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

KazeoHin said:
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.
Click to expand...

Its all hooked up with a HD7870, BX100 SSDs, USB3, Xonar audio etc. etc.

It's no slouch for having quality 2008 tech at its heart.:cool:
 
KazeoHin said:
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.
Click to expand...

I get what you're saying but the X5470 is just a Core 2 Quad based Xeon. I purchased an X5450 a while back to use in my DFI 775 board but screwed up the mod and killed both the CPU and motherboard. Probably my biggest fail messing with computers lol
 
TheEschaton said:
You know, Nehalem doesn't really have an IPC advantage over Piledriver... source, I'm running Nehalem right now... see my sig.
Click to expand...
comparing in Cinebench R11.5 single thread
FX@5GHz score 1.33
Nehalem @4GHz score 1.35
Similar but FX need 1GHz more to cath up to Nehalem thus it have worse IPC. Nehalem is 25% better here.

FX@4GHz score 6.94
Nehalem @4GHz score 6.77
FX is faster here but it is due to better multi-thread scaling of threads from modules compared to Hyper Threading and not due to better IPC. If we used 4 threads then Nehalem @ 4GHz would easily win with FX @ 4GHz

Then there are games. Bulldozer have inefficient cache design and loose quite a lot of performance due to this. Performance which it cannot regain by having better scaling on its additional threads as games use two to four cores.

Frankly if Bulldozer had Nehalem IPC but could go 5GH, had scaling it have then no one would ever say it is bad CPU.
 
XoR said:
comparing in Cinebench R11.5 single thread
FX@5GHz score 1.33
Nehalem @4GHz score 1.35
Similar but FX need 1GHz more to cath up to Nehalem thus it have worse IPC. Nehalem is 25% better here.

FX@4GHz score 6.94
Nehalem @4GHz score 6.77
FX is faster here but it is due to better multi-thread scaling of threads from modules compared to Hyper Threading and not due to better IPC. If we used 4 threads then Nehalem @ 4GHz would easily win with FX @ 4GHz

Then there are games. Bulldozer have inefficient cache design and loose quite a lot of performance due to this. Performance which it cannot regain by having better scaling on its additional threads as games use two to four cores.

Frankly if Bulldozer had Nehalem IPC but could go 5GH, had scaling it have then no one would ever say it is bad CPU.
Click to expand...

sincerely I would be happy if Zen can deliver This ivy Bridge Single Threaded Performance score I would happily replace my 2 FX machines for Zen with 8 or more truly cores (ofc depending of the price) I just hope to not see another aberration like the 900$ FX9590 anymore. but also I have to take in consideration that Ivy Bridge its 3 generation old and a good 15%-25% behind Skylake in IPC. but again, I would be happy with ivy bridge single thread performance.

5ed312.jpg


don't forget that at stock settings a FX8350 have a Single Thread performance of just 1.05 so is a big number to achieve..
 
Araxie said:
sincerely I would be happy if Zen can deliver This ivy Bridge Single Threaded Performance score I would happily replace my 2 FX machines for Zen with 8 or more truly cores (ofc depending of the price) I just hope to not see another aberration like the 900$ FX9590 anymore. but also I have to take in consideration that Ivy Bridge its 3 generation old and a good 15%-25% behind Skylake in IPC. but again, I would be happy with ivy bridge single thread performance.

5ed312.jpg


don't forget that at stock settings a FX8350 have a Single Thread performance of just 1.05 so is a big number to achieve..
Click to expand...

That is not a stock clocked 3770 at all. However, a stock clocked one would still be faster than the 8350. My I7-6700k at 4.2GHz is 2.06 so, if my 6700k was clocked at 3.5GHz, I am absolutely certain that it would not be slower than a stock clocked 3770k.
 
