AMD's Ryzen CPUs (Ryzen/TR/Epyc) & Vega/Polaris/Navi GPUs

4GHz on all cores is overclocked. The 1950X has a 3.6 all core, 4.0 4-core, and 4.2 XFR (one or two cores) if memory serves. So, this is definitely overclocked. I posted the exact clock rates a couple pages back.

Sent from my SM-G900P using Tapatalk

Edit: post #2549 on page 255.

 

3,277 is much better. Still want too see what 3600 MHz yeilds though. Also nice too see 4.0 GHz stable but it would have been sweet to get that 4.2 GHz on all cores.

 

So far, 3.9-4.0 overclock have been shown. I'm thinking binned ones can get 4.1, but time will tell. Still, 4.0 just a little over 1.4V means it is a daily driver at those speeds.

Sent from my SM-G900P using Tapatalk

 

Surely at 4.0 GHz I'll be a happy camper. I know supposedly with Ryzen 7 above 3200 MHz has diminishing returns. This is why I am so curious with TR as is it worth the extra steps to get too 3600 MHz. That is the memory and board sections required.

I am figuring it really takes, at lesat for now, the 5 GHz intel cores to really utilize the 1080 TI the water cooled Vega may be the best option. You just can not configure one yet for that.

 

Also, I'm betting it does well with graphics cards in professional work...

Sent from my SM-G900P using Tapatalk

 

Level1 News August 1 2017: Reduce, Reuse, Reeeeeeeee!
25:58 - AMD Ryzen 3 1300X And 1200 Processor Review: More Affordable Zen
27:52 - AMD Ryzen Threadripper systems are now available
29:15 - MSI Video Outlines Threadripper Installation Procedure In X399's TR4 Socket

Spoiler: Full Index

Links to articles referenced in video:
https://www.one-tab.com/page/w1AJ86QOSPKGrELYY5reuw

0:34 - Lawsuit seeks Ajit Pai's net neutrality talks with Internet providers
2:05 - Apple Has Pulled Anti-Censorship Apps from China's App Store
3:22 - West Virginia Tries To Improve Broadband Competition, Incumbent ISPs Immediately Sue
6:04 - China Forces Muslim Minority to Install Spyware on Their Phones
7:39 - Feds Crack Trump Protesters' Phones to Charge Them With Felony Rioting
10:13 - India's transport minister vows to ban self-driving cars to save jobs
12:25 - Microsoft's default font is at the center of a government corruption case
13:10 - Sweden Accidentally Leaks Personal Details of Nearly All Citizens
15:17 - U.S. indicts suspected Russian 'mastermind' of $4 billion bitcoin laundering scheme
18:03 - Oh ****, the SEC Just Ruled That Ethereum ICO Tokens Are Securities
19:36 - Hacker Steals $8.4 Million Worth of Ethereum From Veritaseum Platform
21:23 - German Court Bans Mass Usage of Keyboard-Tracking Software by Employers
22:52 - US Army Seeks Internet-of-Battlefield-Things, Distributed Bot Swarms
24:59 - U.K. to Require Registration and Safety Tests for Drone Owners
25:58 - AMD Ryzen 3 1300X And 1200 Processor Review: More Affordable Zen
27:52 - AMD Ryzen Threadripper systems are now available
29:15 - MSI Video Outlines Threadripper Installation Procedure In X399's TR4 Socket
30:51 - Samsung ends Intel's 24-year reign, becomes the largest chipmaker in the world
32:19 - Intel Discontinues Joule, Galileo, And Edison Product Lines
35:25 - Microsoft Adds Support for Linking Android Phones to Windows 10 PCs
36:49 - Microsoft Paint is not being killed in next Windows 10 update
38:57 - P&G Cuts More Than $100 Million in "Largely Ineffective" Digital Ads
41:09 - The Tesla Model 3 interior doesn't look like any car you've ever seen
43:33 - Twitter's user growth disappoints despite Trump tweets
44:49 - 100x faster, 10x cheaper: 3D metal printing is about to go mainstream
46:30 - Facebook reportedly building smart speaker with touch screen
47:07 - Elon Musk Says Mark Zuckerberg's Understanding of AI Is Limited
47:51 - Hackers break into voting machines in 90 minutes
49:37 - Flush times for hackers in booming cyber security job market
50:40 - Hacker cracks smart gun security to shoot it without approval
51:36 - A Wisconsin company will let employees use microchip implants to buy snacks and open doors
53:52 - Jeff Bezos Briefly Tops Bill Gates as the World's Richest Person
54:38 - A recycling robot named Clarke could be the key to reducing waste

AMD Crossfire Losing Support!

