Zen 2 Review Summary

DuronBurgerMan

[H]ard|Gawd
Joined
Mar 13, 2017
Messages
1,340
Been reading the reviews and benches this morning, and I think a clear picture is emerging.

Single thread

Single-threaded performance has been hugely improved over Zen/Zen+, and clock-for-clock it's equal or maybe a pubic hair better than Skylake. However, the high boost clocks of the 9700k and 9900k still deliver an overall single-threaded win for Intel. Correspondingly, Intel retains the gaming performance crown, though the lead is much slimmer than it has been, and you have to go down to 720p to really push it outside the margin of error much of the time. Also, I note a few outright wins for Zen 2 here and there (as I suspected). But the 9700k and 9900k win more often. And frankly, the Intel wins are a bit more frequent than I thought they might be. I was hoping for a more 30/70 or 40/60 split, but it appears to be more like a 20/80 split, roughly.

Still, AMD has a lot to be proud of here. It's clear they did a buttload of work on this.

Zen 2 is also winning some single-thread productivity/content creation benchmarks. I noted a mild win in Cinebench 20, and in some of the web browser metrics. But as with gaming, the 9700k and 9900k still win more single threaded benchmarks. Notably, single-threaded Photoshop tests still mildly favor Intel. Overall this paints a picture of single-threaded performance within ~5% of Intel and identical-to-slightly-better IPC. It's also worth noting that single-threaded application performance is more competitive than single-threaded gaming performance.

Worth noting that PCWorld ran Puget System's test suite for Photoshop, which resulted in a win for the 3900X against the 9900k, but that is a mix of single and multithreaded tests. Still, no AMD CPU has beaten the 9900k here until now.

Multithread

In multithreaded workloads, Zen 2 crushes it. Even the 3700X is extremely close to the 9900k at times, and demolishes the non-HT 9700k. The 3900X is in a league of its own - the 9900k can't touch it, overall, and only wins a handful of specific benchmarks. AMD's superior SMT gains really shine here. The power efficiency of the 3700X makes it a more efficient part than the 9900k even when it doesn't outright win, but I saw a number of outright wins for the 3700X over the 9900k anyway (the 9900k still wins overall vs. the 3700X though).

There's not much else to say here. It's an obvious, huge, and clear win for Zen 2, and it lacks many of the caveats that came with saying Zen/Zen+ was 'better at multithreaded tests, except X, and Y, and Z." The win is conclusive, and needs no disclaimer. Zen 2 makes huge gains even in benchmarks that traditionally did not favor Zen/Zen+ (Geekbench being an excellent example).

Overclocking

Overclocking is a bust. I've noted no good all-core overclocking results in any of the reviews. However, it is consistently mentioned that heat is the big problem here, more so than voltage. Techpowerup, for instance, recorded a crappy 4.0 GHz OC (to be fair, this was by far the worst result seen this morning), but at a voltage of only 1.2v, because heat generation was overloading the stock cooler past that - so don't try OCing this on the stock cooler, as clearly that is a bad idea. Other results were more along the lines of 4.3-4.4 GHz for the 3900X, but again, heat came into play. I suspect a ~4.5ghz all core ceiling for Zen 2 even on a good AIO or water cooler.

This does indicate that a good water cooler may achieve better OC results than we've seen this morning, and we should see more on this in the coming days as people experiment with them. Nonetheless, it looks like the PBO +200MHz overboost is probably the best way to go with overclocking anyway, since single-threaded performance is still the weakest link (albeit not bad anymore), this will address that better than an all-core OC. Multi-threaded performance by Zen 2 is so dominating that an all-core OC is probably a waste of time anyway.

Missing 3800X review

I've seen no benchmarks on the 3800X this morning. Since the 3700X loses by a bit (but wins on efficiency) against the 9900k in outright performance, it would be interesting to see if the 3800X correspondingly wins by a hair, or effective ties the 9900k. As it stands, the price/performance/efficiency sweet spot still favors the 3700X, but it would be nice to see if AMD can claim the multithreaded application crown on a per-core basis. The 3700X doesn't quite get there. The 3800X might. I still wouldn't bother with a 9900k even as a pure gamer, though. The 3700X is close enough, cheaper, and comes with a cooler. And the 9700k still shows slightly better performance in gaming in many cases than the 9900k.