ManofGod said:
That is not a stock clocked 3770 at all. However, a stock clocked one would still be faster than the 8350. My I7-6700k at 4.2GHz is 2.06 so, if my 6700k was clocked at 3.5GHz, I am absolutely certain that it would not be slower than a stock clocked 3770k.
Click to expand...

exactly, this is the minimum single threaded performance I would consider good to consider zen a success to be at the level of ivy bridge at 4.5ghz which ins't uncommon for this kind of chip.. Anything below this performance will be in my opinion disappointing in every sense and it will make zen just look crap and a huge fail.
 
Araxie said:
exactly, this is the minimum single threaded performance I would consider good to consider zen a success to be at the level of ivy bridge at 4.5ghz which ins't uncommon for this kind of chip.. Anything below this performance will be in my opinion disappointing in every sense and it will make zen just look crap and a huge fail.
Click to expand...

And then I look down into your sig and see a 4.5 GHz overclocked 3770k. :eek: Good overclock! I could not wait for Zen which is why I upgraded to a 6700K. However, I still use my FX8350 at work running at 4.2GHz and it is very fast overall. I don't game on it much though and anything less than a 5820k at work probably would not be worth it.
 
Am I the only one that would be happy with a Zen that wasn't miles better than Piledriver, as long as it had Intel equivalent power efficiency due to the smaller process? 25% IPC over my FX-9590, stock Turbo of 4.1 GHz+, 24/7 air overclocks of 4.4 GHz+, and less *:mad:* heat and I would be perfectly happy. It wouldn't be enough IPC to catch up with Intel, but it would be enough for the games I play.

When you clock the FX-9590 to 5 GHz constant it isn't slow in real world gaming, it's just ridiculous how much electricity it sucks down and consequently how much heat you have to dump into your room. I didn't use to care because I had a wide open living room with vaulted ceilings for my PC. Now I have it in my bedroom and it's quite noticeable how much it alters the ambient air temperature.
 
XoR said:
comparing in Cinebench R11.5 single thread
FX@5GHz score 1.33
Nehalem @4GHz score 1.35
Similar but FX need 1GHz more to cath up to Nehalem thus it have worse IPC. Nehalem is 25% better here.

FX@4GHz score 6.94
Nehalem @4GHz score 6.77
FX is faster here but it is due to better multi-thread scaling of threads from modules compared to Hyper Threading and not due to better IPC. If we used 4 threads then Nehalem @ 4GHz would easily win with FX @ 4GHz

Then there are games. Bulldozer have inefficient cache design and loose quite a lot of performance due to this. Performance which it cannot regain by having better scaling on its additional threads as games use two to four cores.

Frankly if Bulldozer had Nehalem IPC but could go 5GH, had scaling it have then no one would ever say it is bad CPU.
Click to expand...


Just ran mine @ 4Ghz. Almost 1.4 with slow RAM.



Even my 980BE does 1.34 @ 4.5Ghz. So to say Nehalem doesn't have IPC advantage over Piledriver is just plain wrong. Even Stars has IPC over Piledriver, it just doesn't clock as high on average.
 
Piledriver is definitely no slouch, but yeah...

But why Cinebench 11.5 and not a later version, if I might ask?
 
People get used to a bench and never upgrade because then they can't direct compare to hardware they no longer have.
 
Nobu said:
Piledriver is definitely no slouch, but yeah...
But why Cinebench 11.5 and not a later version, if I might ask?
Click to expand...

A couple reasons. 1st, that's what was used to compare in the post preceding mine. Secondly, because we're talking about CPUs several years old, well find more results on the older version than the newer one. I have and am more than willing to compare on R15 as well.

And here's good old Stars at 4.4Ghz for comparison. What does yours get on multicore for 4 threads?

4400Mhz-2640NB-1466MhzRAM-SP1522.jpg
 
*The*Dark*Lord* said:
I thought this was a ZEN thread and all I can see is people comparing their intel CPUs lol
Click to expand...

Well its more fun to talk about tech that actually exists and its not like we have much else to talk about in the AMD section.