I don't think it's AMD that's dropping support, they are just observing the rough start for DX12 multigpu and lack of developers putting multigpu support at launch. There are still a lot of games that support Crossfire and scale 2x well, IDK about 3x/4x, but would be fun to try on ThreadRipper MB's

Threadripper CPU Installation Guide! How Socket TR4 Works

 

@ajc9988

I knew I read that the Asus Zenith Extreme would support 3600+ RAM (supposedly all of their X399 boards will support this speed). https://edgeup.asus.com/2017/07/31/x399-motherboard-guide-threadripper-asus-rog-prime/

I do understand that Asus is advertising 3200 on their "main" website but that Edge Up article isn't the only place I saw that pointed to the 3600. I canceled my pre-order to wait on reviews.

 

So, here's the issue I have, the box marked 2800, the website and QVL lists 3200. The M8E I have supports 3866 officially only as 1 dimm. I got the ram to 4000 on this board, but that is lucky, as binning boards was needed to reach that. I settled on 3733 just because the tighter timings made up the difference. So, I'd wait for more confirmation and more QVLs before making the final call (which you said you were doing).

Sent from my SM-G900P using Tapatalk

 

If you can get 30 day + returns without charge, it's worth a gamble just to get in on the fun early.

 

It's back up...named differently, but same graphed results.

AMD Threadripper - Memory Speed & Latency on Cinebench (Standard & Performance ram)

 

Nvidia responds to AMD by removing Sandbagging in drivers to unlock 3x compute performance


Gulskjegg 8 minutes ago
"I buy AMD products exclusively not because AMD is awesome, but because Intel and nVidia are the absolute scum of the earth."

TITAN Xp Enables New Levels of Performance for Creative Professionals
https://blogs.nvidia.com/blog/2017/...ls-of-performance-for-creative-professionals/

Titan Xp Enables New Levels of Performance for Creative Professionals | NVIDIA Blog
https://www.reddit.com/r/hardware/comments/6qtp99/titan_xp_enables_new_levels_of_performance_for/

Titan X (Pascal) benches with the new 385.12 driver
https://www.reddit.com/r/nvidia/comments/6qtf9b/titan_x_pascal_benches_with_the_new_38512_driver/

In the reddit post, be aware the poster highlighted highest Titan Xp scores, but AMD still wins in a number of tests, look at the scores for comparison not the bold highlighted scores.

 

^^ highlights why we need competition to nvidia

 

It has long been known nVIDIA is gimping their products.. and Intel is barely innovating. Well, knock knock Ryzen smashed down your door, and on top of that Vega smashed compute performance to another dimension, I know gaming is absolutely everything, but if you create any content what so ever, this boost is everything. On top of that, FreeSync monitors are dimensions cheaper than G-Sync monitors and let us not forget that AMD giving fierce competition might get nVIDIA and Intel to pull their head out of their ar**.

110% going AMD my next build. This is too absurd..

 

Sent from my SM-G900P using Tapatalk

 

http://www.asrock.com/mb/AMD/X399 Taichi/index.us.asp#Memory
@TANWare @Rage Set @hmscott
So, Asrock has published the QVL list on memory today. My ram are confirmed for 3600 (verified overclock), but has not been tested in quad channel. The 3733 version of the Trident Z line has been tested and confirmed in quad @ 3600, but it is a quad channel kit.

 

I've been wondering about quad-channel in the ThreadRipper, and usually how the fewer the DIMM's/SODIMM's the better the chance of maintaining high speed low latency settings.

With ThreadRipper isn't the memory access "shared" across the Infinity Mesh, such that if you got 4 x DIMM's, it's really 2x to each CCX, so running 4 DIMM's is the "minimum" configuration for quad-channel and low enough count to run highest speed and lowest latency?

Or, is there an advantage to 1 DIMM per CCX, and run memory in a 2x configuration?

