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

EPYC™ Tech Day: Kevin Lepak

 

http://www.tweaktown.com/news/58385/amd-takes-radeon-rx-vega-road-gamers-next-week/index.html

 

RX Vega = 1080 Ti & 1070?!
It looks like there will probably be 3 RX Vega GPUs, plus AMD is introducing Vega earlier and Ryzen deals.

 

Gigabyte AB350 Gaming WIFI ITX AM4 Motherboard Review

 

Some more news on Vega:
http://wccftech.com/amd-rx-vega-xtx...m-rx-vegas-ready-mass-production-late-august/

Wccftech is a notoriously unreliable website though, so treat the 'release dates' and most other information with healthy skepticism.
So much nonsense came out of that website that it's obscene.
the only reason I'm posting it is because this time it seems to have sources that confirmed the names of different vega GPU's and we might have an estimate on release dates (again, grains of salt).

 

https://www.extremetech.com/computi...window-speeds-prices-details-upcoming-ryzen-3
https://www.pcgamesn.com/amd/amd-threadripper-price
http://www.anandtech.com/show/11636/amd-ryzen-threadripper-1920x-1950x-16-cores-4g-turbo-799-999-usd
https://www.pcper.com/reviews/Proce...-and-1920X-Announced-Flagship-Performance-999

Disapointed only a little over 3000 on R15. Then again early in the game.

 

http://forum.notebookreview.com/index.php?posts/10563858
New info on TR!!!

Sent from my SM-G900P using Tapatalk

 

Remember how much was gained going from 2400 ram to 3200 on performance. Here, 2400 and 2666 are listed as the top speed, meaning ram speeds above this were not used! we're going to see better scores. And Epyc showed as the speed of ram increases, IF throughput scales amazingly. Who wants to throw some 4133 in and see if they can get 3600 or above stable. I do!!!

Sent from my SM-G900P using Tapatalk

 

And here we go!

 

From the announcement, hopefully these are stepping 2;
https://en.wikichip.org/wiki/amd/ryzen_threadripper/1920x
https://en.wikichip.org/wiki/amd/ryzen_threadripper/1950x

Note it does not show 1TB as max ram but 128GB.

 

Some dude already received his watercooled Vega FE.










source: reddit

 

Yes, but they usually set that by what is available to the market. This is quad channel, with 8 slots. In regular systems, that comes to 8*16GB= 128GB. 32GB DIMMS are not common in the regular system. Also, very few consumer boards, even high end consumer boards, support over 16GB per DIMM. So that doesn't necessarily give insight into stepping.

Sent from my SM-G900P using Tapatalk

 

EPYC™ Tech Day: Gerry Talbot
AMD Epyc I/O and Infinity Fabric Electrical Interfaces
New re-issue by AMD, not sure yet what's changed from the first one AMD pulled...

Original link, in case it ever comes back...

 

Look what we have here! LN2 and Ryzen!

http://www.tomshardware.com/reviews/overclocking-amd-ryzen-ln2,5116.html

edit: tl;dr at approximately -200 degrees Celsius, each Ryzen chip hit on average ~5GHz

 

Ryzen and overclocking is a sad chapter. A no from me. And what a silicon lottery A Ryzen equipped laptop @Mr. Fox

 

I guess it is a good balance.

Intel for clockspeed. AMD for cores.

 

Put the other way. If I go for a medium priced desktop setup with 8 core chips. AMD 1800X vs. i7-7820X is a bad choice if you enjoy OC'ing. AMD offer of course a lower price. But I would gladly pay the extra $210 for the chips here home for more clockspeed aka more fun. And the lower priced 1700/1700X isn't a choice due the bad binning.

 

Completely agree.

However, there's a reason the 1600x, 1700x, and 1800x exist.

People do not feel comfortable overclocking, at all. I can with high certainty say that this accounts for most of the people out there, most gamers even!

Simple things such as updating the BIOS is out of the question for most people. I have loads of friends that don't even update their drivers lol.