So while we wait for the next round of utter disappointment to arrive...;):D




Plus the AMD boys could post up some figures too.
 
daglesj said:
So while we wait for the next round of utter disappointment to arrive...;):D
Click to expand...

Or hope for something competitive. In my case its hope for an upgrade (not a side grade) from an i7 970 that I purchased in 2011 for $365 just after the bulldozer release. The upgrade must have better single threaded and better multithreaded performance for me to consider upgrading.
 
FlawleZ said:
A couple reasons. 1st, that's what was used to compare in the post preceding mine. Secondly, because we're talking about CPUs several years old, well find more results on the older version than the newer one. I have and am more than willing to compare on R15 as well.

And here's good old Stars at 4.4Ghz for comparison. What does yours get on multicore for 4 threads?

http://img.photobucket.com/albums/v321/FlawleZ/Screenshots/4400Mhz-2640NB-1466MhzRAM-SP1522.jpg
Click to expand...
Cool, just wondering. Multicore I get 3.38pts, but I could swear I got at least 4 in Linux--I'll have to reboot to confirm. I ran the single threaded bench again since everything's settled in Windows land over night, got 1.18.
 
Finally found that old Cinebench from back when I set up my server and briefly had Windows on it.

Turns out I ran R15 though...

14877877026_deb2fb3684_b.jpg


For comparison purposes, dual Xeon L5640's at stock clocks. With two of these, I'm still using less power than a single FX-8120 or FX-8350
 
Yeah, 1.18/3.33 on Linux. On R15 on Linux I get 105/353. At 4.9GHz (unstable) single-core increases to 114. I think the biggest factor is memory speed, so maybe the jump to DDR4 will help a bit?
 
daglesj said:
Plus the AMD boys could post up some figures too.
Click to expand...

It is easy Bulldozer doesn't do single thread well the design won't allow single thread improvements with a margin that it will be competitive.
It is futile to run benchmarks if you know that is something you are not looking for.

Gaming benchmarks under Mantle/DX12/Vulkan is something I could more or less care about. Something that pushes cores over ipc. Then the "discussion" will not be about how bad certain designs/ipc is but how well 8 cores function on technology that matters rather then stuff that really does not.

I don't run cinebench I don't like some of the compiler shenanigans either I seriously have a problem with people yapping about benchmarks that are not optimized for Bulldozer (round peg square hole). You get drawn into absurd corners as the x87 debate which matters not...
 
Last edited:
Pieter3dnow said:
It is easy Bulldozer doesn't do single thread well the design won't allow single thread improvements with a margin that it will be competitive.
It is futile to run benchmarks if you know that is something you are not looking for.

Gaming benchmarks under Mantle/DX12/Vulkan is something I could more or less care about. Something that pushes cores over ipc. Then the "discussion" will not be about how bad certain designs/ipc is but how well 8 cores function on technology that matters rather then stuff that really does not.

I don't run cinebench I don't like some of the compiler shenanigans either I seriously have a problem with people yapping about benchmarks that are not optimized for Bulldozer (round peg square hole). You get drawn into absurd corners as the x87 debate which matters not...
Click to expand...


Well I guess you guys are depressed enough already.
 
Pieter3dnow said:
It is easy Bulldozer doesn't do single thread well the design won't allow single thread improvements with a margin that it will be competitive.
It is futile to run benchmarks if you know that is something you are not looking for.

Gaming benchmarks under Mantle/DX12/Vulkan is something I could more or less care about. Something that pushes cores over ipc. Then the "discussion" will not be about how bad certain designs/ipc is but how well 8 cores function on technology that matters rather then stuff that really does not.