What's the best memory layout per the design spec / manual?

 

Great question. This goes back to the discussion of memory on Epyc. Intel has one massive, consolidated memory controller, but Ryzen has one per die. So, when the other die needs information from the other set of dimms connected to the other die, it has to jump to the other die, then jump to the memory, then follow it back. For 2P, they made a big deal on limiting it to 2 jumps before hitting the ram. Here, you just have the one jump.

This means you are achieving that speed per controller. So, two kits should work fine instead of one kit with all four tested as matching. The timings are still set together, though, meaning you still may have to find the variance in timings between the two kits.

Now, if you only populated dual channel to one die, you would permanently handicap the other die. If you run one in each channel for each die, you are not approaching as good of performance as if you did dual channel. So you do need 4 dimms to get the higher speeds and lowest latency.

 

http://www.gigabyte.us/Motherboard/X399-AORUS-Gaming-7-rev-10#support-doc
So, Giga and Asrock are the two MB that have official QVL certifications at 3600MHz. That means Asus and MSI need to step up to the plate. Half the boards support it, while the ones with 2x8-pin connectors do not.

 

I don't doubt either of them are far behind, though I'm surprised that ASUS wasn't the first.

 

Look at the OC Formula from Asrock--the first board to support 4133 G.Skill memory on the 6700K Z170 board. For Asus, you had to bin boards, only the Maximus VIII Impact ever officially supported it, and you still had to bin boards across the Asus line to find ones that could really support that memory speed. So Asrock being first wasn't a surprise at all. Asus has lost that part, but they have really good EUFI features. They also are the first E-ATX, support double 8-pin (more important for LN2), have the DIMM.2 slot, etc. So, it isn't a surprise they lacked focus elsewhere. With Intel chips, you don't see much, if any, scaling with memory speed. With AMD Ryzen based CPUs, because of IF scaling, it becomes more paramount. So, this isn't to knock them, but rather to say don't overly bet on them to get the best memory speeds. Asrock showed they can take that crown, but then you may have little things that upset you with its UEFI, etc. I have a feeling all boards will have their strengths and weaknesses, but I also feel that 3600 will be the base best, with binning to get above that level.

Take for example Asus, Gigabyte, and MSI. They all tested the ram kits on whether they could run at the rated speeds. Every kit listed runs at the rated XMP. Asrock took a different approach. They showed the max speed they achieved with each kit tested. So, my 4133 Trident Z are verified to run at 3600, NOT their rated speeds. That means they played and you will have to tune the ram to get it working. It also means that the tuning algorithm used for the secondary and tertiary timings is able to allow use of a LARGER set of kits, whereas sometimes Asus sets them too tight and you get issues (which is why Asus gives two sets of secondary timings to choose from, in case the more aggressive set fails). So there is a lot more to this. In fact, most of the 3600MHz approved kits ARE NOT rated at 3600, instead being rated at 3733 and above. That is a very important consideration, especially if you want to plug and go. Asrock may require more tuning, but you usually can get more power out of it once tuned. Because of this, you need to consider time to operation and deployment, whether in the house or in a business. Just to give a bit more context.

Edit: In fact, the Avexir AVD4UZ136001704G-4BZ1RR is the only 3600MHz ram kit to run at the rated speeds with the Asrock board.

 

Some more RX Vega news (more like 'leaked benchmarks from an anonymous source).
http://www.tweaktown.com/news/58635...-leaked-benchmarks-gtx-1070-killer/index.html

If the numbers turn out to be accurate though, it might be pretty good for the naysayers. Plus, it might turn out to be consuming less power under gaming anyway (without an undervolt).

Still hype and 'leaks' though.

 

So long as the water cooled version is in the vicinity of the GTX 1080 I will be game. Since the 1080 TI is a waste on about anything but a 7700K in some games I should be fine. Now we just need the ability to order one. I am fine with the ASUS Zenith and only 3200 MHz ram on a boutique build but I will just have to wait a bit longer.

I found even then I can save with some of the add on items threw NewEgg.

