AMD launches Zen+ 12nm Ryzen and X470 motherboards

[richaron]

Has anyone tested the 2000-series for the linux compilation bug of the 1000-series?

The Linux segfault bug was apparently fixed in silicone around week 25 of 2017. My 1700 was made week 28 and with limited testing I haven't been able to replicate the issue.

 

[tamz_msc]

The CPU results and mini-drama of AT's results has been largely hashed out, but what about Motherboards? Not just reviews of 470's, but how does 470 compare to 370?

One quick review(link below) of my last question seems to indicate little difference, performance wise, but the review was quite limited. Further testing of this comparison seems warranted.

370v470

You can expect to shave off a bit from memory latency with Summit Ridge and X470. Otherwise, they're basically identical, except for chances of running >3200 MHz memory which is higher with the Pinnacle Ridge + X470 combo.

 

[Atari2600]

The CPU results and mini-drama of AT's results has been largely hashed out

So the conclusions were?

==> Testing at rated DRAM frequencies (2666 MHz for 8700k, 2993 MHz for 2700X)
==> Free (stock) AMD cooler, budget Intel cooler
==> Win10 Balanced profile (not Ryzen balanced)
==> All latest patches & microcode applied to both platforms


Which meant the results are valid results, just the input settings are somewhat different (and arguably more legitimate) than others?

 

[sandorski]

You can expect to shave off a bit from memory latency with Summit Ridge and X470. Otherwise, they're basically identical, except for chances of running >3200 MHz memory which is higher with the Pinnacle Ridge + X470 combo.

I wish they had run those tests with 2700 in both mobos.

 

[exquisitechar]

https://www.reddit.com/r/Amd/comments/8dkruc/xfr_is_on_the_chip_not_the_board_get_whatever/dxo5k66/

Interesting post on reddit.

Traditional all core overclocking does seem useless like some of us had thought based on the early leaks, but I can't wait learn more about this.

 

[EXCellR8]

Anyone snap some pics of their prism installed yet?

 

[sandorski]

So the conclusions were?

==> Testing at rated DRAM frequencies (2666 MHz for 8700k, 2993 MHz for 2700X)
==> Free (stock) AMD cooler, budget Intel cooler
==> Win10 Balanced profile (not Ryzen balanced)
==> All latest patches & microcode applied to both platforms


Which meant the results are valid results, just the input settings are somewhat different (and arguably more legitimate) than others?

Still needs verification and further investigation, but that seems to be a summary of where we're at. I just feel we're all just spinning our wheels on that discussion right now and that motherboards are being overlooked.

 

[Christopher Bohling]

You can expect to shave off a bit from memory latency with Summit Ridge and X470. Otherwise, they're basically identical, except for chances of running >3200 MHz memory which is higher with the Pinnacle Ridge + X470 combo.

Bitwit did say that he had just as easy of a time running the 3400Mhz kit at rated speeds on the X370 board, to be fair. Which is honestly something I was worried about going forward - I don't plan to upgrade this gen but I wouldn't want my future Ryzen on 7nm limited due to poor RAM support.

 

[Dasa2]

So the conclusions were?
==> Testing at rated DRAM frequencies (2666 MHz for 8700k, 2993 MHz for 2700X)
==> Free (stock) AMD cooler, budget Intel cooler
==> Win10 Balanced profile (not Ryzen balanced)
==> All latest patches & microcode applied to both platforms
Which meant the results are valid results, just the input settings are somewhat different (and arguably more legitimate) than others?

fairly much although we don't know if the patches are making any difference a this stage
its more than 2666 vs 2933 its also the affect 4 sticks of ram have on sub timings vs two
considering anandtech benchmarks show a 150% increase between ryzen 1600x and 2600 in rocket league and the civ benchmarks are well below what my 6700k manages at the same detail settings i believe further investigation is required so it will be interesting to see what they come up with

nice info on the mb so ti looks like 470x has lower latency or is this just a more mature bios at work running tighter sub timing? otherwise there seems to be little difference with the main boost to memory clock speed coming from improvements the cpu memory controller itself? it will be interesting to see more tests with b350 included and how well vrm stand up without airflow due to aio\tower cooling

 

[Tup3x]

I really don't know what to do... I'd finally want to upgrade but the memory prices are crazy at the moment and I'd really like to get 32GB this time (but it costs as much as mobo+2700X). Then I also wonder if DDR5 and PCI-E upgrades are around the corner... So far it I haven't felt like my 3770K @ 4,4 GHz has been a bottleneck in games but but... I do not know.

