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

Ohhh, it's a lot bigger than I thought it was...

 

What the hell they got in there, I know Threadripper is big, but not that big! Oh, hang on, do these come with coolers bundled - could almost fit that in there right!?

 

Did you see the Noctua TR4 heatsink I posted earlier? That whole thing is huge, and wouldn't fit in that ThreadRipper Product box.

It also doesn't look large enough for a 2x120 or 3x120 radiator + pump + CPU mounting, does it?

Looking through the "eye" you can see the top of the CPU, and the ThreadRipper CPU itself is huge, and it looks like it's centered in the box, still less room for other stuff.

It's hard to tell from a 2d photo

 

That packaging is mahoosive! (Maybe he's got small hands!)

 

Think you mean "she's".

Sent from my SM-G900P using Tapatalk

 

It's odd to hear, but AMD Bulldozer has HPC cred's, through Cray SuperComputers

Oil And Gas Upstart Has No Reserves About GPUs
July 24, 2017
https://www.nextplatform.com/2017/07/24/oil-gas-upstart-no-reserves-gpus/

"If you work it out, 1 billion cells on 700,000 processors, that’s 1,300 cells per processor. So it turns out they’re only really running in L3 cache, they’re not even using main memory. They’re using an entire cluster to basically use the L3 cache of the AMD “Bulldozer” Opteron chip, which is what the Cray nodes use. And also, since reservoir simulation is memory bound, after two or three, maybe four, cores on a single CPU chip, you don’t get much more out of it. You can add more cores, but you’re not gonna get more performance."

This is the kind of application that Epyc and Vega/Navi are going to help expand AMD's presence.

Cray is still building supercomputers, and with Epyc / Vega / Navi it's going to have a great source of parts for new builds over the next few years. I hope they keep mixing it up with Intel and AMD solutions.

Cray Announces New, AI-Focused Supercomputers
May 11, 2017
https://www.extremetech.com/computing/249149-cray-announces-new-ai-focused-supercomputers

This was a few months ago, amazing how much things can still change in the course of only a few months.

"AMD has made plans to enter these markets with deep learning accelerators based on its Polaris and Vega architectures, but those chips haven’t actually launched in-market yet.

By all accounts, these are the killer growth markets for the industry as a whole, and they help explain why even some game developers like Blizzard want to get in on the AI craze. As compute resources shift towards Amazon, Microsoft, and other cloud service providers, the companies that can provide the hardware these workloads run on will be best positioned for the future.

Smartphones and tablets didn’t really work for Nvidia or Intel–making AMD’s decision to stay out of those markets retrospectively look very, very wise–but both are positioned well to capitalize on these new dense server trends.

AMD is obviously playing catch-up on the CPU and GPU front, but Ryzen should deliver strong server performance when Naples launches later this quarter."

 

This is kind of an odd one, Epyc Security vs Intel Management.

Level1 News July 25 2017: Existential Crisis, Will Robinson
5:30 - AMD Confirms It Won't Opensource EPYC's Platform Security Processor Code

Spoiler: Full show Index

Article Links used in the Show
https://www.one-tab.com/page/-In9aVQyT561-TlDFMYz8A

