Zen3 reviews are out - hail the new King!

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

azev

Well-Known Member
Jan 18, 2013
769
251
63
Too bad there's not much stock out there right now, I tried waking up early to hit the 9AM ET release and did not managed to get one before it goes out of stock everywhere :(
 

Wasmachineman_NL

Wittgenstein the Supercomputer FTW!
Aug 7, 2019
1,880
620
113
5950X it is for me, but i'm still (impatiently) awaiting the release of a suitable board which is taking for-****ing-ever. Until then I can't build.
 

EffrafaxOfWug

Radioactive Member
Feb 12, 2015
1,394
511
113
Assuming a BIOS update comes out for my X470 board, I'll be replacing my not-even-a-year-old 3950X with a 5950X. The 10-20% gains in single-threaded workloads alone were already more than I could have hoped for.

Most of the review sites are showing pretty astonishing gains of 20% or more, but the performance margin under linux is seemingly even better. And all of this from smarter design - power draw seems to be identical to the Zen2 equivalents. Can't wait to see what the next-gen Epyc's turn in to - it's not every day you can legitimately say "wow, I wish we could have waited before buying all these Epyc Rome servers for work"! :D

Absolutely top-drawer stuff, well done AMD. To go from bulldozer to this in the space of three and a bit years is frankly astonishing.
 

Mithril

Active Member
Sep 13, 2019
356
106
43
So I've got a 3800x right now (due to a nice sale it was close to the 3700x price last year), and while I do some things on my desktop where single-thread is king, I feel like it would be silly to drop core count. I'm hoping for a 5700(x).

"But you could 'just' spend 100 more for the 5900x!". True but, same TDP for more cores in theory means more slower cores in some workloads, plus you go back to a split L3 across chiplets and a higher memory latency (due to 2 chiplets).

If I had money and time, I'd buy one of each and benchmark the workloads I care about going "as fast as they can". :)
 

zer0sum

Well-Known Member
Mar 8, 2013
849
474
63
So I've got a 3800x right now (due to a nice sale it was close to the 3700x price last year), and while I do some things on my desktop where single-thread is king, I feel like it would be silly to drop core count. I'm hoping for a 5700(x).

"But you could 'just' spend 100 more for the 5900x!". True but, same TDP for more cores in theory means more slower cores in some workloads, plus you go back to a split L3 across chiplets and a higher memory latency (due to 2 chiplets).

If I had money and time, I'd buy one of each and benchmark the workloads I care about going "as fast as they can". :)
I'm not sure what your workloads are, but from the reviews I've read the 5900X smokes the 3XXX series, and even the 5800X is killing it.
These new Zen3's are pretty amazing all round :D
 

Wasmachineman_NL

Wittgenstein the Supercomputer FTW!
Aug 7, 2019
1,880
620
113
What motherboard are you waiting for?
Crosshair VIII Dark Hero. As much that I hate ASSUS for their slow as hell BIOS updates and Supreme(me)FX they do have the best BIOS and boards when it comes to features and overclocking. (The Stilt's Fmax enhancer and dynamic OC is one of them)

Really leaning towards the CH8 non-WiFi though, because I can get it for €335 after cashback and WLAN doesn't work under Windows 7 anyways. (not without downgrading the WLAN card to a 8260AC lol)

EDIT: I'll wait until the 21th, if the C8DH isn't out by then i'm going with a CH8 non-WiFi because of the cashback.
 
Last edited:

3nodeproblem

Member
Jun 28, 2020
49
12
8
I invested heavily in Zen2 quite recently so I'm doing my best not to look at all this too much but wow, the 5600X looks amazing for cluster builds at 65WTDP.
 

zer0sum

Well-Known Member
Mar 8, 2013
849
474
63
Crosshair VIII Dark Hero. As much that I hate ASSUS for their slow as hell BIOS updates and Supreme(me)FX they do have the best BIOS and boards when it comes to features and overclocking. (The Stilt's Fmax enhancer and dynamic OC is one of them)

Really leaning towards the CH8 non-WiFi though, because I can get it for €335 after cashback and WLAN doesn't work under Windows 7 anyways. (not without downgrading the WLAN card to a 8260AC lol)

EDIT: I'll wait until the 21th, if the C8DH isn't out by then i'm going with a CH8 non-WiFi because of the cashback.
Did you take a look at the Pro-WS-X570-ACE?
 

