AMD Zen Performance Preview

http://blenchmark.com/cpu-benchmarks

Naples seems to have appeared on the site. 32 AMD cores=10 core IB(2680v2). But the 115W 2680v2 is running roughly twice the clock. The 8 core IB based 2.6Ghz 95W 2650v2 is right after.
 
Last edited:
Shintai said:
http://blenchmark.com/cpu-benchmarks

Naples seems to have appeared on the site. 32 AMD cores=10 core IB(2680v2). But the 115W 2680v2 is running roughly twice the clock. The 8 core IB based 2.6Ghz 95W 2650v2 is right after.
Click to expand...

cageymaru said:
What's this anomaly hanging out in the top 10? Completed the test in 69 seconds. Is that decent?
http://blenchmark.com/cpu-benchmarks

Saw it on Reddit.

Click to expand...

Your both talking about the same benchmark right?
Edit: Meaning the AMD Engineering Sample located at the top of the list.
Edit2: Any chance it could be faked?
 
Last edited:
gamerk2 said:
Speaking as a software developer; NO ONE USES HARDCODED CPU OPCODES ANYMORE. Its all done compiler side. Unless you're writing a device driver, a benchmarking program, or working with a realtime system that actually cares about performance, NO ONE trys to outsmart the compiler anymore.
Click to expand...

Wow you know every developer on earth and can say that non one does, that's impressive... but also wrong. I can give you at least a handful of people that does it.
Dont uses ultimative words if you cant back it up with more than just your personal opinion

its all depending on what they work with. and if you think we are sitting on a lot of CPU power that noone really needs, Your need to think outside your box mindset of games a he only demanding software to run.
Much of my data analyze programs use a level of highly hand optimized opcodes, and it takes weeks to run due to the massive power needed. So just a 5% increase in overall performance shaves off hours of wait time
 
SighTurtle said:

Your both talking about the same benchmark right?
Edit: Meaning the AMD Engineering Sample located at the top of the list.
Edit2: Any chance it could be faked?
Click to expand...

Ha ha no idea if it can be faked. For all we know it could be a 20, 30, 40 core monster chip for servers. (We can pray it's the desktop even though I SERIOUSLY doubt it is.)
 
Zarathustra[H] said:
because for most programming performance doesn't matter anymore, because most users are sitting on computers with way more RAM and CPU cycles than they'll ever need.

There are exceptions where performance is important, but for the overwhelming majority of software code, that simply isn't the case.
Click to expand...

Sure, if you are writing code for client based apps. Ever heard of the internet? Developers seem to have this mentality they dont need to do any performance optimization anymore and wonder why the production mega-ultron specced server can't handle more than 50 users.
 
In reference to the blenchmark AMD Engineering Sample, if the linked post is correct, then this benchmark is fake.
https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/page-62#post-38554162

Windows 2008 Server being used (if that is accurate) is a pretty clear sign that this result was not done on a Zeppelin based product ;)
Most likely it was done on C32 / G34 ES Opteron parts, in 2P config.

Also since the "Blenchmark" acquires (cpuinfo.py) the CPU name string (which is "AMD Engineering Sample" in this case) from Windows registry (Hardware\Description\System\CentralProcessor\0 \ "ProcessorNameString") it tells that this is no Zeppelin of any kind either.
The name string which gets copied into this registry entry comes from CPUID registers. On Zeppelin based parts it is "AMD Eng Sample:" instead of "AMD Engineering Sample", as we've seen on in AOTS and Geekbench leaks.
Click to expand...
 
Well I definitely hope they are all unlocked processors. Would be another reason to go AMD vice Intel. The waiting is getting long in the tooth.
 
noko said:
Well I definitely hope they are all unlocked processors. Would be another reason to go AMD vice Intel. The waiting is getting long in the tooth.
Click to expand...

Maybe it is not so much the waiting that is problem we would be happily waiting for that cpu that does 4.5ghz when overclocked. Zen has to have a decent base clock but it does not seem that way.
When you buy a FX 8320E or most of the 8xxx series you can get a decent overclock going on it. Even if all Zen cpu are unlocked that does not mean they will get the same range as with the older generation.

It would mean that by the time Zen gets a new stepping B1 or C1 and you still can't get past 3.5ghz it makes little sense to "upgrade".
 
Last edited:
Well I am hoping 4ghz +. I don't see it going too high just because all the engineering samples are in the low 3ghz range. 3.2ghz 3.6ghz turbo my best guess for the 8core/16thread. OCing I hope will have some lead way. Sad part Broadwell E may look like a good OCing cpu compared to Zen -> We will see.

