AMD - Zen chitchat

[[DW]Cougho]

These overclocking numbers are reminding me of the Fury X....no headroom at all. Essentially AMD are selling pre-overclocked parts.....makes me wonder how stable they will be in extreme conditions.

The Samsung process isn't built for high clocks from what I hear, Intel will have the lead there for a while yet.

I also wouldn't call a 900mhz overclock shabby, remember this CPU (1700) is only meant to do 3.7GHz on 1 core, it's 3.0GHz base. There's a little more to come from it I'm sure, it reached 3.9GHz so easily and although I've been overclocking for many years I'm not very good (I don't have the patience).

[scaryjim]

... Essentially AMD are selling pre-overclocked parts.....

Sorry, but that's nonsense. The only reasonable definition of "overclocked" would be working outside of their frequency range design. The fact that we're talking about a 3.6GHz base clock 8-core part at 95W strongly suggests to me that they're working within their frequency range design. They simply happen to be made on a process that incurs significant penalties for operating outside of the design range. But the fact that they don't overclock much doesn't mean that they're "pre-overclocked".

We already know that if you give them LN2 and a ton of voltage they'll do well over 5GHz. That basically puts them on a par with Intel's 6000-series HEDT parts, which have similar stock clocks (3.5GHz - 4GHz) and overclock to about the same level (a bit over 5GHz). Haven't heard anyone claiming the 6900k is a pre-overclocked part though....

[shaithis]

Sorry, but that's nonsense. The only reasonable definition of "overclocked" would be working outside of their frequency range design.

I'd call that nonsense as well. You can argue over the meaning over overclocked all you like but good luck finding a general consensus

No overclock headroom (without using LN2) = preoverclocked IMO. Regardless of what happens under extreme cooling results. That's running at the extreme edge of the silicon and that's when things start to go wrong....hello summer around the corner!

The 6900ks seem to hit 4.5GHz+ and while not a massive overlock is certainly a lot better than barely surpassing the turbo multiplier!

[scaryjim]

... You can argue over the meaning over overclocked all you like but good luck finding a general consensus ...

At which point talking about something being "pre-overclocked" is pretty much meaningless.

However, if you're using "no headroom" as the definition, then I don't think that applies here: the 1800X seems to hit 4GHz all-core without too much issue, which is an 11% overclock on stock. It certainly doesn't have much headroom (and thanks to its intelligent boost tech it can actually acheive those clocks on single threaded loads without manual OC), but an extra 10% isn't exactly "no" headroom - it's in line with what a lot of Intel's top SKUs have done over the last few generations...

AMD have sold "pre-overclocked" parts before; the FX 9000 series unquestionably, and arguably the Fury X (I'm on the fence about that one); you might even persuade me that the entire Polaris line was pre-overclocked. The Ryzen 7 X chips simply don't fall into that category, IMNSHO...

[kalniel]

If the retail processors are 'pre-overclocked', what do you call these then?

https://www.scan.co.uk/shop/computer...-ryzen-bundles

[scaryjim]

... what do you call these then? ...

post-overclocked

[watercooled]

I think I'll have to return the board as it's broken for all intents and purposes - that PC is not usable any more. Any recommendations on what to swap it with? I'd prefer a microATX board in case I want to repurpose it as an APU system in the future, but that's not essential.

WRT overclocking, Wikipedia's first line is

Overclocking is configuration of computer hardware components to operate faster than certified by the original manufacturer

with which I agree. Regardless of what clock a CPU ships at, it's guaranteed to work at those clocks. Overclocking is any aftermarket adjustment to those clocks. You can't make the argument of CPUs being past the point of diminishing returns in terms of perf/W as that generally comes at very low clocks, way below what desktop CPUs ship at. The point some CPUs made on a certain node means they don't have much headroom to clock higher, doesn't mean they're 'pre-overclocked' - it just is what it is. If you look at the voltage curve of various nodes this becomes pretty clear - some have a very long tail with decreasing efficiency as you go, others hit a voltage wall earlier on. That's the sort of thing that distinguishes the several different versions of even TSMC's 28nm node.