I don't run cinebench I don't like some of the compiler shenanigans either I seriously have a problem with people yapping about benchmarks that are not optimized for Bulldozer (round peg square hole). You get drawn into absurd corners as the x87 debate which matters not...
Click to expand...

relax, is clearly you have never used Cinebench or you don't know what is it, because then you would know it's one of the few benchmark, probably together with 7-Zip benchmark and POV-Ray that actually favor cores over IPC and highly favor Bulldozer in those aspect even to the point it can be tied with sandy bridge or even a FX9590(to speak about stock speed) tied with Ivy Bridge 3770K.(at stock too). if you want to run benchmark to make a FX chip see running at equal levels of sandy and ivy bridge then those are what you should run. Cinebench, POV-RAY and 7-Zip. don;t be so butthurt because your new FX don't benchmark and don't run well =).
 
Cinebench is a fucking joke for benchmarking and I'm surprised people still rely on it so much. It's almost mythological at this point.
 
"Regarding Zen performance, a guy who worked for AMD (at least his linkedin profile says that) and who, as he claims, worked on designing L2 cache for Zen and K12 said that their focus was to be competitive against Intel. He no longer works there but apparently his old colleague who still works there said Zen chips have already been tested and so far "it has met all expectation" and they "haven't found any significant bottlenecks". Apparently they haven't finalized the specifications for the clocks and TDP, but their partners in server market are "very excited".
It's not much detail, but I think if there was a problem from having only 2 AGUs, it would count as a significant bottleneck.
Also this is my first post ever, I just usually lurk here and this is the first time I have something useful to add to the discussion. Please no bully"

http://www.realworldtech.com/forum/?threadid=154302&curpostid=154823
 
NaroonGTX said:
Cinebench is a fucking joke for benchmarking and I'm surprised people still rely on it so much. It's almost mythological at this point.
Click to expand...

I disagree.

It's one of the best synthetic CPU benchmarks, giving a good example of what any platform is capable of, if used by running software to it's maximum ability.

it is also fantastic because it offers single threaded and multi-threaded benches, so you can learn more about what is going on.

Short of in game, or in app benchmarks in the specific App/Game you plan on running, I'd argue that there is nothing else I find more accurately predicting CPU performance.
 
Zarathustra[H];1041941448 said:
I disagree.

It's one of the best synthetic CPU benchmarks, giving a good example of what any platform is capable of, if used by running software to it's maximum ability.

it is also fantastic because it offers single threaded and multi-threaded benches, so you can learn more about what is going on.

Short of in game, or in app benchmarks in the specific App/Game you plan on running, I'd argue that there is nothing else I find more accurately predicting CPU performance.
Click to expand...

This. Cinebench is one of the only benchmarks that provides a metric that scales to almost 100% exact linear increases in CPU and RAM performance for both single and multithreaded.
 
The simple fact that benchmarks which are not open source nor able to compile means that you don't know what it does really ..
 
rvborgh said:
"Regarding Zen performance, a guy who worked for AMD (at least his linkedin profile says that) and who, as he claims, worked on designing L2 cache for Zen and K12 said that their focus was to be competitive against Intel. He no longer works there but apparently his old colleague who still works there said Zen chips have already been tested and so far "it has met all expectation" and they "haven't found any significant bottlenecks". Apparently they haven't finalized the specifications for the clocks and TDP, but their partners in server market are "very excited".
It's not much detail, but I think if there was a problem from having only 2 AGUs, it would count as a significant bottleneck.
Also this is my first post ever, I just usually lurk here and this is the first time I have something useful to add to the discussion. Please no bully"

http://www.realworldtech.com/forum/?threadid=154302&curpostid=154823
Click to expand...

Looking at what is said is not that surprising it is pretty well worded where you know that these things tend to be said to suggest things rather then "prove" things.

We are still a year away from launch of Zen at this point in time no one at AMD will take the risk of talking to much, hoping that some news of AM4 motherboards will clarify the power envelope of Zen..
 
Cinebench R15 is probably a better tool as it showed me the 5025 MHz overclock was not 100% stable (worked fine in R11.5).

KpGQJR8.png


n1xlCoi.png




FlawleZ said:
And here's good old Stars at 4.4Ghz for comparison. What does yours get on multicore for 4 threads?
Click to expand...
4.4 GHz is an exceptionally high overclock for a Phenom II. What cooling was that with?
 