1:13 - Meet Ataribox, Atari's Retro-Themed Gaming Console Based On PC Tech
2:54 - Facebook may begin testing a paywall for selected media stories as soon as October
5:30 - AMD Confirms It Won't Opensource EPYC's Platform Security Processor Code
7:24 - Porn ID checks set to start in April 2018
8:45 - New sex robots with 'frigid' settings allow men to simulate rape
11:34 - Facial Recognition Coming to Police Body Cameras
12:36 - Ontario gives $1.5M to Carrot Rewards app to help people make healthy choices
14:31 - Report: Many firms are "AI washing" claims of intelligent products
15:42 - Tesla Autopilot Isn't the Cause of the Rollover Crash That Injured Five
17:18 - Why is Comcast using self-driving cars to justify abolishing net neutrality?
20:07 - Google, Apple, Amazon hit record lobbying highs
21:46 - U.S. appeals court upholds gag orders on FBI data surveillance
22:57 - Paywall Shenanigans/China able to intercept images and video before they are even sent
24:24 - Telecom Lobbyists Downplayed "Theoretical" Security Flaws in Mobile Data Backbone
27:33 - US Navy drone-killing laser weapon USS Ponce
29:21 - That Devious Plot to "Zombify" Russia: The Fidget Spinner
30:20 - Intel Launches Movidius Neural Compute Stick: Deep Learning and AI on a $79 USB Stick
32:42 - US and Australia finish a key round of hypersonic missile tests
33:44 - Alphabay shutdown: Bad boys, bad boys, what you gonna do? Not use your Hotmail...
36:17 - Sony's legal quest to remove its leaked developer kits from the Web
37:30 - Elon Musk says he has a green light to build a NY-Philly-Baltimore-DC hyperloop
39:15 - Microsoft cuts off Windows 10 Creators Update from some older devices
41:10 - SpaceX appears to have pulled the plug on its Red Dragon plans
42:46 - Game of Thrones Pirates Being Monitored By HBO, Warnings On The Way
44:33 - Hacker Uses Parity Wallet Vulnerability to Steal $30 Million Worth of Ethereum
47:44 - The GhostCtrl Android malware can silently record your audio and steal sensitive data
49:19 - Nasty Bug Left Thousands of Internet of Things Devices Open to Hackers
50:16 - Trademarks show Amazon has sights on meal-kits, "single cow burgers" and other fast food options
53:20 - Robot Security Guard Commits Suicide in Public Fountain

 

This is why they were protective of the code. With this, it is similar to ME, but it runs off of the ARM chip on the CPU (separate from the dies). This is what manages the security controls for the memory, allows the VMs to be secured and unsecure running next to each other, etc. When has Microsoft given the full look over of every detail of the ME source code? Never! So why ask more of Ryzen? You don't get it from Intel, so it's a wash.

Sent from my SM-G900P using Tapatalk

 

Ha, holy sh*t! Gotta apologise to her, it was her garb mostly!

 

Can Ryzen Play eSports Games at 240 Hz? CS:GO, Overwatch, Rocket League & More!


HW News: HBM2 Production Boost, Threadripper Prep

 

Two bits of news coming today:
1) X399 Motherboard event
2) AMD Earnings Report (after the market closes)

Now, Barclays has estimated a 35% drop and short sellers are preparing, so expect market volatility as you will see it in after hours trading AND tomorrow. Barclays has joined Goldman Sachs in having a low target price for the stock and think that this earnings report will cause a market correction. Expect larger hedge funds and investors to lead the sell, which may cause some to react (retail investors), which would then allow the larger investors to come back in at a lower price.

On the motherboard lineup, we will get our first view of the entire launch lineup for the TR line. In a couple days, we should see more on this leading up to SIGGRAPH.

 

AMD Zen Cpu Architecture Core Scaling - 1700X 1950X 76001 in Cinebench R15
Epyc didn't scale well then with 128 threads, dropping down to 43.33 points / thread from 96.94 for ThreadRipper and 97.94 Ryzen 7. TR and R7 are so close, something's odd with CB R15 @ 128 threads, or did SMT accidentally get turned off??

 

http://www.pcgameshardware.de/Mainb...en-Threadripper-High-End-TR4-Kuehler-1234091/
First report I've seen on the new X399 motherboards, including VRM heatsink redesign and pictures. Also has a picture of the MSI board.