 

Links;
http://wccftech.com/amd-x399-motherboard-ryzen-threadripper-roundup-msi-asus-asrock-aorus/
http://www.zdnet.com/article/intel-i9s-new-rival-latest-ryzen-threadripper-is-amds-cheapest-yet/
http://techreport.com/news/32340/noctua-offers-up-a-trio-of-big-booty-heatsinks-for-threadripper

http://www.barrons.com/articles/amds-radeon-rx-vega-threadripper-to-take-share-says-wells-1501554644
http://www.fudzilla.com/news/graphics/44217-amd-radeon-rx-vega-56-comes-in-september
http://4k.com/news/amds-new-new-hig...l-deliver-fantastic-4k-gaming-at-60fps-20723/
http://www.amd.com/en-us/press-releases/Pages/amd-redefines-2017jul30.aspx
https://websetnet.com/spotted-amds-radeon-rx-vega-air-liquid-cooled/

 

AMD Ryzen Threadripper 1950X, 1920X & 1900X....EXPLAINED!!

Cheapest* Threadripper PC + $1500 Gaming/Streaming Setup - August 2017 Builds

Smallest Ryzen Build Yet! The NCASE M1

AMD Ryzen 5 1400 BUILD AN INSANE $700 GAMING PC BUILD 2017! [1440P & VR Gaming!]

BUILD THE ULTIMATE $500 GAMING PC 2017! [AMD Ryzen 3 PC Build!]

 

I'm up to a little over 2,300 with the basics. I am going with 3600 RAM even if the Asus Zenith doesn't support it (initially).

https://pcpartpicker.com/list/87JpPs

 

Ok, so I noticed a lot of news articles are now saying MSI and the Zenith boards support 3600, but they have none on QVL. Asrock has it, but only using higher speeds (the 3733 version of what you got or above). I think all of the companies are likely to support 3600 this round, except maybe the strix boards at launch, but you may need to go one or two rungs higher to achieve it. Just a thought.

Unfortunately 4133 is around $440-450, 3733 is around $400. I do think you hit a nice price, but wonder if the 133mhz is needed to get 3600 stable...

Sent from my SM-G900P using Tapatalk

Edit: btw, nice build!

 

Hardware Unboxed.

 

Asus ROG Zennith Extreme X399 Threadripper Motherboard

Threadripper 1950X & 1920X Reviewer's Kit Unboxing!

HOLY MOTHER OF THREADRIPPER UNBOXING

Unboxing of the AMD Ryzen Threadripper Reviewer Kit

AMD Ryzen Threadripper Unboxing With A Custom-Etched Chip - HotHardware

AMD Ryzen Threadripper Unboxing and Media Kit

How to install the AMD Ryzen Threadripper Processor

AMD Ryzen Threadripper Montage

 

Sent from my SM-G900P using Tapatalk

Edit: GamersNexus comparison of AIO to die placement.

 

How NOT to unbox AMD Threadripper - It was an accident... I swear

 

AMD Project 47 Supercomputer SIGGRAPH 2017

1st PETAFLOP SuperComputer, the IBM / AMD RoadRunner May 25, 2008

AMD EPYC / Radeon Instinct MI25 GPU 1 PETAFLOP SuperComputer July 30, 2017

80 GPU's support 16 VEGA remote sessions, 1280 simultaneous sessions.


AMD Demos Petaflop-in-a-Rack Supercomputer
https://www.top500.org/news/amd-demos-petaflop-in-a-rack-supercomputer/

 

That's helpful, I was wondering what the contact area was like on some of those.

 

Thank you. I'm going with a full custom water cooling solution in the next month. I am thinking instead of a Ryzen color scheme, I will do orange and white. Hopefully EKG will eventually release a block that covers the entire chip (I know it doesn't need it but I think it will look better).

The RAM is a stopgap. When support is better, I'll get better. Over the next 6 months, I'm going all out on this build. The only thing I'm undecided on are gpu's. I'm still not certain whether I will go with a single card or stick with Sli.

 

That video was very informative and I am now going to have to reconsider any AIO not made for TR4. The ideal solution is to have the cold plate positioned parallel to the dies, but there is no way to know that. I hope this doesn't turn out to be a snafu for AMD. People are going to be using AIO or even custom cooling kits designed for single die chips. In the video, he does hint at the chip possibly not getting proper cooling.