DedEmbryonicCe11 said:
Cinebench R15 is probably a better tool as it showed me the 5025 MHz overclock was not 100% stable (worked fine in R11.5).



Click to expand...

sorry to derail with this, but instability with only 4 thread in that chip at that speed and that voltage isn't anything common. RAM maybe?
 
Araxie said:
sorry to derail with this, but instability with only 4 thread in that chip at that speed and that voltage isn't anything common. RAM maybe?
Click to expand...
I'm not sure if it is the RAM, memory controller, or both. Loosening tRCD from 11 to 12 worked with the CPU-NB at 2211 MHz, but still locks up the system at 2412 Mhz. Raising the CPU-NB voltage from 1.25v to 1.275v did not improve stability. I could always raise the multi .5 instead and crank the voltage but I'm not interested in suicide runs. At this point I'm not fiddling with it more because it's not like I'm going to see drastic gains. AM4 can't come fast enough.:p
 
FlawleZ said:
I get what you're saying but the X5470 is just a Core 2 Quad based Xeon. I purchased an X5450 a while back to use in my DFI 775 board but screwed up the mod and killed both the CPU and motherboard. Probably my biggest fail messing with computers lol
Click to expand...

Shit! That's steep, man. I know the 775->771 mod is a great way to get a cheap kickass CPU on old boards, but damn I didn't know it could go south like that! I've had some pretty large screwups in my computing career (we all have!) so the fact that you can look back and laugh is good!
 
DedEmbryonicCe11 said:
Cinebench R15 is probably a better tool as it showed me the 5025 MHz overclock was not 100% stable (worked fine in R11.5).





4.4 GHz is an exceptionally high overclock for a Phenom II. What cooling was that with?
Click to expand...


Was just air. A Scythe Mugen 2. I actually hit 4.7Ghz over the winter one night just opening my window and bringing the ambient temp down. Wasn't stable other than booting into Windows, SuperPi, and browsing web.

 
Zarathustra[H];1041941448 said:
I disagree.

It's one of the best synthetic CPU benchmarks, giving a good example of what any platform is capable of, if used by running software to it's maximum ability.

it is also fantastic because it offers single threaded and multi-threaded benches, so you can learn more about what is going on.

Short of in game, or in app benchmarks in the specific App/Game you plan on running, I'd argue that there is nothing else I find more accurately predicting CPU performance.
Click to expand...

Actually it isn't really multi-threaded, it is numerous instances of single thread. If it were true multithread it would produce each frame grade one at a time using all threads/cores.

11.5 is not accurate across platforms, only comparing within an architecture. For example looking at clock scaling. Being it used compilers ICC that obviously would favor Intels architecture, it wouldn't give a true depiction of performance outside of Cinema 3D.

R15 is supposed to be free of that issue but haven't seen any real effort to prove one way or another, so cant say.

But this brings up the topic of Architectural software building and how software can be a bigger factor over architecture in determining final performance.

A DOUBLE EDGED SWORD.
 
JustReason said:
Actually it isn't really multi-threaded, it is numerous instances of single thread. If it were true multithread it would produce each frame grade one at a time using all threads/cores.

11.5 is not accurate across platforms, only comparing within an architecture. For example looking at clock scaling. Being it used compilers ICC that obviously would favor Intels architecture, it wouldn't give a true depiction of performance outside of Cinema 3D.

R15 is supposed to be free of that issue but haven't seen any real effort to prove one way or another, so cant say.

But this brings up the topic of Architectural software building and how software can be a bigger factor over architecture in determining final performance.

A DOUBLE EDGED SWORD.
Click to expand...

Whether that is correct or not does not really matter. I have two systems, one I upgraded to a I7 6700k with 32GB of DDR4 and one system that I swapped in my FX 8350 and it has 32GB of DDR3 ram. Although my FX 8350 work system is fast, my I7-6700k is straight up noticeably faster in all tasks. Besides, the 990FX is an ancient chipset and the Z170 is faster even with loaded games off my hard drives.