 

You have to remember that quite a few people enjoy Overclocking as a hobby and sport. If this is your primary goal AMD is not your choice. I was locked heavily into it myself for a good period. There is nothing wrong with it but it is only one small aspect of computing.

Once I realized allot better gains in the computing experience were to be had my interests drifted elsewhere. Walking away from the higher clocked C2D into a C2Q running SSD's and better ram etc stated showing me the light. I would not even think of running a primary HDD again, you just live and learn.

 

So true.

I've updated many systems from the BIOS to the drivers to the software they use to 'current' levels and one even asked me what I changed (hardware-wise) to make their system so much faster/smoother (the system was left untouched, as delivered, for over half a decade...).

I always tell my clients the potential risks of flashing firmware on their systems but after I verify that the UPS they're connected to is capable of keeping the system up for at least 5-10 minutes while I do so; they almost always let me.

 

Links;
https://www.extremetech.com/computi...f-intel-skylake-sp-xeon-massive-server-battle

 

OC'ing isn't an eSports event, it's very low numbers of people.

Besides, who wants the computer to come out of the box not providing full performance. We shouldn't need to hunt for another 10%-20% speed improvements. Ship'em that way to begin with.

Like Ryzen

 

I agree with running toward the top, but I also enjoy squeezing as much as I can out if it!

Sent from my SM-G900P using Tapatalk

 

No matter who makes the system, proper maintenance of firmware, drivers and removal of bloatware etc. goes a long way. I will leave a system and inevitably get the "It's better than the day I got it bit. That is so long as it is not so anemic there is nothing to be done.

 

And once again how I read the above is 'the discerning customer, wins'.

 

I thought I already posted this, but search doesn't find it...

AMD Ryzen™ Threadripper™ and Ryzen™ 3 Product Updates

 

AMD Outlines New AGESA Memory OC Settings For Gaming, Explores Frequency Versus Timings
http://www.tomshardware.com/news/amd-ryzen-agesa-memory-bios,35012.html

Memory OC Showdown: Frequency vs. Memory Timings
Posted by rhallock Employee in Gaming on Jul 14, 2017 11:20:08 AM
https://community.amd.com/community...emory-oc-showdown-frequency-vs-memory-timings

 

AMD Ryzen Threadripper 1950X & 1920X: Specs, Pricing & My Thoughts!

Intel is LOSING its DIGNITY - WAN Show July 14, 2017

Spoiler: Topic Timing Index=Several of Interest Highlighted

0:00 Intro...
4:08 Zillow orders popular blog to take down photos
11:55 Stupid juice machine now “only” $200
16:55 Luke talks about being at Twitch in Silicon Valley
17:55 Luke's mum calls up for a guest call
20:25 Throwing Shade: Intel Xeon SP Slide Deck attacks AMD EPYC Server Lineup
38:35 Sponsor : Comic Bento
41:35 Sponsor : Fresh Books
43:49 Sponsor : IFIXIT
45:19 AMD Ryzen Threadripper and Ryzen 3 Product Updates
49:10 The Skylake-X Mess Explored: Thermal Paste And Runaway Power
52:57 $200 Wireless Oculus Rift in the works for 2018
58:45 KFC Phone. Yup.
1:02:17 OnePlus 5’s Display is Upside Down
1:04:05 Google Pixel 2 design leaked
1:05:03 Logitech Acquires Astro Gaming

 

AMD Ryzen Laptop Preview: Asus ROG Strix GL702ZC Benchmarked
Written by Matthew Lambert July 13, 2017 // 7 p.m.
https://www.bit-tech.net/previews/amd-ryzen-laptop-preview-asus-rog-strix-gl702zc-benchmarked/2/





More benchmarks on the site

Reddit thread:
AMD Ryzen Laptop Preview: Asus ROG Strix GL702ZC
https://www.reddit.com/r/hardware/comments/6n3f3r/amd_ryzen_laptop_preview_asus_rog_strix_gl702zc/

 

Thought the Linus Tech video had a good breakdown of Intel's Slides. Some funny crap Intel put together

Starts about 22 minutes in

Gets funnier 29 minutes in



Moderator; (moved here as the media was already linked within the thread)

 

CPU hitting 95 deg C at full load and maintaining 3.3 GhZ?
Is Asus skimping on the cooling or something?

