New Zen information, AM3+ info, APU presentation, and video card information

[Shintai]

it shows L3 fine for my test. For now I am taking this with a grain of salt. Cuz my 4 cores are beating 64cores. I am like scratching my head. Even at 1.4ghz 62 cores should make up for it. I think the ES sample might be crippled or something.

Remember scaling issue. If you look at some parts of the test like LZMA. You get 38x scaling. The Haswell scales 11x in that.

HTML5 parse scales 35x or so. Jpeg scales 44x. Camera scales 45x.

And this is with MCM and clustered cores that goes against scaling.

It just looks like another example of hype meets reality. Just as the single threaded results.

 

[gamerk2]

Leaked Geekbench Results: AMD Zen Performance Extrapolated | PC Perspective

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.

 

[lolfail9001]

Well, Excavator outside of few fringe cases is around Penryn level per-clock, actually.

Would be weird if +40% over that would suddenly lead to it being competitive with Haswell and higher.

 

[Chimpee]

These numbers don't make any sense and are based off an engineering sample clocked at 1.44Ghz. I would take these results with a grain of salt as they are not likely to be indicative of Summit Ridge especially considering the difference in clock speeds. It's also one benchmark. How about Cinebench or other relevant benchmarks (Blender was one of them).

If we learn anything from AMD, all benchmarks should be taken with a grain of salt until reviewers get their hands on it.

 

[lolfail9001]

These numbers don't make any sense and are based off an engineering sample clocked at 1.44Ghz. I would take these results with a grain of salt as they are not likely to be indicative of Summit Ridge especially considering the difference in clock speeds. It's also one benchmark. How about Cinebench or other relevant benchmarks (Blender was one of them).

These numbers make perfect sense actually

 

[Spirit_Retro]

These numbers make perfect sense actually

At face value they do- but it's not final silicon, and it's extrapolation of a very underclocked processor.

Who knows if the cache was even functional?

I'll predict two possible outcomes- this will be a massive win for AMD or a massive fail. Not sure there's much in between.

Sure- I'd love AMD to pull a rabbit out of the hat. But the rabbit needs to be alive.

 

[lolfail9001]

No they don't and you haven't explained why they make sense. The 8150 was in direct competition with the i5 2500K and that was Zambezei. Summit Ridge is not going to be 10-30% slower than Ivy Bridge considering clock for clock in Blender it was slightly faster than Broadwell-E 6900K. Geekbench is not be relied upon considering what Linus Torvalds said on Geekbench:

Real World Technologies - Forums - Thread: Charlie re: Apple and ARM

That's not mentioning Geekbench and L3 cache's.

They make sense after you look at it bench-per-bench basis.
Next, 8150 was only in competition with 2500k in a set of cherry picked benches. It's well known how much of trainwreck it was otherwise.
Third, read my previous point about cherry picked benches. Excavator module (!) throughput is in general around the level of Haswell i3 core per clock, while their single thread performance is more like ~60%. Merge "cores" into a core, removing CMT penalty and adding SMT and add 40% IPC and bah-dum, you land at the level of Broadwell SMT core throughput while still lagging behind it in single thread performance per clock.

Finally, Linus said that about GB3, IIRC he is of better opinion about GB4.

 

[Deathroned]

Rumors.

Compared to a Haswell Xeon with normalized clocks, the ST performance is really low. The Haswell is about 50% faster. (1.6Ghz Xeon as example.).

It also seems a 1.45Ghz Zen fits around a 1.8-2Ghz Pilediver in ST.

it must be nice pulling stuff off your rear, bashing an unfinished and an unoptimised product.

 

[lolfail9001]

it must be nice pulling stuff off your rear, bashing an unfinished and an unoptimised product.

Well, he had source for his stuff, so you better edit out "off your rear" part.
And finally: define "unfinished". Unfinished microprocessors don't run at all, you know.

 

[lolfail9001]

So Geekbench is a better and more indicative benchmark of general architecture than Cinebench or Blender or any other multi-threaded application? Last I checked most reviewers don't use Geekbench even for recent processor generations. I don't recall Geekbench being widely used in Haswell-E or even Broadwell-E reviews and benchmarks.

To start with, Geekbench is a set of benchmarks, like SPEC, but more memory sensitive, and probably worse, but i have not seen the source and ain't got time for other sorts of analysis just yet.

http://www.anandtech.com/bench/product/434?vs=288

I wouldn't consider that to be cherry picked benchmarks.

Well, i see a complete blow-out by 2500k in most of tests here. So yeah.