It maybe just clock limited to maintain that 95w power target as well. This makes system builder builds much easier then a 125w or greater rating. Now if AMD can really surprise us then that will be fun.
 
Pieter3dnow said:
To be honest I found the story far fetched ;) and have my doubts , it takes around 6 months for a re-spin of a stepping with improvements ;) .
Click to expand...


Didn't think of that. At the same time that idea of the bios being on the chip is ridiculous. No one would put the bios programming on the processor.
 
Pieter3dnow said:
http://carbonite.co.za/f20/amd-zen-141214-post972366/?s=0f8a8306530c2f672db98e8c93e49975#post972366

disclaimer:
Just someone on the internet ;)
Click to expand...

So, more information is given out throughout the thread (not sure if people naturally read through everything) so for easy reading:

OP:

Apologies if there's another thread about AMD's upcoming CPU architecture and of course desktop CPUs, if there is, delete or perhaps merge this info.
This is the current state of the retail CPUs, which have been improving by the month.

- There are some errata issues present in the current testing samples, similar in a way to the TLB bug of the Phenom. The workaround right now is done via the BIOS. The workaround however, strips around 30 ~ 40% of the CPU performance.

- The CPUs are well behind schedule and every day there's real progress and bug fixing being done. Unlike with INTEL's E0 CPUs which make it to the wild that are almost completely final silicon. AMD's samples will continue to get bug fixes right up until retail spec sampling to partners.

- In August Clock speeds were 3.8GHz, right now 4.2GHz overclocking is possible, with LN2 5GHz is doable. Again this will change of course, but it is just the current silicon that is behaving like this.

- AM4/ZEN uses an SOC design, that means even CMOS/BIOS configuration is on package (not necessarily on silicon, I can't confirm this) so it is possible to clear the "BIOS" and still have old value applied 30 minutes later. How this will be addressed remains to be seen. Perhaps it won't be the same scenario for final silicon

- Operating voltages (nominal) are 1.3v and all the way up to 1.5v should be fine it seems for AIO cooling. Frequency scaling isn't a strong point but again that may have everything to do with the process at this point rather than an inherent design limitation.

- Performance is particularly strong at this point vs. INTEL's latest offerings. Single thread performance is matching Haswell-E and of course multi-threading performance as well. Tests that are memory bandwidth dependent may go to the INTEL platform simply as a result of having more memory channels, but I can't confirm that right now and have no info on that. The important thing here is that the 16Thread/8-Core CPU is minimum 5960X performance if not better actually. (Based on Cinebench R15) with the error fix disabled.

- Can't speak to how well the IMC is working as current samples are locked to low DRAM frequencies (2133MHz and lower) and of course this has an impact on performance.

- As stated in the beginning, every week is progress and AMD is working at an unprecedented rate to get these ready by March.

- You're unlikely to see any high end boards for the CPUs prior to launch or at launch, simply because no vendors can commit to too much right now as plenty is changing at a rapid rate.
Click to expand...

2nd:
No idea on any pricing yet, but from board vendor side, Highest SKU is Less than $700USD.
March is the target time for the CPUs. I don't suspect 1st generation boards will be all that great though.

We finally have native/chipset NVme, PCI-E 3.0 + USB 3.1 with a number of lanes for each.
This one is shaping up well, and a definite contender for INTEL and AMD users alike.

That Cinebench R15 MT score
6900K @ 5.1GHz 2,100~
8C/16T ZEN @ 5.1GHz 2,000 (workaround disabled)
Click to expand...

3rd:
Let's not get ahead of ourselves. IPC right now is unlikely to go up
However be advised that while it can match Haswell-E for sure, it doesn't match Broadwell-E. If you are not aware, a 6900K @ 4GHz is about the same as a 5960X @ 4.5GHz. The difference in IPC is more than one would think oddly enough.

As for TDP, INTEL measures TDP entirely differently from AMD. AMD never measures max power draw or instantaneous power draw while INTEL does. It's is a lot more complicated than this but suffice to say they aren't directly comparable and of course this stands to reason because operating voltage for the the CPUs right now is 1.3V vs. Intel's 1.2v or so for a higher clock speed and a denser CPU.

Catching up to the leader is a simpler task than charting new ground and pushing the frontier. AMD has the road map laid out for it and the advantage of not having had to pioneer an architecture to match an untested node or process. That work has already been done by others and and if you look at tri-gate fabrication processes, they are so much better than they were when we first got them at 22nm.
Single thread performance still matters if only for the low power states and what options it allows you for power gating the CPU. Not necessarily the power it gives you at full tilt.

