Podcast #441 - GTX 1080 Ti, FCAT VR, Ryzen, Kaby Lake De-lidding

Subject: Editorial | March 16, 2017 - 12:38 PM |
Tagged: podcast, ryzen 5, ryzen, nvidia, mobileye, jetson, gtx 1080 ti, fcat vr, delidding

PC Perspective Podcast #441 - 03/16/17

Join us for NVIDIA GTX 1080 Ti, AMD Ryzen Scheduler Discussion, AMD Ryzen 5 Announcement, Intel Kaby Lake de-lidding, and more!

You can subscribe to us through iTunes and you can still access it directly through the RSS page HERE.

The URL for the podcast is: http://pcper.com/podcast - Share with your friends!

Hosts: Ryan Shrout, Josh Walrath, Jeremy Hellstrom, Morry Teitelman, Ken Addison

Program length: 1:24:48



March 16, 2017 | 01:45 PM - Posted by Xukanik

So since AMD Rysen uses CCX 2 die system.
8C/16T = 2x4c/8t
6C/12T = 2x3c/6t
4C/8T = 2x2c/4t

Those can slow down performance on multi core applications due to latency.

How will this effect APUs'?
If they can replace one of the dies with a video die, this could possibly make the APU the best performing CPU for games discrete or not.

March 16, 2017 | 04:28 PM - Posted by Tim Verry

I was under the impression raven ridge will have 1 ccx but i guess we will just have to wait and see.I hope that the fabric is enough for the gpu.

March 17, 2017 | 12:30 PM - Posted by Anonymous (not verified)

Well for any APUs with a single CCX unit 4 core die the only usage for the off CCX unit Infinity fabric will be for maybe the CPU to GPU communication on the die(If using a monolithic die APU). So AMD's Infinity fabric is also going to be used on Vega and there is always the possibility of any APU's CPU to GPU cache coherency over the Infinity Fabric to maybe be used to keep any slower DIMM DRAM memory transfer issues to a minimum. It would be nice for the APU’s CPU/CCX unit to maybe be able to transfer kernels directly transfer to the Vega(Integrated) GPU's L2/HBC cache more directly sans the trip through slower main memory even if only used for some high priority CPU to GPU kernel workloads.

It's really good to go over to the Semiaccurate website and read Charlie D’s take on what the AMD Infinity Fabric may be able to achieve for APUs and even for CPU's to direct attached discrete GPUs(Directly attached over the Infinity Fabric on Zen/Naples). The Infinity Fabric will be used on AMD’s HPC/Server/Workstation(Hopefully) SKUs as more of an NVLink type of higher bandwidth coherent connection fabric in a direct attached CPU to discrete manner for the GPU accelerator to CPU server/HPC, and hopefully workstation market.

March 16, 2017 | 02:35 PM - Posted by Goofus Maximus (not verified)

I do wonder if the Ryzen locking problem could be related to the performance slowdown problem, and the "scheduling" problem is actually a power management problem...

March 16, 2017 | 02:49 PM - Posted by Anonymous (not verified)

You will take windows 10 FORCED up the A$$ and you will smile while taking it (real good) from the Lords of Redmond! (1)

Can anybody say Linux/Vulkan and NO to any FORCED M$ Serfdom on Ryzen and Kaby Lake based hardware forever!


"Microsoft Locks System Updates for Windows 7, 8.1 on Ryzen, Kaby Lake Systems"


March 16, 2017 | 11:30 PM - Posted by StephanS

anyone tried to overlock a r7 using only 4 core ?

A big chunk of the TDP is from the cores, so I would imagine with 50% less core active, the TDP must drop dramatically ..

and we are told that ryzen clock is only limited by voltage (and in turn heat)

So since the 1800x does 3.6ghz with all 8 core loaded at 95w ...
Could you get a 4 core at 4.2ghz at 95w ?

And push it to 4.8ghz at 120w ?

March 17, 2017 | 07:35 AM - Posted by Anonymous (not verified)

Relationships between voltage, power, and frequency are usually not linear; it is likely that 4.8ghz would require such a high voltage that the chip would not last long, even if it could be cooled (and it would be a lot more than 120w).

Still, it would be interesting to see overclocking on something like a chilled water setup. Unlike Intel chips, which hit a wall somewhere in the mid to high 4ghz range, Bulldozer and derivatives continued to scale well with voltage, as long as you could keep them cool - I remember reading about some 24/7 stable 6ghz setups on phase change, and it would be nice if Zen would also scale like that.

March 17, 2017 | 12:03 PM - Posted by Anonymous (not verified)

Smaller 14nm process node on newer CPUs so less atoms available to resist any over voltage/heat induced atom/elemental migration issues. Those larger process nodes for Bulldozer had more atoms available to fend off any atom/elemental migration issues caused by high heat, voltage, and electrical current related transistor degradation issues!

Intel's 14nm FenFet Process is more tweaked and the Fin Heights made taller to increase transistor area ratios and reduce leakage so Intel can run at higher clocks on its 14nm process relative to the Samsung 14nm process that GF licenses!

The smaller the process node the less resistance to any atom/elemental migration issues there will be as those smaller circuits have less atoms to loose to any atom migration, other atom/elements, migration that happens at higher current/voltage and higher temperatures, especially on overclocked processors!

Also at 14nm the circuits are going to be packed closer together causing more thermal related dissipation issues, and add to that the types of automated layout libraries used to get tighter circuit packing on any CPU layouts can also affect any thermal dissipation issues.

GPUs are engineered with high density libraries to get more shader cores per unit area and GPUs can not be clocked higher than CPUs that are laid out using nominal low density CPU high performance design libraries. Wait for the 7nm processes and there will probably be not much room for any CPU overclocking going forward as there will be too few atoms to lose before the circuits/transistors become fried!

March 19, 2017 | 04:25 PM - Posted by Anonymous (not verified)

Over at Anandtech Ryzen: Strictly technical, there is a very interesting comparsion between an Opteron 6140 (Magny-cours) SKU's behavior and Ryzen's! The full post is rather long but the benchmarking results are interesting!

Poster oegat(Post #1009) states:

"I joined just to share a test result of mine, which might interest some in this discussion. I ran tests on a virtual Windows 8.1 instance with 8 virtual processors, all pinned 1-1 to the 8 cores on a single Opteron 6140 (Magny-cours) die. The reason for the test, and why I bring it up in this thread, is that Magny-cours has a couple of key similarities with Ryzen when it comes to CPU topology:

•Each Magny-cours die consists of two ”modules” which share 4-6 (4 on the 6140) cores and an L3 cache
•L1+2 are private per core, L3 is shared per module and is a victim-cache of L2" (1)


Post by oegat, Today(Mar 19, 2017) at 9:16 AM


March 19, 2017 | 09:39 PM - Posted by Anonymous (not verified)

Vulkan it is for Star Citizen's crowdfunded project!

"Star Citizen Plans To Go Vulkan-Only"