 

[Jimzz]

I really don't know what to do... I'd finally want to upgrade but the memory prices are crazy at the moment and I'd really like to get 32GB this time (but it costs as much as mobo+2700X). Then I also wonder if DDR5 and PCI-E upgrades are around the corner... So far it I haven't felt like my 3770K @ 4,4 GHz has been a bottleneck in games but but... I do not know.

Yea I have a 3770 in the 4Ghz range I like to upgrade. No major reason for other than the upgrade bug biting.

I was waiting for Zen+ for the same reason I waited for Ivy. Fix the first gen bugs and tweak the low hanging fruit. If memory, and GPU to a extent, would come back down to reasonable levels I would upgrade now.

The reviews of Zen+ make me want to upgrade even more now.

 

[PingSpike]

Pretty much same place as Tup3x and Jimzz. I want to go to 32GB of ram now since I'm already running 24GB of DDR3 with ivybridge but its a large outlay of cash to step up to those 8 cores. Worse, it looks like high end ram is actually worth buying these days.

 

[krumme]

Pretty much same place as Tup3x and Jimzz. I want to go to 32GB of ram now since I'm already running 24GB of DDR3 with ivybridge but its a large outlay of cash to step up to those 8 cores. Worse, it looks like high end ram is actually worth buying these days.

Well what matters is total cost. I think the 2700x is the better cpu than the 8700k but the 8700k is simply cheaper due to lower cost for the ram you would use especially if you use 32GB. Depends on cooler cost ofc.

 

[IllogicalGlory]

So the conclusions were?

==> Testing at rated DRAM frequencies (2666 MHz for 8700k, 2993 MHz for 2700X)
==> Free (stock) AMD cooler, budget Intel cooler
==> Win10 Balanced profile (not Ryzen balanced)
==> All latest patches & microcode applied to both platforms


Which meant the results are valid results, just the input settings are somewhat different (and arguably more legitimate) than others?

Personally, I’m still unconvinced of the gaming results, given that they match the bizarre results from the original 8700K review, where an i5-7400 is 30% faster than the 8700K in Rise of the Tomb Raider.

https://www.anandtech.com/bench/CPU/1827
I don’t think that an i3-8350K is 50% faster than an 8700K or that a Pentium G4560 is 10% faster.

One other thing: where does AT say which power profile it used for Ryzen?

 

[IEC]

If anything, it exposes the complete lack of scientific method (or in some cases, any reasonable method at all) when it comes to reproducible testing.

1) Every CPU must absolutely be up-to-date microcode and Windows updates and benchmarks re-run. It's pretty clear things change over time so any reviewer re-using old results to compare a new processor is suspect.
2) Looking at the aggregate of reviews, i7-8700K still wins by a small margin in 1080p gaming for average FPS, but not always for 99th percentile frametimes. 1440p/4K no substantial difference, might as well go 2700K if you need the threads.
3) I am unable to reproduce TechRadar's results with 0x84 microcode and latest Windows 10 Pro updates so far, but I am manually OC'd to 4700MHz no AVX offset with all kinds of power limits disabled/maxed out. It's quite possible they were throttling due to either cooler or motherboard (strict power limits), or maybe Windows power management settings (I am on High Performance/all power saving features disabled).

 

[tamz_msc]

I wish they had run those tests with 2700 in both mobos.

The difference isn't that pronounced in term of the performance scaling you get with increased frequency. With Pinnacle Ridge though there is a chance that the IMC might be better, so you can try pushing lower latency.

Bitwit did say that he had just as easy of a time running the 3400Mhz kit at rated speeds on the X370 board, to be fair. Which is honestly something I was worried about going forward - I don't plan to upgrade this gen but I wouldn't want my future Ryzen on 7nm limited due to poor RAM support.