This has not posed a problem with mainstream Ryzen, but I am going to take a wait and see approach. On the flip-side, any AIO or custom water cooling kits developed for TR4 are going to cost more.

 

I'm doing the same thing, but would recommend a monoblock. You might as well get something that cools the VRM if you are watercooling (since it is so simple at that point). I'd almost recommend the Giga board for color and for direct listing of running the Trident Z 3600 at its rated XMP. Complementary colors and what have you (although unnecessary). Before you do, I think JayzTwoCentz showed why he steers away from darker colors because of draining an orange out of a system (looks good to start, looks bad later sometimes). Don't know if you were talking the liquid with the orange or just the colors of components. EK plans on both custom blocks for it and monoblocks for some boards. They primarily make monoblocks for Asus, but have also done it for MSI second most, then Giga. They never have made one for Asrock, so I shot a message to ask if they plan to this round.

I hear ya on the ram. Just wanted to point out speed and price because the prices will continue to suck until at least Q1 of 2018. Horrible time to buy ram (I say as I just bought a 16GB kit of 3200 Trident Z on sale for $135 for my router/firewall build). But I hear you. If you can wait until next year, you get more data on compatible dimms AND you get a cheaper price. So, plan on the stopgap being a little longer, if possible (I understand if you can't wait, just give advice to help your wallet; that is when I might go to 64GB).

For the AIO, I'll wait to get info on monoblock support, but that still means relying on my TT Water 3.0 Ultimate in push/pull to make it through until it is available. I don't think it will be the issue some think it is, but we'll find out. Every single review came with a 360mm AIO. You don't do that unless you think that will give good performance for reviews. But, the design of heatsinks is third party support, not something to blame on AMD. AMD cannot force them to make a product. I recommend monitoring the heat per core for all cores to make sure you don't overheat any core while waiting for better solutions.

As to them costing more, it is a niche processor compared to mainstream, so the research and components have to be paid by a smaller pool of users, so of course it will cost more. Name of the game.

 

Myself I would recommend a custom application of ICD on the water block to CPU. At least have it cover the entire area of possible copper contact. Liquid metal may be better but with that large of a contact patch lapping may be required.

 

That video is a much watch video if you are going to use water-cooling, as the adapter plate doesn't solve the cooling patch lack of coverage - single heat source vs 2-4 heat sources now with ThreadRipper / Epyc.

It's going to take some experimentation to find the right paste cover for that situation, LM is bad when you don't have a tight fit, and ICD is a pain to clean between applications - scratching off the printing.

With that poor coverage by the center paste screen application, I wonder how good that's going to cool, adding paste further out from the center circle of paste might help, but then you've got the indented screw holes to deal with - they are *right on top of* the hot cores - certainly a dangerous place to put LM due to leaking potential.

It was helpful to see the internal fins for the cooling to see it's coverage - aligning it long way with the spread of the cores should be helpful - along with expanding the paste circle into an area that matches the internal fin "plate".

Hopefully it's enough, if not we are going to need to see new designs with plates and fins that cover the exact area of the core coverage, either monolithicly covering all 4, or 4 fin plates for covering the exact area over the cores - it's going to be interesting to see how the OLC blocks will be designed.

Fun stuff, with improvements to come.

Here the CLC with unpluggable fitments will come in handy, to install upgraded components as the need is established for cooling TR4.

I would imagine Epyc is going to need this even more than TR4, with all 4 cores active. Even with only 2 of the 4 cores active, the other 2 inert cores are still going to be conducting heat from the carrier through the IHS, so they need cooling coverage as well.

Asetek at least need to move those screw holes over so that they are not right on top of the active cores

Threadripper Cooler, Thermal Paste Coverage vs. Die Area & IHS
http://www.gamersnexus.net/news-pc/3008-threadripper-cooler-and-thermalpaste-coverage-vs-die-ihs

 

Buildzoid rambles about the RX VEGA cards

 

untill AMD can come up with cuDNN , it will not be player in data center, AI workloads that nvidia is on fire about.

Its really stupid from AMD not to after this market

 

AMD has pledged to pursue this market with Epyc and AMD Radeon Instinct GPU's:
https://www.amd.com/system/files/2017-06/TIRIAS-AMD-Epyc-GPU-Server.pdf

Have you looked at the AMD MIOpen software and AMD Instinct hardware? Do they support OpenACC, OpenNMT, Google TenserFlow, or any other open standards?