Edit: Asrock boards: http://adrenaline.uol.com.br/2017/0...al1ty-e-x399-taichi-para-ryzen-threadripper-/

Asus X399: http://www.legitreviews.com/asus-rog-zenith-extreme-x399-motherboard-pictures_196357

MSI: http://adrenaline.uol.com.br/2017/0...aming-pro-carbon-ac-para-ryzen-threadripper-/
https://videocardz.com/71100/msi-showcases-x399-gaming-pro-carbon-ac

 

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-eventDetails&EventId=5259318
AMD earnings call will be webcast. If you want to tune in, it is in an hour. ( @hmscott )

 

http://www.marketwatch.com/story/amd-jumps-after-beating-on-earnings-raising-forecast-2017-07-25

Update - stock is rising as AMD beat estimates by $60M on an adjusted basis and gave $0.02 EPS. All those people that shorted are going to get hurt in this!!!

 

Who in their right mind would short AMD right now?

They all need to switch their "shorts" to Intel

 

Barclays estimated a 35% stock tumble. Goldman put the price very low as well. The market has been lamb-basting the Epyc release and limited availability (Su said supply chain is strong, but demand was stronger than expected, which is why they have increased CapEx during the quarter and preparing for next quarters launches, which will include the Pro line, Epyc wider availability, Ryzen 3, and Threadripper, as well as Vega). They also revised revenues for the year from low teens to mid to high teens. So, it was people listening to analysts that base on history rather than facts on the ground and that go with herd mentality. Many prepared for a short in case things moved, that way to make on the move.

Right now, 9% up in after hours trading.

 

Someone confused "not able to meet demand for Epyc" with "not going to make numbers".

When you are running at maximum output - well past "wildest expectations", that's a good thing, and revenue is better than normal.

Yeah, those analysts are going to need to re-tune their thinking and listen more carefully instead of jumping to conclusions.

Wow, if investors were expecting a 35% drop there were a lot of losers where they should have been a bunch of happy winners.

 

Well, I listened to the conference call and most of the hedge fund and investment bank analysts were asking things that were disproved rumors (like the "can't do virtualization" based on a little used instruction set that was fixed within the month in March, and the other based on not updating the linux kernel, which this didn't effect Windows Server, etc., although there were minor compatibility issues on certain virtualization software that AMD has worked closely with vendors to overcome or add support), or silly questions where Su had to explain the difference between OEM and channel sales on supplies of Ryzen and that OEMs didn't start selling until the end of June, the Epyc demand, to ignore the down numbers in mobile because that refresh isn't coming until the end of the year, etc. Stuff we have already discussed in this forum tirelessly. They talked about the thin channel inventory on graphics cards and that Su expects to have strong growth aside from the portion related to crypto, etc.

It is more they did not know what we knew by studying release schedules, following the product availability, etc. To be fair, they have to manage a lot of stocks and keeping up with any one company takes a lot, so let's not overly criticize them for not seeing it (except for Barclays and Goldman). There was (from Barclays, if I remember) a question on how AMD is doing with the smaller dies with the glue fabric compared to monolithic chip design, as well as posing a loaded question to make it appear that there may be concerns on Epyc gaining traction in the $16B server market, as though the climate was so different compared to a decade ago that no one will want the product compared to Intel's. Then which market segments (cloud, enterprise, etc.) they are focused on and what to expect on the revenue from those and how they will appear in earnings.

Some of the questions are good, getting at needed information and/or just wanting to get the CEO or CFO on record. Some seemed skewed. Some just didn't seem to have what was needed on prep work (but they may have been busy with other work). But, overall, nothing really groundbreaking on the call.

 

Here's a few more x399 articles:

AORUS X399 GAMING 7 is ready for AMD Ryzen ThreadRipper
http://www.tweaktown.com/news/58564/aorus-x399-gaming-7-ready-amd-ryzen-threadripper/index.html

It looks like X399 pre-orders will start in less than 24-hours
https://www.overclock3d.net/news/cp...pre-orders_will_start_in_less_than_24-hours/1

GIGABYTE X399 AORUS Threadripper motherboards soon available for preorders
https://videocardz.com/71145/gigaby...per-motherboards-soon-available-for-preorders

MSI showcases X399 GAMING PRO CARBON AC
https://videocardz.com/71100/msi-showcases-x399-gaming-pro-carbon-ac

