sure about 115x, though.

Yeah my current workstation is a dual CPU X5470/771 unit but it still keeps up nicely. Just the power consumption is a tad high. Cost me next to nothing to buy a couple of years ago so I'm still good on the deal.

The fact that as mentioned the kit is solid quality really helps. Once you have gone this route normal domestic kit with dragons and robots on the boxes is just for kids.
 
Yeah my current workstation is a dual CPU X5470/771 unit but it still keeps up nicely. Just the power consumption is a tad high. Cost me next to nothing to buy a couple of years ago so I'm still good on the deal.

The fact that as mentioned the kit is solid quality really helps. Once you have gone this route normal domestic kit with dragons and robots on the boxes is just for kids.

Ah, I half agree. Simply put: enterprise-class hardware is top-tier, but as far as gaming, you need a consumer/gamers card: as the enterprise stuff is slow as snails for realtime rendering. That said, anyone like yourself running 8 cores on 771 still has a competitive amount of CPU horsepower years onward.
 
Back
Top