The cadence that INTEL uses for platform development even right now, puts them quite ahead even though they've not been necessarily focused on improving performance per say, but rather performance per watt, much like NVIDIA. The Kaby Lake is the Oregon dev team I think, while Skylake-X and Cannonlake should be the Israeli team. (could have the swapped) Point is Skylake-X platforms for instance, only due Q4 2017 are already available at all the vendors with CPUs at a later development stage than Zen is at present.

Either way, 2017 is going to be exciting for AMD and us of course. New GPUs and new platforms that bring AMD right up to modern computing performance. Pricing is not going to be cheap because if you have a comparable part, you price it accordingly. So yeah for high end look to $700 or so
Click to expand...

4th:
Huge news and an update coming up soon.
Just spent a portion of the day with.... and yeah I'm worried, but extremely excited as well. Have some SKU numbers as well...

We are in for a revolution people.
smile.png

Single core performance is strong, yep, it is very strong. As for price... heheheh
cool.png

All I can say is INTEL isn't panicking, but they should be. Yes they should be
Click to expand...

5th:
Those that go back to Planet Mars days, when even old Henk himself was still into some overclocking, will know I was hellishly defensive of AMD. Of course That was around 2001 or so maybe all the way up to 2004 perhaps. There wasn't a single AMD CPU or chipset I didn't get my hands on, even if it meant starving. Hopped on the Mobile Athlon-XP and ran what I think is SA's first sub-zero OC with DICE and a container made or rather stripped from cast iron (came from the leg of a table). AMD yeah, I lived and breathed AMD. However as you learn and get exposed the truth has a very sobering effect on one.

I'm excited because of the potential I see even this early. AMD can't help but get in their own way, but things are still looking up. They can only be better with 2nd gen Zen.
Anyway here goes
==============================================

* All overclocking is done via Overdrive, you can't change any performance features at all in the BIOS (on to that next) at all.

* BIOS or UEFI is actually built into the CPU, so only AMD can update the "BIOS" or microcode. All overclocking must take place within the Operating system

* Right now it takes up to 30 minutes to clear the BIOS. If you remove the CPU and place it on another motherboard, it'll have the same settings applied as on the previous board. So debugging is a nightmare

* 6850K SKU (May not be final designation) is wait for it.... $300 roughly. That's 8 Cores and 16 Threads

* AMD's Hyper Threading is called SMU and it is damn good. The same efficiency as Intel's HT.

* Performance is really good, be it SuperPi, Cinebench, 3DMark etc, it's FPU performance is incredibly good and easily matching that of what Intel offers.

* Current performance is staggering even though it is limited to 2133MHz (as mentioned before) and NorthBridge Frequency is limited to 2400MHz

* There will be a nigher SKU than the 6850K, but it is a higher bin so it will certainly overclock better than 6850K and that may carry a premium price, but unlikely to be double.

* There's plenty of excitement from all board vendors about the platform, so we will see how it all pans out. (Especially with the hot mess that INTEL has in store for us H2 2017, that we can leave to another thread)

* For Gaming, the CPU is neck and neck with INTEL, even at low res where CPU bound.
================================================== ============

AMD is coming for em, NVIDIA and INTEL. 2017 is going to be a real kicker of a year as AMD will be able to provide premium performance at under $1,000 for the entire platform (CPU, board and VGA) whereas the equivalent from INTEL is about $1,000 just for the CPU.

If there's any questions I'll answer what I'm allowed to.
Click to expand...

6th:
AMD essentially has no Chipset for the CPUs. It's an SOC of sorts so everything is just about contained within the CPU or at least the CPU package. That does simplify the board in some ways, but the worst part is that it takes away the small differences boards had between them. Right now there's literally no difference at all between any board (all of which are far too early to be relevant). Essentially all control is in AMD's hands and in the user hands for overdrive.

You can still access traditional BIOS for SATA config, Audio, boot sequence etc but that's it really. Even DRAM frequency you'll have to set inside the OS and reboot as I don't think you can change multi once the mem training is done. We will have to see.

Again there are still way too many bugs to iron out, but the performance is there so they must work to iron out the rest of it. There's too much of a spread right now between CPUs as I said
4.1GHz vs 5.2GHz on some. But progress is being made rapidly so we will see. Overclocking is only on the OC SKUs, but as we have seen from AMD, that's almost all of them.
Click to expand...
 