Memory Performance

If overclocking results were a bust, the memory OC performance results are excellent. Most reviews have noted that Zen 2 does not share the bizarre memory OC problems of Zen and Zen+. One quote from Techpowerup was "...and can confirm that DDR4-4000 was not any more difficult than on Intel." So perhaps the days of nitpicking your RAM kit with Ryzen are over. Much to my personal relief, really.

Linux Results

There was some kind of bug reported by Phoronix on some Linux distro versions (specifically the very latest releases). They weren't sure if it was a chipset thing or something else. So testing was done one version back.

However, Linux results were otherwise a huge win for Zen 2. The 3900X crushed the 9900k, beat out AMD's TR 2950X, and almost beat out the 7960X. The geometric mean of all tests is on the last page.
 
Last edited:
how would you say a 3600x + rx 5700 compare to a similar intel/nvidia pairing? these higher end parts look good, but i may need to build a system just a step or two down from there. focused on 1080p gaming mostly.
 
how would you say a 3600x + rx 5700 compare to a similar intel/nvidia pairing? these higher end parts look good, but i may need to build a system just a step or two down from there. focused on 1080p gaming mostly.

For gaming, the 9600 still wins, but not by a huge margin and the 3600 is a better overall performer. It's a better value CPU but if you only care about games and nothing else maybe the 9600 is still a better option. The 9600 also seems to OC better but i wanna see someone really try to push the 3600. A lot of the reviews aren't particularly trying (i've seen a few that were using 2666 memory, like, the hell). Still, word seems to be the Ryzen 3000 just isn't a monster OCer.

Navi i've seen reviews where it goes neck and neck with a 2060 super and others where it barely wins over a regular 2060. Hard to say, will be worth it to wait more and see better benchmarks.
 
The overclocking portion is disappointing. I'm going to upgrade, but I'm going to wait until this aspect is understood a bit better by the community.
 
yea i pretty disappointed with the overclocking headroom. Maybe its just early bios revision....I mean PBO can hit 4.5ghz it seems so I am wondering if its just a matter of tweaking etc. Great processors, but AMD needs to get those clocks up.
 
I think the only case for a 9600k at this point, is budget high refresh gaming. And only if that is your absolute main focus.
 
DerBauer has a comprehensive look at O/C potential:

tl;dw: Expect max 4.4Ghz all core as standard with enough cooling. Delidding is not worth it since you may net only 4C temp gains (or about 5-10Mhz).

Edit: I accidentally a word or 2. Max 4.4 Ghz, and 5-10 Mhz more speed with delidding at best.
 
Last edited:
DerBauer has a comprehensive look at O/C potential:

tl;dw: Expect 4.4Ghz all core as standard with enough cooling. Delidding is not worth it since you may net only 4C temp gains (or about 25Mhz).


How did you get to that conclusion? He literally said 4.2-4.3ghz is probably all you will get with AIO/watercooling.
 
I think the only case for a 9600k at this point, is budget high refresh gaming. And only if that is your absolute main focus.
how does it perform better with high refresh, since that's actually why i'm sticking with 1080p, i still have some pretty ridiculous reflexes so i'm trying to keep things snappy.
 
How did you get to that conclusion? He literally said 4.2-4.3ghz is probably all you will get with AIO/watercooling.

Normally I'd believe him, but we're seeing a lot of 4.4 GHz results in the reviews. Of course, a lot of 4.25 results too. I dunno. I think we'll have to wait for these to get into private hands in numbers to get a good feel for the OC abilities - but it doesn't look very good.
 
how does it perform better with high refresh, since that's actually why i'm sticking with 1080p, i still have some pretty ridiculous reflexes so i'm trying to keep things snappy.
Look at the gamernexus Ryzen 3600 review. Games at 1080p on an overclocked 9600k, beat the Ryzen 3600 by a fairly healthy amount.
 
I made this post in a prior thread, but it's worth noting that unless your budget is unlimited, you are better off getting a Ryzen 3000 vs a 9900k, then toss the savings from not getting Intel into going up a GPU bracket, which again gives you more gaming performance than if you brought the 9900k. Tldr: unless your fucking rich go AMD and Nvidia. Good job AMD, you just fucking ate Intel's sandwich. Intel may technically be king, but at this point, unless Intel drops pricing hard, go Ryzen and toss the savings into GPU.
 
Worth noting that Hardware Unboxed did extensive latency, IPC, and memory testing.