 

There is only one defensible position for Intel here I can see. And it requires a bit of history. If everyone remembers back at just before Ryzen 7 launch Intel knew of the HEDT and its plans at the time to keep up with AMD was only up to 12 cores in the next iteration. At the time I am sure they were preparing press material as well.

I am sure Intel was keeping track of Epyc and TR which early on, like Ryzen 7, was based on stepping 1 of the Zen core. I think with it there were optimizations just not doable under AGESA that were detrimental to performance. This probably left the new architecture much in the same performance boat Bull Dozer would have been in under the same multi core chip.

This can be bolstered somewhat by all the leaked Benchmarks under Stepping 1. Primarilly of the TR but it shows that silicon did exist at some point and this is what AMD was using. It show where IF was just not giving as good scaling result to the multi CCX complex chips at all. This where is seems with Ryzen7 it scaled reasonably well.

At some point Intel got wind of the enhanced performance capabilities of both Epyc and TR under stepping 2 of the cores and this started the scramble of both HEDT and XEON chips to now compete with the upcoming AMD chips. Not that Intel lacked the tech but was now going to have to call on it to both upgrade all of their chips, prepare for the price competitive market and possible subsequent market share loss.

Now this is evident in the partner slide as we know Epyc and Supermicro exist. Ryzen and Gigabyte, MSI at least exist as well up to TR. Microsoft exists, there are I am sure others as well on the slides. So to me these are leftovers from the time Intel felt ZEN just posed no threat, and if it were true were no better than Bull Dozer and a "Glued Chip" as it was.

Now I doubt this will happen as it means admitting to Ryzen forced reorganization, industrial spying, possible competition, possible revenue loss etc. etc. etc.. But this is just my thoughts here.

 

EPYC™ Tech Day: Gerry Talbot
AMD Epyc I/O and Infinity Fabric Electrical Interfaces
New re-issue by AMD, not sure yet what's changed from the first one AMD pulled...

Original post:
http://forum.notebookreview.com/thr...ga-polaris-gpus.799348/page-232#post-10564531

Nino Magazinovic 15 minutes ago
"Without AMD we would today from Intel get only 32bit 2core cpu for 300$. No 4k videos and games, No VR, nothing. Now, AMD with infinity fabric scalling brings us again BACK TO THE FUTURE, 8K gaming and VR."

 

AMD Ryzen 3 1200 & 1300X Unboxing

 

Something is amiss here.
The RX 580 is 65W with radically lowered clocks (over 30% it would seem) and it still manages keep up with 1060 (120W) while losing only 5-10% performance it would seem in DX11 and maintaining an advantage in DX12.
The CPU's in comparison are 65W on AMD's part (Ryzen 1600 - 6 core part no less) vs a 7700HQ (4 core, 4 threads and 45W).

The quarrel I have here is with the premise that a 65W Ryzen CPU is seemingly unable to maintain its turbo speeds and approaches high temperatures as a result, whereas we see Intel and Nvidia laptops (7700 HQ and 1060, both of which have a combined TDP of 165W - while on AMD's side, there's a combined TDP of 130W - 35W less) capable of maintaining their turbo speeds and achieving lower temperatures.
Why is it seemingly impossible to keep the AMD parts (CPU in this case) from throttling and achieving high temperatures, whereas a higher TDP laptop system that has Nvidia+Intel can more or less manage to do just that at about 15 degrees C lower temps?

On average, I would imagine AMD would have an advantage on load power draw (which is higher on both systems) and OEM's would have an easier time cooling it.

