AMD ZEN Engineering Sample AOS - Further Performance Analysis
Click here to post a comment for AMD ZEN Engineering Sample AOS - Further Performance Analysis on our message forum
Noisiv
Not the same thing as fps result. It's from schmidtbag's post:
poornaprakash
The scores of Zen ES is a bit worrying because 5960X is considerably faster than Zen ES despite both running at similar clocks. If the retail Zen is clocked some 3.6 -3.8Ghz then it should score much higher than these scores but still it seems clock for clock performance Zen is considerably slower than Haswell and AMD may needs to clock Zen high in order to match Haswell performance . This reminds me the comparison between Phenom II X4 vs Core 2 Quad (45nm) where Phenom always struggled to come close to Core 2 Quads. Hopefully the retail version of Zen should be fast enough that can offer good performance/price CPU. A special thanks to Mr. Hilbert for this great article :cheers:
DARKSF
Ok mabye i already forgot i run both the HT link and the CPU-NB at 2200.
You can check this ages old review and same story apply to AMD 990 chipset with Phenom II and Corsair Dominator 1600 C7 banks at 1800 (the maximum of this old Phenom with any set of RAM).
anandtech.com/show/3877/asrock-890fx-deluxe-full-review-and-an-investigation-of-thuban-performance-scaling/7
Still i bet on the results being affected more by platform bottleneck than CPU itself.Any way will wait and see if it is a crap i'll just switch to a second hand 3770K based platform which means Intel for first time in 16 years of K7,K8 and K10 usage.
schmidtbag
An engineering sample is very much experimental. The primary purpose of them is so motherboard manufacturers have something to test (in other words, experiment) with. If it were a retail part it would be referred to as "retail sample" or just simply a sample. Engineering samples are frequently known to not accurately resemble the future retail parts; every once in a while they're better.
At this point you're just disagreeing for the sake of disagreeing. You seem to be really nit-picky about vocabulary. A program that can take advantage of 8 threads can be referred to as optimized for 8 threads. Regardless, I don't see how my link is terrible when it proved my point. I said that the game doesn't take advantage of more than 8 threads, which the chart clearly showed. With an optimized scheduler, the game can prioritize physical cores over logical cores, or intentionally focus post-processing on the logical cores. I hope you realize that there are some (albeit, few) situations where hyper-threading is known to decrease performance; this is how optimization comes into play.
Yogi
Ah. Thanks for clearing that up.
Osterwhite2
Hello,
Hilbert Hagedoorn, can you add the Intel I7 4790 and the I5 4570K for your analyse?
These two are present in the "leak" bench and can give us a more precise picture.
🤓
zehoo
I bag of salt is best consumed while reading these benchmarks. Though I still don't have high expectations even when real results eventually show up. It would be nice to be surprised though and have some competition again. Might throw some money AMD's way and use a zen in my HTPC rebuild next year if this is the case.
Fox2232
Well, unless someone sets CPU into manual 3.2GHz stable, that CPU runs 2.8GHz only under load and 3.2GHz is achievable only while few cores are loaded.
And not single person here knows how it was set up, what memory was used, ...
waltc3
There is no Zen anywhere around here...;) All I see is an i7 faking as a Zen which in turn is tuned to match some very dubious Internet "performance" information. Also, the higher you ratchet the res and the IQ the more GPU-limited the game becomes, and the less the cpu matters at all--which is why, regardless of cpu, the higher the res and IQ the closer the frame-rate performance will be, assuming the same GPU.
I guess it was a slow day for HH...;) Not only was the baseline "leaked" *cough* (invented?), but it's not only a matter of an engineering sample, it's also a matter of the motherboard core logic, ram, etc. ad infinitum.
There is no performance info about Zen here, which I believe HH clearly stated...;)
PrMinisterGR
My guess is that we will see soon enough, but people vastly underestimate how much CPU drivers and power profiles matter for proper performance.