We can't effectively analyze what Excavator performance is on a HEDT platform because there are no Excavator FX processor SKU's. Not to mention that there haven't been any since Vishera. So even if you compared their APU's it wouldn't be a proper comparison. In addition you can't directly compare performance of Excavator to Zen as Bristol Ridge parts have not been thoroughly benchmarked and reviewed extensively (or even properly leaked).

Well, i am operating off of your own assumption that Blender scales absolutely linearly with core/thread count, because otherwise you cannot use it to make an IPC claim on different SMT implementations.

Geekbench as a benchmark regardless of version is not widely used or reviewed. It might appear for select review sites but generally it's not widely relied upon so claiming that 10-30% of Ivy Bridge performance is incorrect. Explain why Summit Ridge would be 10-30% behind the IPC of Ivy Bridge when at the onset of being taped out it was designed to compete with Haswell (and later with Broadwell)

Bulldozer when it was taped out was aiming at competition with Westmere servers now and Sandy-EP servers later or something. Look, how that turned out

Yes they do that's why there is errata and various steppings and revisions hence "engineering sample".

Yeah, those serve to make sure the microprocessor actually works correctly on desired conditions. Because if you have to go back to drawing board after it turns out you have to redo design itself, you've just delayed your CPU release for a few months, if not years (see original 45nm Bulldozer plans).

 

[gamerk2]

I did a major analysis of the geekbench results over a Toms. We concluded a few things:

1: We discarded the multithreaded results. Be it scaling or simply an unpached Linux Kernel, the numbers simply didn't make sense.

2: Compared to various architectures:

12.73% faster then Piledriver
22.52% slower then IB
44.13% slower then Haswell [Desktop]
48.22% slower then Haswell [Server]

Disregarding memory based benchmarks, for obvious reasons.

Didn't get a chance to test my trusty 2600k over the weekend, but the numbers put IPC somewhere around SB level, which is at the low end of my expectations. That being said, the nature of the host OS (Linux) is a concern, since it almost certainly doesn't have a clue how to use the CPU properly, and depending how it's being handled, could be eating some performance. Still, numbers are close to what I'd consider reasonable for Zen based off what I know so far.

 

[Shintai]

Two very interesting comments from the PC Perspective comment section:

I think the Geekbench results should be discarded as they wouldn't accurately show the current clocked ES Zen Summit Ridge at 3Ghz.

It doesn't matter what its clocked at, since we can compare with equally lower clocked Xeon parts. With roughly the same equal I/O.

Its right tho, those using the 1.45Ghz Zen to scale up gets a wrong number from a "3Ghz" Zen. Because the scaled up number will be higher than what the real 3Ghz Zen will perform.

 

[Luka]

Microsoft made 'em do it: The latest Kaby Lake, Zen chips will support only Windows 10.
Microsoft made 'em do it: The latest Kaby Lake, Zen chips will support only Windows 10

I read some months ago that they would work in older OS, but don't expect certain optimizations to be enabled as they are only found in WIn 10.

There is nothing as far as optimization goes with Windows 10, OS is a junk. I use Windows 7 with x99 platform and Nvidia 1080 and shit runs and it runs incredible well. Kaby Lake and Zen will work fine with Windows 7.

 

[a3venom]

There is nothing as far as optimization goes with Windows 10, OS is a junk. I use Windows 7 with x99 platform and Nvidia 1080 and shit runs and it runs incredible well. Kaby Lake and Zen will work fine with Windows 7.

Funny how people are still complaining about windows 10, this is the most perfect OS they have ever made from a performance standpoint.

 

[lolfail9001]

Funny how people are still complaining about windows 10, this is the most perfect OS they have ever made from a performance standpoint.

Better than WinXP with olive theme? Hehe.

 

[Darakian]

Funny how people are still complaining about windows 10, this is the most perfect OS they have ever made from a performance standpoint.

\
It's great on the performance front. If you want a computer to spy on you very quickly then it's everything you could ever want.

 

[a3venom]

\
It's great on the performance front. If you want a computer to spy on you very quickly then it's everything you could ever want.

Privacy is an illusion, as long as MS doesn't steal my bank account, I could care less.
I'm sure the NSA will find me if i was trying to join ISIS even if I was on win 95

 

[Araxie]

\
It's great on the performance front. If you want a computer to spy on you very quickly then it's everything you could ever want.

And the problem is?. are they going to know now that I watch some Pron? damn, im concerned about that.. lol I still don't understand the overly concern about "privacy".. shit, that doesn't even exist.. as a3venom said above, privacy is an illusion..

 

[Darakian]

Privacy is an illusion for you maybe. As snowden put it
'Arguing that you don't care about the right to privacy because you have nothing to hide is no different than saying you don't care about free speech because you have nothing to say.'

 

[Ruoh]

Funny how people are still complaining about windows 10, this is the most perfect OS they have ever made from a performance standpoint.