Last edited:
For some reason i suspect i know the poster.

Anyways, while that sounds really good, the variance he mentions on obvious subzero OCs is weird. GloFo still has not sorted out their FinFETs or what?

And of course, the SKU name is BAAAAAD.
 
$300 for a 8 core 16 thread? That sounds too good to be true. Then again they have 32 core chips? for enterprise.
 
cageymaru said:
$300 for a 8 core 16 thread? That sounds too good to be true. Then again they have 32 core chips? for enterprise.
Click to expand...
Not when you are really far behind and require making a huge splash to earn back market share. AMD is lucky to even be in business. Giving up making a lot of profit for a few years to gain back market share and reinvest that money into their business may be their only path to success.
 
bigdogchris said:
Not when you are really far behind and require making a huge splash to earn back market share. AMD is lucky to even be in business. Giving up making a lot of profit for a few years to gain back market share and reinvest that money into their business may be their only path to success.
Click to expand...

Yes, but according to the article, depending on the silicon lottery, Zen can be faster than a comparable Intel chip. And it's $300. In my mind this does not compute. Then again the SOC has everything inside it. So in the long run it could be cheaper to produce?

I guess what I'm saying is that I'm taking the skeptical wait and see approach to this leak. ;)
 
razor1 said:
Ah yeah sometimes still do ;) really depends on what the developer is going for, and if there is time (resources) to do it.
Click to expand...

It's totally dependent on what you're writing and the budget. Speaking as a software developer (apparently we're all software developers) I have NEVER had to ASM code anything ever and in fact I'd probably be fired for "wasting all that time" if I did. I write mostly IO-bound code and that's where all the time is spent so optimizing that is generally where I look if I need better performance. And don't say it's an issue with more users, we're got thousands of users per front-end server. CPU cycles are so cheap these days that unless you're doing AI, physics modelling or 3D it's not generally an issue.

P.S. We use the Microsoft compiler ;)
 
cageymaru said:
Yes, but according to the article, depending on the silicon lottery, Zen can be faster than a comparable Intel chip. And it's $300. In my mind this does not compute. Then again the SOC has everything inside it. So in the long run it could be cheaper to produce?

I guess what I'm saying is that I'm taking the skeptical wait and see approach to this leak. ;)
Click to expand...

300 probably for the models that are cheap and worst overclockers probably bottom of the barrel chips that is what you pay what you get. Probably sufficient enough to run on advertised speed and not much higher. I think AMD has this perfectly planned. Then sell premium chips at a premium. up to 500-600 or even more but quiet a bit cheaper than intel?

If they are launching 8 core chip first they can't simply price them selves out. Probably having a few different models and charging premium for binned chips.

But I might have just gave myself what I wanted to believe to be true. lol
 
Last edited:
Quix said:
It's totally dependent on what you're writing and the budget. Speaking as a software developer (apparently we're all software developers) I have NEVER had to ASM code anything ever and in fact I'd probably be fired for "wasting all that time" if I did. I write mostly IO-bound code and that's where all the time is spent so optimizing that is generally where I look if I need better performance. And don't say it's an issue with more users, we're got thousands of users per front-end server. CPU cycles are so cheap these days that unless you're doing AI, physics modelling or 3D it's not generally an issue.

P.S. We use the Microsoft compiler ;)
Click to expand...

Well- I'm a hardware engineer. Though I've written software. Hopefully that counts.

In general I agree with you. But in a server environment there's only so much optimization that can be done considering hardware always presents brick-wall limitations. No matter what compiler you use.
 
NKD said:
300 probably for the models that are cheap and worst overclockers probably bottom of the barrel chips that is what you pay what you get. Probably sufficient enough to run on advertised speed and not much higher. I think AMD has this perfectly planned. Then sell premium chips at a premium. up to 500-600 or even more but quiet a bit cheaper than intel?

If they are launching 8 core chip first they can't simply price them selves out. Probably having a few different models and charging premium for binned chips.

But I might have just gave myself what I wanted to believe to be true. lol
Click to expand...

Let me reread it. I may have misinterpreted it. I thought he was saying the 8C 16T was $700 at the start and then said $300 by page 4. I was skimming also so. :)
 
Well that thread is stuck behind a login wall now. I think someone quoted it in this Reddit thread. This is what I'm talking about that makes no sense to me. I mean I'd LOVE for it to be true. If putting everything into the SOC allows for this type of pricing then I'm all for it. Does make you think though. At least question why Intel charges so much.

