FEATURED AMD ZEN Discussion (Previous Rumor Thread)

[mackerel]

it's dual channel platform, can compare it to i7 6700k

Will try it when I have a free system...

Having said that, my earlier ram bandwidth limited throughput testing with Prime95 did show comparable performance at 2133 dual channel single rank, with future research required to better understand if it behaves differently with 2R per channel.

 

[Kenrou]

Just wanted to add this here for those saying that Ryzen is "terrible" for gaming

Only thing holding back seems to be clock speed no ? here's to hoping the 4c/8t and 6c/12t reach higher...

 

[notarat]

My 4770K is getting a bit long in the tooth and I'm about 98% certain that I'm going to upgrade to a Ryzen 1800X, but I cancelled my pre-order just before the release because I heard rumblings about memory (and I couldn't find the board I wanted)

I'll hold off a month or so on my upgrade until memory issues are ironed out and boards are a bit more "available".

 

[Tgrable]

Only thing holding back seems to be clock speed no ? here's to hoping the 4c/8t and 6c/12t reach higher...

Going off the amd documentation the 1800x is the highest stock clocked chip they have of the 7 and 5 series

 

[Kenrou]

Ryzen - Amazing Update for Gamers (?)

 

[Tech Tweaker]

I think some of the gaming issues may be related to the high memory latency (90-100+ nanoseconds). I would have thought they could get it between 50-70ns, even AMD socket AM3 Phenom II's managed to be in that range if I remember correctly.

I noticed on my LGA1150 system when I was doing some memory testing a while back if I ran memory at 2666/2667MHz DDR3 I would get 150ns of latency. Even though speed was much higher than my 1866MHz kit, the latency on my 1866MHz kit is 54-60ns. The extra latency was noticeable as while gaming actions on screen seemed to happen more slowly and games felt less "smooth", even though my FPS was higher. Also, programs seemed to load much more slowly in general (not just games) Then, I dropped back to 2400MHz and noticed no drop in FPS, but latency went down by a lot to 50ns (which is much more manageable).

As for the rest, these kinds of things (BIOS issues, memory issues, board issues, etc) are why I don't like to buy into something early (or be an early adopter of new ideas or technologies). My phone I bought a year after it came out, for my computer platform I chose one generation behind the newest (for multiple reasons, bent CPU PCB's among them). I figure it will be at least three to six months before they get most of it straightened around, perhaps longer.

 

[Kenrou]

 

[Johan45]

Here's one for comparison this is the 1700 at stock appears to register boost speed and 2933 ram

 

[Mandrake4565]

My 4770K is getting a bit long in the tooth

Long in the tooth doing what?

 

[Alaric]

Looks like Ryzen may have a little more time for optimization. Cannonlake doesn't seem too healthy at the moment. https://www.fool.com/investing/2017/03/06/intel-corporation-expresses-doubt-about-10-nano-ra.aspx

 

[Dolk]

I can do some write ups on Intel slow down in the material side if some people are interested. It will definitely give some time for AMD to respond with a Zen+ on the same feature size.

 

[MaddMutt]

I just got my $600.00 Ryzen 7 1800x......Still waiting on my $269.99 CH6 that I ordered from Amazon I be able to start test then

Which would be the best OS to bench with for boints???

 

[DaveB]

Here's a comparison of memory latency my AIDA64 test of my 1700 @ 3.9 GHz with the memory at DDR4-2400.

 

[wingman99]

I really don't see how people can say it's so terrible even with limited bandwidth it keeps up clock for clock with it's intel counterparts and in some areas actually exceeds their performance. Memory performance is supposed to be improved as well with upcoming code releases. As for gaming if you compare apples to apples they're the same, if you have titles that take advantage of the extra cores then it really shines. It has the IPC just doesn't have the speed of a skylake i5 which is where they're taking advantage. CPU performance isn't all about gaming.
There definitely are areas for improvement and I'm sure it will. Was it released a bit early probably yes. There will be some growing pains and I have had my own.
My biggest mistake was using an older Win10.ISO which caused all sorts of instability issues. I have since used the newest at MS and it has made a world of difference almost night and day.

I don't see Ryzen shining in these gaming benchmarks Link: http://www.gamersnexus.net/hwreview...review-premiere-blender-fps-benchmarks/page-7

You keep saying "die" which is confusing. A two die chip would be the core2quad. Ryzen is a single continuous piece of silicone, and a single die. Inside that die many different interconnected integrated circuits. The two largest blocks of these integrated circuits are labeled as a single ccx which contains 4 cores, l1-l3 cache, etc. These are essentially duplicated hence having 8 cores. The interconnect is called the "infinity fabric" and operates at the actual clock speed of the RAM. E.G. when ram is running at DDR4-2667, the fabric runs at 1333 MHz. It may be better to think of the CCXs as 'modules' from FX, but the new versions. The Infinity Fabric should be thought of as the new version of the CPU-NB from FX. There are some differences, etc, but the CCXs are not connected by a bandwidth of single channel ram. Just because the fabric has a clock speed of the actual ram clock speed, doesn't mean it's overall data throughput is the same. It is possible to have something very wide (interface wise) with multiple IO points (think memory interfaces on a GPU, I.E.- 256-bit interface vs 128-bit interface, if both are operating at the same clock speed, then one has double the throughput).

My understanding is they are connected via shared L3 cache, and part of the problem is how Windows 10 handles looking at each CPU core and it's dedicated resources. Slilt does go into it a bit with someone concerned about Vega which is supposed to use a similar infinity fabric, particularly regarding bandwidth access to the memory. Maybe that can shed some light into the issue?

I'd wager that unless Dolk has dug that deep on Ryzen, one of the better places to get info would be within that thread itself. I'm merely an undergrad EE student who is focused more on RF applications and theory, so am a bit outta my element in these realms. If you get your questions answered, and understand them, please let us know here as well. Slilt seems to translate things into laymen terms pretty well, and my understanding of the architecture design could WAY wrong, but it's just how I processed it

This picture shows two dies for Ryzen

 

[Clutch_Head]

This picture shows two dies for Ryzen

two dies, or two solder pads?

 

[Johan45]

@wingman99 I posted a video at 2620 that shows it keeps up just fine.

 

[Alaric]

I can do some write ups on Intel slow down in the material side if some people are interested. It will definitely give some time for AMD to respond with a Zen+ on the same feature size.

That was my thinking. A little breathing room for a brand new architecture can't hurt while they fine tune the rest of they system. And I'd certainly be interested in a more in depth look at the 10 nm situation. Knowledge=good.

 

[wingman99]

Just wanted to add this here for those saying that Ryzen is "terrible" for gaming

Thanks for finding that video, my i5 6600k beats Ryzen 1800X according to the 16 game average.

 

[Bluefalcon13]

This picture shows two dies for Ryzen

You both are correct. I was looking at the CCX die shots incorrectly. It still stands that any core can access the whole L3 cache at any given time. The 'infinity fabric' interconnects the separate dies and has significantly higher throughput than RAM. This has a fairly deep dive into it within the review:

http://www.anandtech.com/show/11170...-review-a-deep-dive-on-1800x-1700x-and-1700/6

I can see how the Stilt's cache issues could cause all kinds of problems. Thanks wingman and sentential for pointing me to better my understanding!

 

[Dolk]

Theory for those that want to test memory latency (NOT CACHE):

If you have a 1700 or up to 1800x, please enable/disable a set of 4Cores (SMT OFF) and test memory latency. Repeat test for other set of 4 cores.

The set of 4 cores should be Cores 0 through 3, and Cores 4 through 7. I want to verify something that I've noticed in the Ryzen die shot.