AMD Ryzen, eventually better than Piledriver/Bulldozer/Vishera!

Quote:

---

Originally Posted by rafal_iB_PL

Latency may not have decrease as much, but bandwidth ever increases, and it's useful in certain applications.

As for CPUs themselves, I'm starting to think we're close to the performance wall for single thread. Multithreaded can always be increased as long as you don't mind extra power draw.

---

What bandwidth? That's more or less theoretical, synthetic number. Those about 3 % applications get SOME benefit from it, but it scales very far from linear.

I think you are one of those who don't get my point: in real world, you wait all the time between usefull (read, write) operations. The higher the clock, the longer (more clocks) you wait. In reality the actual time between operations is almost the same for a decade, hence you can do pretty much the same number of operations per second. The real-world bandwidth and performance is also almost the same.

LOL?

Bandwidth is especially useful whenever you need to process large amount of data at once (e.g., video editing, image editing, games less often, but sometimes too). Latency itself has been reduced when memory controller was moved to CPU die too.

Do you realize that CPUs have cache and don't go to system RAM everytime they need to access data? Once it is filled, trips to main RAM are not needed until you access data that isn't in cache. Less trips to RAM = less impact of latency, but bandwidth to fill cache and its size are still important.

Exactly - the most performance advantage each generation is from accessing RAM as little as possible, hence so many levels of ever-larger CPU cache. All the gigabytes of RAM today are close to "static" data - large amounts just sitting there and every now and than (hundreds-thousands of CPU cycles) something gets read and written.

It is always way more important to have more than enough memory available, than how fast it actually is.

Video editing gets magnitude faster if you use GPU for the computation.

Regarding Ryzen performance beside Vishera,Bulldozer, etc. Have you guys seen the ryzen Blender test?
The new Ryzen(they didnt mention the model) finished the test in 35seconds
I made the test with my FX 6350 3.9Ghz (stock clock) and i got weird timings:

Blender 2.78a:
- Win7 = 3 minutes 27 seconds
- Win10= 2 minutes 54 seconds

Blender 2.7b:
-Win10 = 1 minute 49 seconds
-Linux = 1 minute 26 seconds

The hardware settings were the same.
What is going on? Whats the meaning of this? Doesnt Blender know how to utilise all the CPU power? Is Blender optimized for Linux?

What scores do you all got?

O.o

http://imgur.com/a/3ALE7

Apparently ryzen isn't an upgrade for me. I'm actually confused at how my time is so low...

minus0: it has 10 more threads and 50 % better IPC, I guess it makes a difference :D

@Behemot , indeed, the new AMD 7 cpus has new different architecture, more cache and more "smart" core/threads.
I know my old FX 6350 (3cores-6threads) is slow, but how come the big difference between the Blender versions? I am not talking about ~2 seconds, i am talking about ~40% ~50% difference.
And comparing with the new Ryzen, i think it should be a big gap between them".

Quote:

---

Originally Posted by ridgid13579

O.o

Apparently ryzen isn't an upgrade for me. I'm actually confused at how my time is so low...

---

What is your CPU ? i saw your score, 20seconds, but what is your rig specification?

It is very settings dependent as to what your score is and that is where there is a problem with the threads on TPU which were getting people to compare how fast their CPUs were to Ryzen

Quote:

---

Originally Posted by minus0

What is your CPU ? i saw your score, 20seconds, but what is your rig specification?

---

5960x at 4.5.. usually run it at 4.7 or even 4.8 but it's quite a hot thing :p

Sure it's got 8c 16t but my friend with a 6600k got 59s (granted his also overclocked).