Some points:

1. Core-to-core latency has improved a lot with Zen 2, but is still about 10% worse than Skylake at equivalent clocks. This probably explains the remaining gaming performance deficit.
2. IPC measured in Cinebench shows a huge gain for Zen 2, and outpaces Skylake significantly - by almost 15%. It's enough to overcome the clockspeed deficit and give AMD the Cinebench R20 single thread crown.
3. The memory controller has been hampered by half on write performance, but somehow this does not appear to cause any real world problems, since it brutally crushes Zen+ everywhere - it shows up in synthetics as worse than Zen+, though.

This likely means that in throughput tasks, Zen 2 is about 15% better than Skylake on IPC. And in latency-sensitive tasks, it's about 10% worse on IPC. Given the clockspeed deficit compared to the 9900k, this means that Zen 2 can sometimes still win in absolute single thread throughput performance (and its superior SMT performance means it wins more often in multithreaded IPC than single threaded), but will generally lose in gaming - at least with games that don't have high thread awareness, or are latency sensitive. Where high thread awareness exists, Zen 2 can still pull a win sometimes. The PCGH benchmarks shows Zen 2 winning 1 out of 5 game benchmarks, getting crushed in 1 out of 5, and hanging just a hair below the 9700k in the remaining 3. This appears to be typical.
 
It's a shame no one seems to have gotten a 3800x to be able to review it. I'm quite interested in seeing some benches because, given the very small gap between the 3700x and 3900x in gaming, I'm struggling to see exactly what the point of the 3800x would be. Unless it magically clocks a good bit higher than a 3700x (doubtful as heat seems to be a limiting factor here), I don't see why anyone would buy it.

Edit: I was able to find a video posted to Reddit that includes the 3800x. If it is at all accurate, it looks like the 3800x maintains ~100mhz boost clock over the 3700x but fps is essentially identical. Definitely doesn't seem worth the extra money or extra TDP. Obviously, take this one with a grain of salt until we see more reviews.

 
Last edited:
I'll be picking one up in three months or so. It wins everywhere that it needs to win. And the gaming losses are tiny like within five frames. Idgaf about something within the margin of error when the price is that amazing.
 
Do these have a one or two core higher turbo or does PBO just go for max all core oc?
 
Going to pick up a 3950X in September. Can't wait. My plan is to watercool it with 2-3x 480mm radiators in a CaseLabs cube and push it as far as I can.
 
Very surprised to see such low all core overclock with the new 7nm process.

Expect this to win solid market share for data center and cloud market due to power efficiency and price vs Intel.

Hopefully TSMC can tweak the process for next 7nm+ release to up the max operating frequency and close that gap to 5.0 Ghz 9900K.
 
Sadly I didn't read the reviews before jumping on board the hype train. I'll be side/downgrading from an 8700K to a 3700X. Newegg won't let me cancel the X570 board I ordered (even though it hasn't even shipped) so I guess I'm in it for the long haul now. Given how long I've been at this game I should have known better.
 
What I’m curious about right now:


PUBG running at 1920x1080 & 2560x1440


Frame rate differences in real time playing with a 3900X vs 9900K

Or 3700 for example vs comparable intel

Both with a 1080Ti
 
Speaking as someone who has been overclocking PCs for the better part of 20 years, starting when I was 14-15, I don't see the "low OC headroom" as a problem. Yes, overclocking is fun, and squeezing more performance out of more mainstream and budget hardware has traditionally been a good way to get the best value, but the fact that you can buy these processors, pop them into the motherboard and get 99% of the performance potential by toggling on a BIOS switch with complete reliability really is something I think we should be praising and not viewing as a negative. Especially for AMD, who is selling to all users and not just people who are on HardOCP, making it easy to get the most of your purchase is a great selling point.
 
Sadly I didn't read the reviews before jumping on board the hype train. I'll be side/downgrading from an 8700K to a 3700X. Newegg won't let me cancel the X570 board I ordered (even though it hasn't even shipped) so I guess I'm in it for the long haul now. Given how long I've been at this game I should have known better.

Just return it? I returned an unused mobo to Newegg about a year ago with zero issues and no restocking fee.
 
Speaking as someone who has been overclocking PCs for the better part of 20 years, starting when I was 14-15, I don't see the "low OC headroom" as a problem. Yes, overclocking is fun, and squeezing more performance out of more mainstream and budget hardware has traditionally been a good way to get the best value, but the fact that you can buy these processors, pop them into the motherboard and get 99% of the performance potential by toggling on a BIOS switch with complete reliability really is something I think we should be praising and not viewing as a negative. Especially for AMD, who is selling to all users and not just people who are on HardOCP, making it easy to get the most of your purchase is a great selling point.