I don't see why you'd be limited by the memory on Zen 2 in the future if you're buying/already have a good B-die kit right now.

 

[lixlax]

A nice review (in french) which shows all the way we've come from- since Yorkfield/K10 to Ryzen 2/Coffee Lake.
http://www.comptoir-hardware.com/ar...-test-amd-x470-r5-2600x-r7-2700x.html?start=3

 

[Hitman928]

Personally, I’m still unconvinced of the gaming results, given that they match the bizarre results from the original 8700K review, where an i5-7400 is 30% faster than the 8700K in Rise of the Tomb Raider.

https://www.anandtech.com/bench/CPU/1827
I don’t think that an i3-8350K is 50% faster than an 8700K or that a Pentium G4560 is 10% faster.

AT uses the DX12 path with ROTR which has been known to behave very weirdly, in particular it seems that CPUs with more than 4 cores suffer significant performance degradation in DX12 compared to running in DX11.

 

[alexruiz]

Ok,
I have been reading the reviews, and a few things that I want to point it out:

• Motherboards used for Ryzen 2000 vary by sites, with most of them using the MSI X470 M7 WiFi. Anandtech used the Asus X470 Crosshair Hero VII. Tech Report used the Gigabyte Aorus Gaming 7.
  • Like Ryzen 1000 series launch, performance between boards is showing a lot of variation.
  • Back in the Ryzen 1000 launch, Gigabyte and AsRock were the best performers.
  • Seems that Asus is the one leading now for Ryzen 2000.
  • The MSI seems to be the slowest, and it is also the ones most used for reviews... Keep the board in mind.
• The versions of windows used for testing also vary a lot.
  • One site (I don't remember which one, but I think it was Tom's) said they used Windows 10 anniversary! That is a 2 year old windows (1607, 14393)
  • Using any version of windows 10 other than fall creators update (1709, 16299) should disqualify the results, simply because ALL PCs not running 1709 will get updated to 1709, like it or not

• Ryzen has been performing better and better with newer versions of windows 10. A Ryzen 1000 series does better on windows 10 1709 than on windows 10 1511. Again, any review NOT conducted on Windows 10 1709 should be immediately disqualified.
• Guess what? Meltdown / Spectre patches are being focused on the newer versions of windows first. You tested an older version of windows 10 than 1709, the patches are probably not there yet.
• As usual, most sites omitted info on BIOS version used, chipset drivers used, graphics drivers used, windows build update (16299.248, 16299.371, etc)

While the Anandtech numbers are being questioned, they are NOT out of line. They used what is currently the best performing board for Ryzen, running the OS that extracts the most of Ryzen, applying all the patches that affect Non-Ryzen platforms the most... Their numbers trend match the configuration used, so it makes sense they had to get the best Ryzen numbers.

So please, if you like a review because it showed your platform of choice performing as you WANTED it to perform, analyze if it truly reflects real life test setup.

 

[guskline]

Just received my 2700x and was able to go home from work over the lunch hour and install it in my Crosshair VI Hero rig below.

I kept it stock and my memory (Flare-X 3200) to the 3200 mark with dram vcore at 1.38. I was also running memory at 3200 for the 1800x at 4 Ghz

CPU is very smooth and slightly faster than my 1800x which I OC'd to 4GHZ on all 8 cores
For Cinebench R15 the 1800x scored @ 1760 while the 2700x stock scored 1792.

Aida64 Queen 1800x=92300 while the 2700x stock was 93300.

I will run more test tonight.

I am using the 6004 BIOS from Asus officially posted on their website today for my mb.

 

[IllogicalGlory]

While the Anandtech numbers are being questioned, they are NOT out of line. They used what is currently the best performing board for Ryzen, running the OS that extracts the most of Ryzen, applying all the patches that affect Non-Ryzen platforms the most... Their numbers trend match the configuration used, so it makes sense they had to get the best Ryzen numbers.

So please, if you like a review because it showed your platform of choice performing as you WANTED it to perform, analyze if it truly reflects real life test setup.

