AMD - Zen chitchat

[watercooled]

I'm honestly not just being obtuse, but in what area is it you can notice/measure a performance difference between Realtek and Intel? It's been a long time since I've ran into any real issues with Realtek NICs, the last time was with a weird intermittent 10/100 integrated Realtek NIC which randomly acted like the cable was disconnected.

I've used more than I can count from a number of brands (another being VIA) and there's been nothing in it WRT performance, and that's from a lot of playing with iperf amongst other things.

If you had a performance issue significant enough to notice, it could have just been a one-off. I've had faulty NICs from Intel too. It happens.

[CAT-THE-FIFTH]

Has anyone noticed this with Prey:

https://forums.overclockers.co.uk/posts/30774700/

DF recently released their performance review of the game and many noticed the GTX1060 seems to be rendering less shadows!!

Edit!!

http://image.prntscr.com/image/5316a...567299cf5e.png
https://i.redd.it/6232uzq8giwy.png
https://j.gifs.com/pg5Gor.gif
https://j.gifs.com/xG4Q9r.gif
https://j.gifs.com/3lD7EQ.gif
http://imgur.com/6w4MBj5

Second Edit!!

It seems PCGH noticed the game can dynamically change settings if less VRAM is detected,etc,so it could be down to the game itself.

[Biscuit]

I'm honestly not just being obtuse, but in what area is it you can notice/measure a performance difference between Realtek and Intel? It's been a long time since I've ran into any real issues with Realtek NICs, the last time was with a weird intermittent 10/100 integrated Realtek NIC which randomly acted like the cable was disconnected.

I've used more than I can count from a number of brands (another being VIA) and there's been nothing in it WRT performance, and that's from a lot of playing with iperf amongst other things.

If you had a performance issue significant enough to notice, it could have just been a one-off. I've had faulty NICs from Intel too. It happens.

The easiest to measure has been inconsistent and lower transfer speeds for large files.

Im not very good when it comes to the community though, as I tend to test and experiment but then not properly document or share my results. I still have a huge stack of photos and screenshot images from my server build in a folder waiting to be put together into a build log. I've left it so long that Ryzen is now released and my build is a pretty poor example to put into the world for any prospective buyer...

Could put it down to laziness but I also have to do testing and documentation for work...

[CAT-THE-FIFTH]

The easiest to measure has been inconsistent and lower transfer speeds for large files.

Im not very good when it comes to the community though, as I tend to test and experiment but then not properly document or share my results. I still have a huge stack of photos and screenshot images from my server build in a folder waiting to be put together into a build log. I've left it so long that Ryzen is now released and my build is a pretty poor example to put into the world for any prospective buyer...

Could put it down to laziness but I also have to do testing and documentation for work...

I would still put it up - its still a useful piece of information to have on the forums. Its like when I got the GTX1080 - I could have put one or two benchmarks out and left it at that since it had been out for quite a while already,but I decided to do a more detailed one anyway.

[watercooled]

The easiest to measure has been inconsistent and lower transfer speeds for large files.

Im not very good when it comes to the community though, as I tend to test and experiment but then not properly document or share my results. I still have a huge stack of photos and screenshot images from my server build in a folder waiting to be put together into a build log. I've left it so long that Ryzen is now released and my build is a pretty poor example to put into the world for any prospective buyer...

Could put it down to laziness but I also have to do testing and documentation for work...

Interesting. Like I say it's something I've not noticed on anything remotely recent - I have a mixture of Realtek, Broadcom and Intel NICs on my LAN and speeds are identical from my NAS to/from any of them (and between them). The only recent occasions I can think of where speeds have been sub-gigabit have been due to CPU load (stuff like first gen Intel Atom running Samba struggles a bit) or PCI bottlenecks e.g. for full duplex transfers.

Could it have been CPU performance or buggy drivers causing issues? E.g. did you try different NICs on the same system to narrow it down?

[Biscuit]

Admittedly no, I don't actually have any add-in NICs to do such a test but it's highly unlikely to be a cpu issue given that it's IVB quad.
This was back in Windows 7 whereas both machines are now on 10. Suppose I could always test it again without too much hassle to see if it was just driver issues at the time.
I would be tempted to get an add in NIC, but given I'm on mATX and still using my Xonar sound card, I don't really have the space.

[kompukare]

So more X399 rumours this time from Videocardz:
https://videocardz.com/newz/amd-engi...samples-update

16C/32T at 3.1GHz/3.6GHz sounds a lot faster than previously expected.
Videocardz list the CPU string decoding table but nobody knows what A9 means in terms of TPU. Something like the Intel Xeon E5-2697Av4 is possibly the closest Intel currently have. 145W with a Turbo of 3.6GHz. That's of a server chip and multisocket whereas X399 is meant to be Workstation.
At 3.1/3.6 with twice the cores, it really seems that logically the TPU should be close to twice of somewhere between R7-1700 (3.0/3.7 @ 65W) and R7-1700X (3.4/3.8 @ 95W). So probably above Intel's 145W.

[scaryjim]

... At 3.1/3.6 with twice the cores, it really seems that logically the TPU should be close to twice of somewhere between R7-1700 (3.0/3.7 @ 65W) and R7-1700X (3.4/3.8 @ 95W). So probably above Intel's 145W.