Better than WinXP with olive theme? Hehe.

\
It's great on the performance front. If you want a computer to spy on you very quickly then it's everything you could ever want.

Privacy is an illusion, as long as MS doesn't steal my bank account, I could care less.
I'm sure the NSA will find me if i was trying to join ISIS even if I was on win 95

And the problem is?. are they going to know now that I watch some Pron? damn, im concerned about that.. lol I still don't understand the overly concern about "privacy".. shit, that doesn't even exist.. as a3venom said above, privacy is an illusion..

Privacy is an illusion for you maybe. As snowden put it
'Arguing that you don't care about the right to privacy because you have nothing to hide is no different than saying you don't care about free speech because you have nothing to say.'

How does this have ANYTHING to do with the thread topic?

 

[a3venom]

Privacy is an illusion for you maybe. As snowden put it
'Arguing that you don't care about the right to privacy because you have nothing to hide is no different than saying you don't care about free speech because you have nothing to say.'

That is good but there is no choice to use a different OS than win 10 for most home users, and complaining against it isnt going to make it go away.

It's same like I could care less about global warming or deforestation as i would be dead anyway before that becomes something major

 

[NKD]

There is nothing as far as optimization goes with Windows 10, OS is a junk. I use Windows 7 with x99 platform and Nvidia 1080 and shit runs and it runs incredible well. Kaby Lake and Zen will work fine with Windows 7.

junk? Use it first before you bash something for your blind love for windows 7. I loved windows 7 too and bashed windows 8 because it deserved it but windows 10 other than people having all hell break lose over privacy stuff is nothing but a performance beast. I can assure you windows 10 uses resources much more efficiently than windows 7. Because I have ran windows 10 on a 4gb machine and widows 7, windows 10 is butter smooth while windows 7 starts choking the more windows I have open.

 

[luminousone]

Eventually all are forced to the latest version of windows, And this is good, And not because the newer kernel is more optimized, or supports the latest hardware, or looks/feels better in any way, regardless of the arguments that any of these points would bring up.

The reason everyone should update, is for maintained security patches, people running out of date security patches or versions of windows that are outside support shouldn't be allowed to connect to the internet, under pain of a severe beating with a large trout.

 

[lolfail9001]

Eventually all are forced to the latest version of windows, And this is good, And not because the newer kernel is more optimized, or supports the latest hardware, or looks/feels better in any way, regardless of the arguments that any of these points would bring up.

The reason everyone should update, is for maintained security patches, people running out of date security patches or versions of windows that are outside support shouldn't be allowed to connect to the internet, under pain of a severe beating with a large trout.

Anyone caring about security would not use windows with internet connection in the first place. Any version of windows, in fact. Some may in fact argue that for any OS.

 

[gamerk2]

Anyone caring about security would not use windows with internet connection in the first place. Any version of windows, in fact. Some may in fact argue that for any OS.

Security through Obscurity is not security at all. If you don't want something exposed to the planet, don't put it on a device connected to the worlds largest public network.

 

[Bandalo]

Anyone caring about security would not use windows with internet connection in the first place. Any version of windows, in fact. Some may in fact argue that for any OS.

If you really care about security, never turn your device on and stay far away from everyone.

Otherwise simply balance the risks and rewards like everyone else.

 

[Nobu]

Security through Obscurity is not security at all. If you don't want something exposed to the planet, don't put it on a device connected to the worlds largest public network.

Anyone caring about security would not use windows with internet connection in the first place. Any version of windows, in fact. Some may in fact argue that for any OS.

I think you missed his point--it's not simply security through obscurity if you deny access to your computer from anyone who does not have direct access.
Anyway, this discussion is supposed to be about "new zen information"...windows security has no relation to that.
As far as CPU support, it's Microsoft who's not supporting older CPUs on future versions of Windows, not AMD or Intel (who have almost no incentive not to support older versions, other than driver/microcode development costs, which probably aren't much compared to what they could lose). What that probably means is they will be using newer x86 and AMD64 extensions which aren't supported in older processors, possibly without fall-back code. At least, that'd make the most sense, and would allow for a better experience with those processors and less work for Microsoft engineers.

 

[luminousone]

I don't see any of the processor extensions being a problem, only reason you would need direct os support is if the register file was larger, this happened with P-III with SSE for example.

Anyone caring about security would not use windows with internet connection in the first place. Any version of windows, in fact. Some may in fact argue that for any OS.

Their is a giant difference between the target on a fully updated and properly firewalled windows machine, and one that hasn't been updated in years and also lacks a firewall.

 

[pendragon1]

I keep popping into this thread for more zen info but only to find people still arguing over windows/security. start a new thread to argue in, FFS...