• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

AMD Ryzen Gen 2 Set For Q2 2018

Page 24 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
krumme said:
Anyone wondered if the change of ceo at GF perhaps is connected to lacking 7nm execution?
Click to expand...
The opposite is actually more likely the case. The outgoing CEO Sanjay Jha is going to handle future business development elsewhere for GloFo's owner Mubadala while the new CEO is part of GloFo's internal staff. So the former was suited to smoothly execute the owner's intended plans while the latter is suited to smoothly proceed with everyday business at GloFo.
 
In the Ryzen anniversary semminar today,Jim Anderson said on the record that Ryzen 2000 would feature improved latencies.
Sadly I was not paying the best attention at that point, so I didn't get details but there probably wasn't anything more specific. But this could suggest those AIDA improvements could be at least part legit.
 
The Stilt said:
Most likely only at 4K resolutions.
At 1080P X265 cannot use more than 10-12 cores efficiently due to it larger default CTU size (64), despite the theoretical limit is 16.875 cores.
With X264 the default CTU is half the size, meaning it can use more cores efficiently.

Currently X265 has no AVX512 code in it, despite it is certainly a work in progress (assembler was replaced with NASM a while ago).
X264 contains a very modest amount of code in AVX512 and it provides < 5% performance improvement over AVX2. VP9 supports AVX512 as well, however the performance of the encoder appears to be so poor (for multiple reasons) that
I don't think it is relevant how much or little AVX512 improves the performance.



Personally I don't believe that Zen2 will increase the width of the execution resources.
I guess it will ultimately depend on the strategy: Is it possible (or viable) this time around to make separate server and desktop designs, and if it isn't does the server design need to have wider resources or not.
Click to expand...
My current (wild) theory for Zen 3 is a 4 way SMT with wider core design. Same core count as Zen 2 designs but each core wider. I'd imagine they add more execution units of the same width rather than make the current ones wider. It would algorithmically decide whether to run in SMT 2 or SMT 4 mode, to avoid the horrible cache thrashing effects that SMT 4 can create.

Of course as I said that's the wilder theory. The safer bet would be AVX256 with AVX512 support for Zen 2 and Zen 3. That would be the sweet spot width in my eyes.

krumme said:
Anyone wondered if the change of ceo at GF perhaps is connected to lacking 7nm execution?
Click to expand...
The new CEO is a great guy, that's all I can say. I believe 7nm is going smoothly, but Tom is the one I'd rather see leading the charge.
 
Last edited:
I hadn't thought about Zen growing to include 4 way SMT. If they did include that, I would hope that they make changes to the L1 and L2 caches to help with handling all of the data that would get shuffled around because of it. I would also hope that it's configurable like in the Power in that you can choose SMT2 or SMT 4, or turn it off completely. I would think that, in x86-64 world, 4 way SMT would usually not be as helpful as it is in IBM's use cases. However, if Zen 3 gets wider, it could allow greater utilization of the extra execution units.
 
https://browser.geekbench.com/v4/cpu/compare/7441750?baseline=7440381
Click to expand...

Strange that 2700X reports for geekbench shows this for L1/L2/L3 caches. I think it is not reading 2700X correctly.

oxLoft9.png

Entry from yesterday.
 
DisEnchantment said:
Damn... the spread on these geekbench scores ...😕
Click to expand...

That is pretty shocking. I don't think much of Geekbench in general, yet the 2700X absolutely murders the OCed 1800x despite a reported clockspeed 300 MHz lower.

edit: looking at it again, I noticed that they used what appears to be DDR4-2400 on both systems. I wonder if the point spread would have been different at DDR4-3200 or higher? Makes me wonder if IF speed ratios are different on Pinnacle Ridge, or something. Summit Ridge is definitely going to see some IF speed problems at such a slow memory speed.
 
Last edited:
DrMrLordX said:
That is pretty shocking. I don't think much of Geekbench in general, yet the 2700X absolutely murders the OCed 1800x despite a reported clockspeed 300 MHz lower
Click to expand...
It's definitely a very strong result, but bare in mind that this 1800X is suspiciously slow for that overclock.