Mithril

Active Member
Sep 13, 2019
356
106
43
I'm not sure what your workloads are, but from the reviews I've read the 5900X smokes the 3XXX series, and even the 5800X is killing it.
These new Zen3's are pretty amazing all round :D
SO interestingly *some* of the benchmarks that ARS did do bear out the 5800x being faster than even the 5950x assuming the thread count is "right". Based on the latency penalty from chiplet to chiplet and the additional memory latency you get from 2 chiplets VS 1 this does make sense. Now, it's possible future schedular or other updates help with that.

I'm curious what benchmarks will look like with more focused testing on overclocking (both PBO and manual) as it seems that ARS was testing fully at stock (which is also valuable!), as well as how well they respond to better cooling (a trend in modern CPUs and GPU is that they tend to auto-boost higher the cooler you keep it).

Anyways, yes great upgrade all around; now we shall just see about availability and if we can actually buy at MSRP.
 

josh

Active Member
Oct 21, 2013
615
190
43
Can someone help me understand the fuss about extra clock speed above 3GHz?

Is there any discernable difference in everyday tasks? Isn't increasing cores more beneficial than increasing clock speed?

E5-2678v3 all core turbo hack setup would only cost about $200 with 12c/24t @ 3.3GHz. I'm finding it hard to justify getting a Ryzen setup for 3-4x the price.
 

alex_stief

Well-Known Member
May 31, 2016
884
312
63
38
Seriously? Compare single- and all-core performance between an old v3 Xeon, and a Ryzen 5000.
Not all tasks, especially those often referred to as "everyday" usage, scale perfectly on many cores. That's why high single-core performance is still nice to have, and often preferred over very high core counts.
 

josh

Active Member
Oct 21, 2013
615
190
43
Seriously? Compare single- and all-core performance between an old v3 Xeon, and a Ryzen 5000.
Not all tasks, especially those often referred to as "everyday" usage, scale perfectly on many cores. That's why high single-core performance is still nice to have, and often preferred over very high core counts.
Obviously a 3.3GHz single core will struggle compared to 4.6GHz in benchmarks but is there really any discernable difference is my question.

Also, why does the date of release matter for the CPU? Specs wise, the Xeon matches the Ryzen in everything except the clock speed.
 

Wasmachineman_NL

Wittgenstein the Supercomputer FTW!
Aug 7, 2019
1,880
620
113
Obviously a 3.3GHz single core will struggle compared to 4.6GHz in benchmarks but is there really any discernable difference is my question.

Also, why does the date of release matter for the CPU? Specs wise, the Xeon matches the Ryzen in everything except the clock speed.
Clock speed != performance. A 5950X will decimate anything Intel made in the past 10 years short of a 8180.
 

Mithril

Active Member
Sep 13, 2019
356
106
43
Obviously a 3.3GHz single core will struggle compared to 4.6GHz in benchmarks but is there really any discernable difference is my question.

Also, why does the date of release matter for the CPU? Specs wise, the Xeon matches the Ryzen in everything except the clock speed.
No, no it doesn't. Both Intel AND amd generally make "IPC" improvements each generation, meaning for the same clock speed the processor does more work. Depending on workload fewer faster (calculation) cores beat more slower cores. This is true for many games and consumer programs where even with multi threading there's still a limit to HOW threaded the program is/can be. This also true for RAM where speed *can* beat more channels, or latency *can* beat speed, and more ranks (not channels, ranks) can beat both. One of the other differences is the amount of cache on the CPU, again depending on the workload a CPU with more cache (or faster cache) could beat a different CPU that it would theoretically "lose" to. One of the big uplifts for ZEN3 is the unified cache where the chiplet of 6-8 core has a shared L3 cache of 32MB.

Newer CPUs also tend to be on a smaller process node meaning more transistors per MM^2, generally less power as well, plus they have new instruction sets which *can* provide an advantage.