This seriously baffles me.
Is Asus really that bad with cooling in their laptops or is this just a pre-draft issue that needs and can be sorted?
Could it also be thermal paste related? (doubt it would result in this much discrepancy).

It would be pretty stupid for Asus to do an amazing thing like putting together a proper FULL AMD system in such a long time (with 6 cores no less on the CPU end) and then skimp out on the cooling, resulting in poor performance and throttling?

Why can they seemingly keep GPU's with much higher TDP's and overall power draw stable and properly 'fed' in terms of power, but have issues with a CPU?

I'm going to wait for the final release and see if these issues have been or can be addressed.
I really want an all AMD Ryzen (8 cores/16th) RX 580 laptop... but not if Asus botches the cooling.

 

That's a lot of doubt for a simple first preview of an engineering sample, maybe wait for real information and production reviews before jumping overboard.

All the vendors botch the CPU cooling to some extent, and that's why so many OCD'OC'rs repaste.

You can get a good one, and most Intel CPU's you can live with by undervolting, but with AMD Ryzen it's unknown how that's going to play out.

I've had 7 Asus laptops, and haven't needed to re-paste any of them.

Only one laptop had a fault develop with a GPU (AMD), and Asus RMA fixed it in 2 weeks with no further problems - and it's still working.

Let's try to stay positive until facts arrive to prove otherwise.

 

I'm aware of the issues with pre-release engineering samples... that's why I said I'd wait for the final release before drawing any conclusions.
It's just that even with an engineering sample this close to release, you'd think ASUS would have things like the cooling sorted - but I have to admit I was indeed skeptical of pre-release info to begin with.

As for undervolting Ryzen... don't know... but I tried a quick search on it and came up with this:
https://twitter.com/BitsAndChipsEng/status/843871819585208322

Those are total system power draw (so actual CPU would be pulling much less), and someone was able to undervolt a Ryzen 1700 on 3.7GhZ to 1.05V it would seem on Reed forums.
Now if we can get that kind of silicon into these Asus units... that would be NICE for undervolting which would of course result in lower power draw and temperatures - but Asus should really work on this to get those in proper ranges on stock speeds from the get go - no more skimping from OEM's.

I'm also aware of the thermal paste issues commonly present with laptop OEM's. I already have a quality thermal paste which I'm using in my Acer.

 

I am not sure if this is true but if so then the 12 core is a 3x3x3x3. I am not too sure as to increase yeilds they could also be a running at 4x4x4x0 but this would change the cache amount as well. The second link has the cache of the 1920x and 1950x being the same amount.

https://www.techpowerup.com/233945/amd-readies-nine-ryzen-threadripper-models?cp=3
https://www.techpowerup.com/cpudb/1910/ryzen-threadripper-1920x

 

All with 3 cores makes more sense with how they connected epyc partners on different sites and different chips.

Sent from my SM-G900P using Tapatalk

Edit: until we see something showing a work around to the mirrored connection, I'd be dubious. Now, the 3x2 complexes could mirror each other. This may be how they want to approach it, but we don't know if the ccx communication to the other complex on the same die is core dependent as well, which also uses IF.

 

Intel's 3DXpoint Optane 32GB - Running on AMD's AM4 Platform with Ryzen R7!!! It works!! But..

 

I lean more toward the 3x3x3x3. I think this is the reason for just the two skews. They took the optimal configurations and are releasing just those. The 12 core was though to be a $599 or so price point but as we see it is quite a bit more. it may be the off counted CCX cores scaled weird such as say a 3x2x3x2 for 10 core. With only $200 between the 1920x and 1950x it does not make much sense for a 4x3x4x3, or does it?

 

You could end up buying a fake AMD Ryzen CPU on Amazon... be vigilant
https://www.pcgamesn.com/amd/fake-ryzen-cpus-sold-on-amazon

"Reports are surfacing of fake AMD Ryzen CPUs being sold through online store giant Amazon. The key’s in the details…

It would appear scammers have been fiddling with our beloved processors, after two Amazon users have posted news of receiving faked Ryzen processors only a week apart from each other. The first report came a little over a week ago, showing an Intel chip with the blue team branding scrubbed off and replaced with AMD markings.