Anyone who has dealt with Linux in a tiny bit lower level can tell you that there are lots and lots of things that can go wrong, just on the software side. Not that the FX family are any crazy performers, but didn't they see a big boost with Windows 8, just because of the OS driver?
Dygaza
Article states next: "DX12 mode this benchmark only properly utilizes four threads". Hasn't it been proven many times already that aots scales for 16 threads. There was a user on anand forums who posted those 16 threads from process hacker running on his 24core xeon.
Also I see the game using 8 threads on my I7, and load spreading equally on all of those. Here is pic of threads on my I7:
[Spoiler]http://i1375.photobucket.com/albums/ag458/Dygaza/threads_zpsndlmi9nf.png[/spoiler]
For me it really sounds like that AMD board is really on it's early phases and ain't properly using all cores like it should be. Which is kinda expected at that state. Hilbert, how low can you downclock that 5960x to see core scaling over single thread performance?
schmidtbag
Yes, this is very true. But it's not just drivers and power profiles. There's also processor microcode, schedulers, and the way the application is compiled. When you have software designed for a specific hardware platform, you can do some pretty amazing things. This is why iPhones or gaming consoles can accomplish so much with so little. And yet some people here think optimizations aren't a thing...
CPC_RedDawn
so many typos where you drunk when you wrote this Hilbert??? :givebeer: 😉
Denial
All this is true but the problem is everyone said all this when the Bulldozer leaks started showing up and since then nothing has changed. AMD lacks the development/marketing resources to push the software necessary for these changes to happen, especially on the compiler front.
It's basically GCN without DX12/Vulkan.
Enticles
I'm not believing a single slide i see until actual retail units get reviewed.
I would love to see Zen close the gap on a similarly spec'ed Intel chip, though deep down I don't think Zen will get as close as we would like them to.
but we can dream, right? 🙂
PrMinisterGR
If anyone has used a tool like cpupower, or used a custom scheduler, it's really night and day. Just the performance bias settings alone are enough to change performance dramatically. No matter about the actual power profile of the OS, and/or the CPU governor.
Yeah, it's not like that exactly for CPUs though. The basic structure and instruction sets are identical between Intel and AMD, and the Intel compiler thing was settled since 2009. There is a tad of a grey area still there, but it's much much better than it used to be before the court, and any programmer worth their salt will either use GCC (which produces surprisingly fast code), or simply patch the Intel compiler and be done with it. The Intel compiler (unlike GCC or the Visual Studio compilers), cannot compile everything. Things like the Unreal Engine can't use it for that reason. The problem that Bulldozer had was lower single-threaded performance and the prevalence of not-really threaded applications/games. Considering its age and tech, it performs quite decently on newer titles.
It's much simpler to support a CPU than a GPU, but it still needs a driver for things like power states. It doesn't have the custom kernel that is almost as complex as an OS for its driver, since it's the OS that actually does that job. These are already in for the Linux kernel since last year, and I'm sure that AMD has submitted similar code to Microsoft too.
Spartan
Hopefully, Zen overclocks like hell, and they will to sell it for a reasonable price.
sunnyp_343
how much slow this gonna from i7 6700K?
fantaskarsef
Thanks for the extended work Hilbert, much appreciated!
Indeed I'd say from my merely educated user standpoint, it looks promising! Especially as you could probably save out on hundreds of quid for small workstations with Intel performance, and of course, it could mean that in the future optimisations for more cores may become a topic since either Intel ups the core count on cheaper CPUs / platforms soon, or multithreading in games might increase to overly make good use of more than 4 cores / 8 threads, or Intel has to give us better performance for comparable prices, once again leading to better hardware on the top end of Intel's lineup.
Now we only got to wait for the final retail product to arrive at Hilbert's and be benchmakred against Skylake CPUs hopefully.
Noisiv
that would mean that:
AMD still hasn't figured out the process quirks
{AND/OR}
Zen is so powerful that they purposly chose to leave performance on the table
its power bill can support entire village for a month