"Many developers rely on standard libraries when writing code. DL developers using CUDA also leverage NVIDIA’s cuDNN DL library. AMD recently released their free “MIOpen” 1.0 machine intelligence library, a functional equivalent to cuDNN."

NVIDIA cuDNN is a GPU-accelerated library of primitives for deep neural networks, and although there are open standards embraced within Nvidia cuDNN, IDK if Nvidia cuDNN is open - Nvidia may not allow AMD to be compatible at their interface level.

"MIOpen 1.0 supports the Caffe, TensorFlow, and Torch 7 DL frameworks. Developers can replace cuDNN library calls with MIOpen calls to finish porting their DL code to run on AMD’s Vega GPU architecture. MIOpen includes commonly used DL functions that are optimized for Vega GPU architecture, such as convolution, pooling, activation functions, normalization, and tensors."

AMD will need to popularize their own equivalent library over Nvidia's cuDNN, while providing replacement / equivalent calls to port code.

Read the rest of the PDF, and these software and hardware documentation / articles for more details

Welcome to MIOpen version 1.0
Advanced Micro Devices, Inc’s open source deep learning library.
https://rocmsoftwareplatform.github.io/MIOpen/doc/html/
https://rocmsoftwareplatform.github.io/MIOpen/doc/html/apireference.html

The Potential Disruptiveness of AMD’s Open Source Deep Learning Strategy
https://instinct.radeon.com/en-us/t...s-of-amds-open-source-deep-learning-strategy/

Introducing Radeon Instinct: The Fusion of Human Instinct and Machine Intelligence
https://instinct.radeon.com/en-us/introducing-radeon-instinct/

The Radeon Instinct Ecosystem: Designed To Be Open From The Metal Forward
https://instinct.radeon.com/en-us/radeon-instinct-ecosystem-open-source-platform/

AMD MI25
World’s Fastest Training Accelerator for Machine Intelligence and Deep Learning
https://instinct.radeon.com/en-us/product/mi/radeon-instinct-mi25/

AMD MI8
Cost-Sensitive, Scalable Accelerator for Machine and Deep Learning Inference Applications
https://instinct.radeon.com/en-us/product/mi/radeon-instinct-mi8/

AMD MI6
Versatile Training and Inference Accelerator for Machine Intelligence and Deep Learning
https://instinct.radeon.com/en-us/product/mi/radeon-instinct-mi6/

 

id Software Optimizes with Radeon GPU Profiler

 

ICD is not ideal, but without modifications is the only answer I can come up with off hand that is fairly easy to implement. It is good as these are soldered to the IHS so that it then stands a decent chance to transfer the heat to the cooler block through the heat spreader.

The best solution is a custom made water block but what will the overall market be like for the item. On that note some secially made air coolers may end working fairly well too.

 

There are lots of newer pastes that are preferred to ICD now, it's abrasive behavior isn't well liked.

 

I have never had that issue. It has always worked well for me and as of yet no removal issues. I should also note with that I install and am done. There is not a re-do every 3-4 months or so. I have only twice removed ICD.

1St (2 times) time on my C2D X9100 to switch from a P7500 board to P79 and the second time to a Q8200xm.

2nd time after 12 months to show the install was as good as day 1 on my C2Q 8200 on my P79

 

It's probably down to their application of paste, the flatness of heatplate, and the air seal around the edge.

Lots of reports of drying out / pumping out - temps ramp up.

Give it a shot again and let us know how it works out

 

Ryzen 3 Review - Performance & Benchmarks

 

AMD Radeon RX VEGA 64 Rumored Hashrate 70-100MH/s!

Radeon RX Vega May Be The Fastest Mining Card Ever At 70+MH/s!

AMD Vega Frontier Edition Ethereum Hashrate

 

Damn with those hashrates... good luck getting them at a reasonable price lol miners are going to buy them out so fast.


Sent from my iPhone using Tapatalk

 

Get them when they first go on sale, not aftermarket or third party. I'd get two and use them just to pay for themselves and expansion. I just need confirmation first. Then put them under water...

Sent from my SM-G900P using Tapatalk