The 3.6GHz 1800X is also specced with a 95W TDP, so you're actually looking at being much closer to the 65W TDP clocks than the 95W TDP clocks. Plus 3GHz is roughly where the first voltage inflection is so you start getting diminishing thermal returns beyond that because you have to increase the voltage disproportionately.

I suspect the good silicon can comfortably do 3.1GHz @ 65W for an 8 core die, and 16 cores @ 3.1GHz is pretty much spot on what I was expecting from a 130W 2-die MCM.

Looks like this is gathering more than just HEDT ES details though - for example, AMD Gardenia is the platform for merlin falcon (embedded APUs) and AMD Mandolin appears to be the test platform for Raven Ridge... so at a guess we're looking at at least two variants on APUs there? Got to be honest, the APU samples are more interesting to me than the HEDT ones!

[CAT-THE-FIFTH]

Looks like this is gathering more than just HEDT ES details though - for example, AMD Gardenia is the platform for merlin falcon (embedded APUs) and AMD Mandolin appears to be the test platform for Raven Ridge... so at a guess we're looking at at least two variants on APUs there? Got to be honest, the APU samples are more interesting to me than the HEDT ones!

One ULV chip for thin and light laptops,and another larger chip for higher performance laptops and desktops??

Apparently Apple was considering Llano for a laptop,so it would be interesting to see if AMD Gardenia might end up in a new MacBook Air at some point.

[kalniel]

Has anyone tried Forza Horizon 3 since the new patch yesterday? Supposedly increases threading and better deals with affinity - could be a big boost on Ryzen CPUs methinks..

https://www.forzamotorsport.net/en-u...eels_available

[scaryjim]

One ULV chip for thin and light laptops,and another larger chip for higher performance laptops and desktops??

Be interesting if they do. Can't remember which thread it was but I was speculating about this recently - whether they've got the resources to produce a smaller 2-core + less shaders APU or whether they'll try to serve the low-end market with disabled 4-core parts. 2 core SMT, no L3 cache, and 4 - 6 CUs (256-384 shaders) should make a tiny but still very capable chip for laptops...

EDIT to add:

Just thinking about it, I'm currently running an A10 4600m laptop as my main computer at home. That's 4 Trinity cores @ 2.3GHz - 3.2GHz and 384 VLIW4 shaders at 685MHz. I reckon an SMT-enabled Zen dual core with 256 Polaris shaders running at 800MHz+ would probably give similar, if not better, performance, and this laptop already has plenty of performance for the vast majority of people.

[CAT-THE-FIFTH]

Be interesting if they do. Can't remember which thread it was but I was speculating about this recently - whether they've got the resources to produce a smaller 2-core + less shaders APU or whether they'll try to serve the low-end market with disabled 4-core parts. 2 core SMT, no L3 cache, and 4 - 6 CUs (256-384 shaders) should make a tiny but still very capable chip for laptops...

EDIT to add:

Just thinking about it, I'm currently running an A10 4600m laptop as my main computer at home. That's 4 Trinity cores @ 2.3GHz - 3.2GHz and 384 VLIW4 shaders at 685MHz. I reckon an SMT-enabled Zen dual core with 256 Polaris shaders running at 800MHz+ would probably give similar, if not better, performance, and this laptop already has plenty of performance for the vast majority of people.

That is why I mentioned Apple,if they wanted a custom 2C/4T part with a decent IGP,they have more than enough money to do so - IIRC,it looks like both Tonga and Polaris 11 might have been developed with Apple in mind.

[kalniel]

Has anyone tried Forza Horizon 3 since the new patch yesterday? Supposedly increases threading and better deals with affinity - could be a big boost on Ryzen CPUs methinks..

https://www.forzamotorsport.net/en-u...eels_available

Ryzen not needed. It just locked out my 950 at 60fps on ultra with vsync. Bodes well for future dx12 titles from MS though.

[kompukare]

More Videocardz leaks:
https://videocardz.com/69428/amd-sno...eat-horned-owl
Tons of slides on Naples, Snowy Owl (X399 apparently), and APUs. And a roadmap to 2018/19 with 7nm parts (server 'Starship' with 48C/96T).
Date Feb 2016. Guess for enterprise partners but still so old and no leaks? Even if it was accurate back in Feb'16, that's so long ago that the schedule is unlikely to be accurate.

[scaryjim]

... Snowy Owl (X399 apparently) ...

Welp, that's just pooped on the enthusiast market*. Snowy Owl is BGA only, no sockets. Otherwise it's exactly what we predicted - a 2 die MCM with quad channel memory and up to 64 PCIe lanes. But without a socket I can't see an enthusiast version happening. Also both it and Naples are designed to run without chipset, which would mean the rumoured X399 "chipset" isn't - it's just a "more PCIe lanes" version of X300 for AM4. I honestly don't see any way they'll release a socketed version of Snowy Owl for enthusiasts if they're not doing a socketed version for servers...

It does confirm that the mid-range APUs will be 4 core with 11CUs (704 shaders) and dual channel DDR4, and will apparently officially support memory speeds up to 3200MT/s. And there's a small die embedded APU with 2C/4T and 3 CUs (192 shaders). The larger APU will come in BGA and AM4, with the smaller one BGA only.

Of course all that's officially only the server and embedded line up, but when was the last time that AMD had significantly different silicon between their consumer and enterprise lines...?


*EDIT: unless, of course, AMD's plans in that regard have changed in the last year....

[Xlucine]

What's alienware doing with a very conservatively clocked 12 core part?