OT: I agree for the most part. Having just being able to stick your CPU in and maybe changing a couple settings around while getting 99% of max performance is nice. But, AMD is advertising boost clocks that nobody seems to be really hitting. I think we might see this improve IF it's a bios implemented lock vs a lock on the CPU's themselves. Seems like there's some super aggressive temperature limits on these CPU's right now

A bit off topic on the overclocking part. I remember my first real OC was my W3520 (i7 920) to 4.2ghz, from a base of 2.66ghz! But that took a while and required a full custom loop, my i7 3770k took less than half the time although it was much less. Giving max performance out of the gate without requiring OC feels like a benefit to 95%+ of consumers who get the most performance without having to get exotic coolers and going through crazy trouble shooting.

I'll be waiting a bit longer, but a cheap B450 for $80-$130 + a 3600X is a pretty good deal.

Also, I wonder when we will see people showing more memory overclocking on these new ryzens.
 
Speaking as someone who has been overclocking PCs for the better part of 20 years, starting when I was 14-15, I don't see the "low OC headroom" as a problem. Yes, overclocking is fun, and squeezing more performance out of more mainstream and budget hardware has traditionally been a good way to get the best value, but the fact that you can buy these processors, pop them into the motherboard and get 99% of the performance potential by toggling on a BIOS switch with complete reliability really is something I think we should be praising and not viewing as a negative. Especially for AMD, who is selling to all users and not just people who are on HardOCP, making it easy to get the most of your purchase is a great selling point.

No OC headroom is something that should be praised? Wtf are you smoking buddy. I've seen some weird weird mental gymnastics of logic when ones favorite company does well but this one is definitely up there in the WTF stratosphere.
 
So in summary, if you game at higher resolutions and want to get it for gaming as they are marketing it, better off keeping your current pc if it was built in the last 6 years and getting a 2080ti.
 
Speaking as someone who has been overclocking PCs for the better part of 20 years, starting when I was 14-15, I don't see the "low OC headroom" as a problem. Yes, overclocking is fun, and squeezing more performance out of more mainstream and budget hardware has traditionally been a good way to get the best value, but the fact that you can buy these processors, pop them into the motherboard and get 99% of the performance potential by toggling on a BIOS switch with complete reliability really is something I think we should be praising and not viewing as a negative. Especially for AMD, who is selling to all users and not just people who are on HardOCP, making it easy to get the most of your purchase is a great selling point.

No OC headroom is something that should be praised? Wtf are you smoking buddy. I've seen some weird weird mental gymnastics of logic when ones favorite company does well but this one is definitely up there in the WTF stratosphere.

For people who tinker and know generally what they are buying, no OC headroom is bad I guess, but for the vast majority of consumers, especially the full system buyers, if they don't ever go near the Bios, having OC headroom is absolutely wasted so if AMD makes it useless to manually overclock, then those people benefit a lot more than the guy who buys a 9900k but then never overclocks it and is running at stock.
 
For people who tinker and know generally what they are buying, no OC headroom is bad I guess, but for the vast majority of consumers, especially the full system buyers, if they don't ever go near the Bios, having OC headroom is absolutely wasted so if AMD makes it useless to manually overclock, then those people benefit a lot more than the guy who buys a 9900k but then never overclocks it and is running at stock.

And I promise you the vast majority of consumers, even the 9900k and 2080ti buyers probably won't overclock. For those people, buying a Ryzen and a 2080ti is a no-brainer in terms of cost, performance, and out of the box tinkering.
 
And I promise you the vast majority of consumers, even the 9900k and 2080ti buyers probably won't overclock. For those people, buying a Ryzen and a 2080ti is a no-brainer in terms of cost, performance, and out of the box tinkering.
Intel stated that only 5% of people who buy K-series processors overclock them.
 
OT: I agree for the most part. Having just being able to stick your CPU in and maybe changing a couple settings around while getting 99% of max performance is nice. But, AMD is advertising boost clocks that nobody seems to be really hitting.
I think you are confused. Boost clocks are not all core overclocks. Boost clocks are not advertised for all core. These are hitting their boost clocks just fine.
 
Last edited:
Back
Top