These things are not coincidental nor unique for me. Both my systems run off of SSD's and loading programs are straight up faster on the I7-6700k system no matter what. (The FX 8350 is no slouch but, it definitely lags behind my new build.)
 
JustReason said:
Actually it isn't really multi-threaded, it is numerous instances of single thread. If it were true multithread it would produce each frame grade one at a time using all threads/cores.

11.5 is not accurate across platforms, only comparing within an architecture. For example looking at clock scaling. Being it used compilers ICC that obviously would favor Intels architecture, it wouldn't give a true depiction of performance outside of Cinema 3D.

R15 is supposed to be free of that issue but haven't seen any real effort to prove one way or another, so cant say.

But this brings up the topic of Architectural software building and how software can be a bigger factor over architecture in determining final performance.

A DOUBLE EDGED SWORD.
Click to expand...

so you made a new account to write all kind of those nonsense crap?..

To make it easy for you, it scale linearly with every kind of thread present in the system; how do yo think multi-thread works?
Assigning one task to each thread: the more threads the faster overall work you will receive. But also, the faster single thread performance will make each task to finish faster, being able to receive a new task in lower time also giving a good overall performance, why so hard to understand?. the whole picture in that image is a single frame and it's divided and processed by the amount of system thread, what do you expect really?. how can you explain the high performance in CineBench in FX chips vs intel chips? according to your "theory" a FX8350 should have the same performance than a FX9590? pure non-sense imagination, if you are going to make a new account appearing to be a genius you are completely wrong.
 
DedEmbryonicCe11 said:
I'm not sure if it is the RAM, memory controller, or both. Loosening tRCD from 11 to 12 worked with the CPU-NB at 2211 MHz, but still locks up the system at 2412 Mhz. Raising the CPU-NB voltage from 1.25v to 1.275v did not improve stability. I could always raise the multi .5 instead and crank the voltage but I'm not interested in suicide runs. At this point I'm not fiddling with it more because it's not like I'm going to see drastic gains. AM4 can't come fast enough.:p
Click to expand...

In my experience AM3+ chips tend to get unstable quickly if you use 4 sticks of ram above 1600Mhz.

With two sticks it's rated for 1866mhz, and overclocks easily above that, but once you add the second two, the official support drops to 1600mhz, and it REALLY doesn't like overclocking above that, at least in my experience.
 
JustReason said:
Actually it isn't really multi-threaded, it is numerous instances of single thread. If it were true multithread it would produce each frame grade one at a time using all threads/cores.
Click to expand...

Semantics really. Splitting up the workload into smaller chunks and running each in a different thread is one way of multithreading :p (There are others too)
 
Araxie said:
so you made a new account to write all kind of those nonsense crap?..

To make it easy for you, it scale linearly with every kind of thread present in the system; how do yo think multi-thread works?
Assigning one task to each thread: the more threads the faster overall work you will receive. But also, the faster single thread performance will make each task to finish faster, being able to receive a new task in lower time also giving a good overall performance, why so hard to understand?. the whole picture in that image is a single frame and it's divided and processed by the amount of system thread, what do you expect really?. how can you explain the high performance in CineBench in FX chips vs intel chips? according to your "theory" a FX8350 should have the same performance than a FX9590? pure non-sense imagination, if you are going to make a new account appearing to be a genius you are completely wrong.
Click to expand...


A lot of AMD fans complain about Cinebench because it performs rather poorly on AMD systems, when in reality it's just showing the true performance deficit the bulldozer-based CMT architectures have compared to Intel current designs.

There are a lot of Reddit conspiracy theories about this, and they are really mostly junk.

The reason AMD's current designs perform poorly on Cninebench is because of an inferior CPU design, not because the benchmark isn't accurate.
 
You must log in or register to reply here.
Back
Top