What is it about the AT review that's so good for Ryzen, or shows the effect of the Meltdown/Spectre bugs? Yes, it takes a lead in a few of the gaming tests over the 8700K; winning slightly in Shadow of Mordor and solidly in GTA V (though Anandtech also shows the 8400 beating the 8700K by a similar margin in the launch review, which of course makes no sense), while winning by a lot in Rise of the Tomb Raider, (though the 8350K crushes them both) and winning by a lot in Rocket League, which does look good for Ryzen. However, in Civilization 6, framerate is not how higher CPU performance presents itself, it's turn time, which by most accounts the 8700K is ahead in.

Starting with Civ VI, we used the AI benchmark to test the time required to compute AI turns, as FPS is useless here. The turn time is about the same at 1440p as it is at 1080p, though we did test both. AVG FPS actually goes up for worse CPUs, because the time spent sitting idle on the screen is longer, as it takes longer for the game to calculate a turn. This makes FPS an unusable metric for this particular AI benchmark.

https://www.gamersnexus.net/hwrevie...vs-ryzen-streaming-gaming-overclocking/page-5

I'd consider both the GTA 5 and the Rise of the Tomb Raider results as questionable because in both cases slower Coffee Lake CPUs beat the 8700K, in RotR's case by over 40%.

As far as most of the go-to CPU-heavy synthetics (Cinebench, Geekbench, etc.), as well as in every game AT tested, the 8700K's performance has not changed compared to launch. The Spectre/Meltdown patches have not affected gaming performance nor performance in CPU synthentics. The Stilt also has no performance degradation in Cinebench on his fully patched 8700K.

That's not to take any credit away from the 2700X; it's performance, especially with tuned RAM is really good (from the computerbase review), and it's a very solid refresh, way better than Haswell Refresh or Kaby Lake, regardless of what Linus says. It's more like Sandy->Ivy or better, and yeah, it's definitely setting up shop in the 8700K's territory. I just don't think that the narrative about the 8700K losing a bunch of performance from the Spectre/Meltdown mitigations holds much water, at least for consumer workloads (gaming, synthetics, emulation, video encoding, etc.).

 

[IEC]

They have 99th percentile results as well which give a clearer picture of what CPUs result in the smoothest gameplay at 1080p.

https://www.anandtech.com/bench/CPU/1852
https://www.anandtech.com/bench/CPU/1828

The 2700X is at the top or near it in pretty much every result.

 

[ondma]

They have 99th percentile results as well which give a clearer picture of what CPUs result in the smoothest gameplay at 1080p.

https://www.anandtech.com/bench/CPU/1852
https://www.anandtech.com/bench/CPU/1828

The 2700X is at the top or near it in pretty much every result.

So you are using a reference from anand tech to try to clarify the seeming anomalous results from the same site? Seems like circular reasoning to me. However, TBH, I put no faith at all in AT's gaming results anymore, even before this test, which seems to differ wildly from what other sites are getting.

 

[.vodka]

https://www.computerbase.de/2018-04..._benchmarks_mit_ddr43466_und_scharfen_timings

Google Translate'd

Finally!

Someone decided to do testing with proper timings. Man, does Pinnacle Ridge also benefit from Stilt's timings!

60ns for 3466MHz memory + proper timings, as predicted and tested a few weeks ago, and quite the performance increase in games.

Lovely sight.

 

[Nate Ortiz]

https://www.computerbase.de/2018-04..._benchmarks_mit_ddr43466_und_scharfen_timings

Google Translate'd

Finally!

Someone decided to do testing with proper timings. Man, does Pinnacle Ridge also benefit from Stilt's timings!

60ns for 3466MHz memory + proper timings, as predicted and tested a few weeks ago, and quite the performance increase in games.

Lovely sight.

Are those overclocked or stock frequencies with tuned memory? Because if you overclock the 2700x you get the same or worse performance because you lose the 4.35 single boost and all core is already like 4.1. But if you overclock the 8700k you can get like a 20% frequency boost from 4.3 all core to 5.1 which should have it pull way further ahead.