A second followed soon after, which redditors have noted featured a different CPU SKU, but the same design of unofficial branding as before, leading to speculation that the two instances are the work of the same scammer (or team of scammers)."

 

Crysis Through the Ages | Nvidia vs AMD

 

AMD Ryzen Threadripper might get AIO LCS Liquid cooling bundled

"This news in fact would likely make Intel even more nervous as that would offer nice additional value. As you know, AMD will do and offer a LOT of make Threadripper as attractive as possible. Both initial processors are fully unlocked and seat themselves onto the new Socket TR4."

 

Level1 News July 18 2017: R.I.P. Gluestick
9:51 - AMD Ryzen Threadripper 1950X 16-Core And 1920X 12-Core CPUs Primed To Undercut, Outperform Skylake-X
13:08 - Intel Says AMD EPYC Processors "Glued-together"
15:45 - Intel required to pay $1.4 billion fine over anticompetitive tactics against AMD (remembering last time)
16:38 - AMD Ryzen Makes Significant Progress Capturing Share Of Desktop Processor Market

Spoiler: Full Show Index + Article Links

https://www.one-tab.com/page/SxZx0QAORVmf263IS0RNEw

0:53 - Visa considers extending "war on cash" business incentives outside US
5:27 - Amazon Prime is on pace to become more popular than cable TV
7:23 - DHS Goes Biometric, Says Travelers Can Opt Out Of Face Scans By Not Traveling
8:26 - EFF Officially Appeals Tim Berners-Lee Decision On DRM In HTML
9:51 - AMD Ryzen Threadripper 1950X 16-Core And 1920X 12-Core CPUs Primed To Undercut, Outperform Skylake-X
13:08 - Intel Says AMD EPYC Processors "Glued-together"
15:45 - Intel required to pay $1.4 billion fine over anticompetitive tactics against AMD
16:38 - AMD Ryzen Makes Significant Progress Capturing Share Of Desktop Processor Market
18:38 - Comcast, Verizon, and AT&T want Congress to make a net neutrality law because they will write it
21:28 - This graph shows how Netflix speeds changed after Comcast deal [Comcast Roundup]
23:41 - Aussie Prime Minister Says The Laws Of Math Don't Apply In Australia When It Comes To Encryption
26:13 - China Tells Carriers to Block Access to Personal VPNs by February
28:24 - Smart Device Breaks Up Domestic Dispute By Calling the Police
30:15 - Amazon may give developers your private Alexa transcripts
31:50 - Millions of Verizon customer records exposed in security lapse
33:59 - Securitas CEO Declared 'Bankrupt' After His Identity Was Stolen
35:49 - Congress Close to Approving a New Space Army
36:55 - What happens when automation comes for highly paid doctors
38:45 - "World's first robot lawyer" now available in all 50 states
39:54 - The Audi A8: the World's First Production Car to Achieve Level 3 Autonomy
42:06 - Ethereum Price Weekly Analysis - ETH/USD Leaning Towards $150
44:09 - SoundCloud Cuts 40% of Staff in Push for Profitability
45:24 - PC shipments declined, Gartner says
46:31 - Chrome Extension Sold to New Dev Who Immediately Turns It Into Adware
48:06 - Lights, Cameras, CRISPR: Biologists Use Gene Editing to Store Movies in DNA
49:58 - Nobody in Hong Kong wants a Tesla anymore
51:27 - Russians now need a passport to watch Pornhub
53:17 - Musk says A.I. is a 'fundamental risk to the existence of human civilization'

 

Facebook Q&A Happening today:

Spoiler

It has been quite the month for datacenter news across the industry. With a lot of information flying around, we want to take a moment to engage directly and answer questions regarding our EPYC™ 7000 series of high-performance server processors, launched in June and designed to once again deliver AMD innovation to the datacenter.