6850K SKU (May not be final designation) is wait for it.... $300 roughly. That's 8 Cores and 16 Threads

 
I'm still running an X58 machine, plan to upgrade in a year or 2 when I feel tech has moved far enough to justify the build. I really hope Zen will be something worth looking at
 
cageymaru said:
Well that thread is stuck behind a login wall now. I think someone quoted it in this Reddit thread. This is what I'm talking about that makes no sense to me. I mean I'd LOVE for it to be true. If putting everything into the SOC allows for this type of pricing then I'm all for it. Does make you think though. At least question why Intel charges so much.

6850K SKU (May not be final designation) is wait for it.... $300 roughly. That's 8 Cores and 16 Threads

Click to expand...

Isn't that a confirmation that it was click bait to begin with ;) makes life so much easier ;) . Person makes post gets quoted drives traffic gets members , for what ?
 
Quix said:
It's totally dependent on what you're writing and the budget. Speaking as a software developer (apparently we're all software developers) I have NEVER had to ASM code anything ever and in fact I'd probably be fired for "wasting all that time" if I did. I write mostly IO-bound code and that's where all the time is spent so optimizing that is generally where I look if I need better performance. And don't say it's an issue with more users, we're got thousands of users per front-end server. CPU cycles are so cheap these days that unless you're doing AI, physics modelling or 3D it's not generally an issue.

P.S. We use the Microsoft compiler ;)
Click to expand...


I agree, compilers now a days are much better then 15 years ago, I would say if the code is written well the compiler can get 95% of the performance out, that extra 5% more isn't worth the time to do ASM isn't usually worth it unless going for something very specific.
 
razor1 said:
I agree, compilers now a days are much better then 15 years ago, I would say if the code is written well the compiler can get 95% of the performance out, that extra 5% more isn't worth the time to do ASM isn't usually worth it unless going for something very specific.
Click to expand...
Basically true, except for proper usage of even SSE registers, let alone AVX/AVX512 ones.
 
cageymaru said:
Well that thread is stuck behind a login wall now. I think someone quoted it in this Reddit thread. This is what I'm talking about that makes no sense to me. I mean I'd LOVE for it to be true. If putting everything into the SOC allows for this type of pricing then I'm all for it. Does make you think though. At least question why Intel charges so much.

6850K SKU (May not be final designation) is wait for it.... $300 roughly. That's 8 Cores and 16 Threads

Click to expand...

$300 would mean slower than 6700K/7700K big time. But as mentioned, its clickbait.
 
Naples (Server Zen) SiSoftware Results ''2S1451A4VIHE4_29/14_N''

- Processor Multi-Media: 643.37Mpix/s
- Processor Cryptography (High Security): 0.17GB/s
- Memory Bandwidth: 0.31GB/s
- Processor Financial Analysis (High/Double Precision): 66.75kOPTS

http://ranker.sisoftware.net/show_system.php?q=cea598ab99ae97af9eb8dfe2cffed8aa97a781e8d5e0c6ae93a680f8c5f4d2b7d2efdff98ab786b0&l=en

Here's Xeon E5-2699 v4 in Comparison

- Processor Multi-Media: 941.16Mpix/s - 2458.56Mpix/s
- Processor Cryptography (High Security): 6.50GB/s - 25.08GB/s
- Processor Financial Analysis (High/Double Precision): 149.42kOPTS - 150.32kOPTS
 
More Zen benchmarks, more showing what every leaked one has. Not even remotely close to the rigged Blender.
 
-Sweeper_ said:
Naples (Server Zen) SiSoftware Results ''2S1451A4VIHE4_29/14_N''

- Processor Multi-Media: 643.37Mpix/s
- Processor Cryptography (High Security): 0.17GB/s
- Memory Bandwidth: 0.31GB/s
- Processor Financial Analysis (High/Double Precision): 66.75kOPTS

http://ranker.sisoftware.net/show_system.php?q=cea598ab99ae97af9eb8dfe2cffed8aa97a781e8d5e0c6ae93a680f8c5f4d2b7d2efdff98ab786b0&l=en

Here's Xeon E5-2699 v4 in Comparison

- Processor Multi-Media: 941.16Mpix/s - 2458.56Mpix/s
- Processor Cryptography (High Security): 6.50GB/s - 25.08GB/s
- Processor Financial Analysis (High/Double Precision): 149.42kOPTS - 150.32kOPTS
Click to expand...

:rolleyes: Yep, of course this processor is going to be slower than the Piledriver Architecture, right.........:wtf:
 
You must log in or register to reply here.
Back
Top