[watercooled]

Rather worryingly for Gigabyte, this is pretty much exactly what happened to my board: https://forums.overclockers.co.uk/th...r-up.18772620/

It's the same board too. I don't have an OCUK forum account, but I wonder if someone (CAT?) would like to point out the similarity to that forum member?

[kompukare]

I don't have an OCUK forum account, but I wonder if someone (CAT?) would like to point out the similarity to that forum member?

Done!
https://forums.overclockers.co.uk/posts/30618271
Always good to know you're not alone!

[watercooled]

Thanks! For the benefit of anyone just reading this thread: http://forums.hexus.net/cpus/371038-...ml#post3785101

And as an update I've given up on the board and am returning it. The state of the BIOS is unacceptable for a shipping product and this was the final straw.

As an update, the BIOS recovery thing is only a temporary fix - it will reboot and work OK-ish for a bit, but it seems as soon as you shut down then power up again, it's broken again. I've even tried both the old BIOS and the newest one which has appeared in the downloads section - no difference with any of them.

The mouse had completely stopped working in any BIOS version, though curiously it did work in a rear USB 3.0 port. All in all a rather bizarre set of problems that has left me with zero confidence in that motherboard and has seriously dented my opinion of Gigabyte.

[kompukare]

CPC Hardware have been tweeting again. First about the mythical workstation 16C/32T part (both Google Translated):

There are many Ryzen 16C / 32T focused HEDTs planned on X399 in 4/6 months. Clocks ~ 2.4 / 2.8G. 2 dies MCM. 4 chan DDR4. Socket LGA SP3r2. ~ 150W.

https://twitter.com/CPCHardware/stat...99198799187968
And then this including a Naples board photo:

As a reminder, the current samples of Naples 32c / 64t (64 MB L3) on SP3 in stepping B0 run at 1.9 / 2.5 / 3.0 GHz for a TDP of 180W.

https://twitter.com/CPCHardware/stat...05726298054656
I guess the second tweet is contrasting HEDT @ 150W vs Naples @180W with twice the cores and the speeds don't look that much better - although no idea what the three numbers are meant to mean? All cores, some turbo, and single core turbo?

BTW, is SFP+ on the actual server port usual? Well, that's what those long connectors on the right look like, but surely even if 10G wouldn't it make more sense to just have Ethernet as those things look like they'd exert a huge pressure on the board. Never mind, a quick google image search shows up plenty of server motherboards like that.

[kompukare]

There's this Reddit thread about a Win10 update from the 16th of March which may have solved some of the scheduling issues:
https://www.reddit.com/r/Amd/comment...tially_solved/


Or maybe something else is going on as 3-4 days is a long time for no site to not have picked anything up.

[watercooled]

Techpowerup review out for Ryzen, and seems like quite a decent, thorough one.

https://www.techpowerup.com/reviews/..._1800X/16.html

Interestingly, they also had issues with a Gigabyte board...

[kompukare]

Yes, W1zzard was very impressed with the power consumption and he's not even benching against any Intel HEDT LGA2011 chips only the consumer ones.
Don't think what gets run for the gaming power load is mentioned (for graphic card, he's use Metro: Last Light for ages, but that's not in the Ryzen review).
Anyway, I can't help but feel that there's something wrong with this picture though:

That is, the power efficiency for gaming seems to be because the chip is under-utilised. Guess a i7-6950K would similarly have most of cores not at 100% load.
Still it's not bad for 1080P gaming at all:

On average it's only 12% slower than the i7-7700K (which is running 200MHz faster) which is a very good showing. If R5 1600 gets anywhere near to that, it will be very hard to justify paying way more for Intel's 4C/8T chip.

[CAT-THE-FIFTH]

Ryzen does have some real issues with certain games - it really does not gel well with Creation based games too.

Anyway,here are some tests with a simulated R5 1600X and R5 1500X:

https://www.youtube.com/watch?v=O5cqOtWz5sU



It looks like if you are gaming they will bought have most of the performance of the R7 1800X for much less money!!

[Biscuit]

It looks like if you are gaming they will bought have most of the performance of the R7 1800X for much less money!!

Which is exactly what we should expect given that we already know games are not heavily affected by 4/4 vs 4/8 vs 6/12 vs 8/16.


I quite like this guys analysis of the gaming situation. He also covers fallout 4.