We invite you to join Forrest Norrod, SVP and GM of Enterprise, Embedded & Semi-Custom Products (EESC), Kevin Lepak, AMD Fellow, and me, Scott Aylor, CVP and GM, Enterprise Solutions on Tuesday, July 18 from 12:00-1:00 pm CT for a Facebook Live Q&A where we will be addressing questions from the community about our “Zen” architecture, EPYC processor performance and workload metrics, and our robust ecosystem of OEMs, ODMs, cloud providers and hardware partners that stand behind EPYC. We’ll also be available to answer other technical inquiries on what this new processor means for the datacenter. » Join Facebook Live Q&A

For more background, below are links to a comprehensive list of EPYC processor information resources, including white papers, AMD Datacenter Tech Day presentations, videos, Launch Day executive presentations and more. You can also access this information in one place on our EPYC landing page.

AMD Server Marketing links:

Spoiler

Performance

• “AMD EPYC SoC Breaks Records with SPEC CPU Benchmarks” Performance Brief
• “AMD EPYC SoC Delivers Exceptional Results on the STREAM Benchmark on 2P Servers” Performance Brief
• “EPYC: Designed for Effective Performance” Whitepaper
• “Power / Performance Determinism” Whitepaper
• “SpecCPU 2017 & Changing Performance” Whitepaper

Architecture

• The Power of EPYC – Tech Day Presentation | Video
• “ZEN”: Built for Servers – Tech Day Presentation
• Architectural Overview – Tech Day Presentation | Video
• I/O and Infinity Fabric – Tech Day Presentation | Video
• Power Management – Tech Day Presentation
• “AMD EPYC Brings New RAS Capability” Whitepaper
• “EPYC Offers x86 Compatibility” Whitepaper

Disruptive Technologies

• “AMD EPYC Empowers Single Socket Servers” Whitepaper
• “Trusting in the CPU: Getting to the Roots of Security” Whitepaper
• SEV and VM Boot Security Video
• “AMD EPYC Empowers Server GPU Deep Learning” Whitepaper

Partner/Customer Ecosystem

• Microsoft Azure Video
• Baidu Video
• 1&1 Video
• LexisNexis Video
• Samsung Video
• Customer Testimonial Video

 

Some news on RX Vega:
http://wccftech.com/amd-radeon-rx-vega-gtx-1080-battlefield-1-comparison/

Apparently this RX Vega was tested in Battlefield 1 against GTX 1080 and ran similarly.
We know nothing of which Vega card was used (since there are reportedly 3 of them being released), nor anything about FPS.

In other news, Vega FE was overvolted (as I suspected) once again contributing to high power draw and thermal throttling.
http://www.gamersnexus.net/guides/2...tion-undervolt-benchmarks-improve-performance

The same thing is likely to be expected in RX Vega in order to increase yields.

However, what is interesting to note that in PRO software, Vega FE frequently outpaces P5000 and in some PRO scenarios beats P6000 (while being only marginally slower the rest of the time by a few %).
So, this particular 'night and day' difference between PRO performance and gaming performance on Vega is surprising at best.

It is difficult to see why would Vega have a gaming IPC on the same level as Fury line, but radically outpace it in PRO software (higher clock frequencies not withstanding as they don't correlate with several times increase as was seen on FE).

So, what's happening here?
Is RX Vega going to be much less capable in games than it seemingly is in PRO software, or is this a problem with drivers simply not being ready?
Or was the RX Vega tested in BF1 a case of cheapest RX Vega?

 

I'm betting it is implementation of primaries shaders on RX Vega, but NOT FE. But we'll find out...

Sent from my SM-G900P using Tapatalk

 

Well, there's something else to take into account.
The Wccftech article was apparently written on a basis of a Reed forum poster who went to the event.
It seems like shaky grounds on which to construct an article.

 

Links;
http://segmentnext.com/2017/07/18/amd-threadripper-1900x/