The reference 1800X scores 4249/21981. Compared to that the 1700X is still nearly 15% faster in multithreaded code, which is very impressive.
Yet there are some 1800X results overclocked to 3.9 Ghz on a similar geekbench version, that get very similar results to the 2700X.

Though to be fair, most of the strong results seem to have much higher mem-clocks (the one i selected is pretty close though), and therefore much stronger results in BW bound tests. So 2700X with similar memory will probably do even better.
 
DrMrLordX said:
That is pretty shocking. I don't think much of Geekbench in general, yet the 2700X absolutely murders the OCed 1800x despite a reported clockspeed 300 MHz lower.

edit: looking at it again, I noticed that they used what appears to be DDR4-2400 on both systems. I wonder if the point spread would have been different at DDR4-3200 or higher? Makes me wonder if IF speed ratios are different on Pinnacle Ridge, or something. Summit Ridge is definitely going to see some IF speed problems at such a slow memory speed.
Click to expand...

Guess we'll have to wait and see when reviews are released. Looks like a decent uplift so far. Probably the more aggressive boost in play.
 
It's also interesting to compare the results to the (very strong) leaked 2600X score (4781/22235). The latter is really good for a 6-core @ stock. It's about the same as the baseline 8700 (non-K) and better than the baseline 1800X.

How does the 2600X perform so well? Faster Clocked memory. Just look at the BW results vs the forementioned 2700X (and how much the 2600X gains from that, also notice the latency 😉)

6g1u3S4.png


Here is another comparison between a 1700X @4.0 (and very fast, low latency memory) vs the 2700X. The 2700X is consistently better, only losing in memory BW and BW bound tests (AES, SQL etc). Yet, because it loses so badly on latency and BW, the scores are practically even.

TL;DR: the 2700X should get a noticeable boost from the leaks with faster memory

Also worth keeping in mind is that all the results so far are on the X370 chipset. The X470 will hopefully do even better.
 
Last edited:
DrMrLordX said:
If that were the case, we would've seen a larger performance delta in the ST score rather than MT. I'm inclined to agree with @Gideon that memory settings are more to blame here.
Click to expand...

Unless the multi core boost has more headroom due to higher tdp.

Reviews will most likely reveal the underlying reason.
 
Out of Curiosity I also ran Geekbench 4.2.2 on My 1700X @ Stock (3.5 All Core Turbo and 3.9 XFR Single Core) only changing the mem-speed

The scores were:
4091/19707 with DDR4 @ 2133 MHz
4527/21920 with DDR4 @ 3466 MHz

And that's @ stock, with different only in memory (and IF) speed.

When comparing my results with the leaked 2600X result I noticed a couple of interesting things.The 2600X seemed a better fit, than the 2700X, as the OS, Geekbench versions are the same and memory is also much closer at 3200 MHz:

  • In Single Threaded tests the 2600X is pretty consistently faster (5-12%)
  • In Multi Threaded tests the 2600X destroys my 1700X in SGEMM and PDF render, despite having less cores (seems to be an issue of mine, some 1XXX series chips seem to run better)
  • The memory latency is 5ns better, despite running slower RAM.

I really hope that these chips can run faster memory stably with the X470 chipset. With 3600 MHz with tight timings we could see latencies in the low 60ies or even high 50ies.
 
There are some rumors, that x470 will support up to DDR4-4000.

The X470 motherboards should support up to DDR4-4000 after overclocking, which is a solid improvement over the 300-Series motherboards. AMD's Ryzen processors benefit greatly from increased memory frequencies because the Infinity Fabric runs at the same speed as the system memory. That boosts performance in memory-sensitive applications, like gaming.

http://www.tomshardware.com/news/gigabyte-amd-zen-x470-motherboard,36309.html
Click to expand...
 
Thats nonsense, X370 already supports up to 4266MT/s with latest drivers. Can you run it at those freq, maybe SC with loads of voltage
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top