ASRock details X399 Fatal1ty Professional Gaming
https://videocardz.com/71110/asrock-details-x399-fatal1ty-professional-gaming

ASRock and MSI X399 Motherboards For AMD Ryzen Threadripper CPUs Detailed – Pre-Orders Start in Less Than 24 Hours
http://wccftech.com/amd-asrock-msi-gigabyte-x399-motherboard-ryzen-threadripper-cpus/

More pictures of ASUS X399 ROG Zenith Extreme emerge
https://videocardz.com/newz/more-pictures-of-asus-x399-rog-zenith-extreme-emerge

 

What I find kind of funny is about every media person is saying how TR is not really meant for gaming but the serious content creator and such that will use all those cores. But just about every motherboard out thee references "Gaming" in the title.

 

See:
http://www.anandtech.com/show/11660/amd-announces-q2-2017-results-ryzen-makes-an-impact


I don't know if it was Ryzen that made this impact for AMD? The GPU shortage for consumers certainly raked in the bucks for AMD too.

 

TANWare, because; marketing.

 

You still didn't answer my question. But thanks for trying.

So many words; so little info...

 

https://en.wikipedia.org/wiki/Glue_logic , First off the term is "Glue Logic" not Glued together and is specific to certain scenarios not a single design SOC. Intel never referenced this type of design so no back pedaling needed.

 

We could tell from context and I definitely agree.

On another note, the Asus Zenith board only has 2800MHz ram officially supported. MSI's is unknown, but these two boards are the only ones with 2x8-pin connectors. The Asrock and Giga boards have 1x8-pin and 1x4-pin. The Asrock is supporting up to 3600MHz ram. If MSI has the support for 3600, then it may lean that way.

On known mosfets, you have 10+3 on MSI, you have 11 on the Asrock (likely 8+3), and don't know yet on the Asus or Giga.

Edit: Giga is using an 8 phase design (likely an 8+3) on the gaming 7.

Edit 2: expect $350-500 for a motherboard, going by X299 pricing.

 

Stop the bickering please.

 

No one last words, next you both get banned as I am tired off this!
Links;
https://www.overclock3d.net/news/cp...edly_include_a_bundled_asetek_retention_kit/1

 

Neat. I recall when I was in a B&M store they were giving away retention kits with each purchase. Cool that they're just going to ship one with it.

 

I'm still waiting to hear about the monoblocks from EKWB and Bitspower. But I'm glad to see the industry getting behind TR early.

Still wondering if lapping will be needed on these large chips...

Sent from my SM-G900P using Tapatalk

 

Maybe, but I'd wait to see how much difference it makes from other people doing it first.

 

Vega looks to be a no go against the GTX 1080 TI, a shame.
https://www.digitaltrends.com/computing/radeon-rx-vega-seen-in-3dmark/

 

I'm one of those people you wait for on lapping info! LOL! It is more a question whether it is worth the effort if you hit a wall on overclock like the Ryzen 7 chips. But, in the review by Wizzarty of Ryzen on LN2, it showed the IHS was not perfectly flat. That is why I'm considering it. Check out what I did to the IHS and heatsink for my P770ZM:

 

Depending on the price point, maybe they were testing something aimed at the 1070?

 

Shiny. How much did it affect heat transfer in that case?

 

We have a couple days and we'll know the truth, whether these are the 1080 competitor, the 1070, or the flagship, and price points.

 

Significant! I'll try to find my notes (it's been awhile since then, but I posted it in the owner's forum in the P770 and 750ZM). I think I got more from going to liquid metal, but I got at least 4C off of it (probably got closer to 8C, my heatsink was bowed when I started so by getting it flat, I got huge gains, but didn't separate fully the effects of the liquid metal used between the IHS and heatsink, I just know that the liquid metal between the die and IHS gave the largest performance boost. So, on a laptop, you maybe get a couple hundred mhz by perfecting contact, but getting rid of the toothpaste is extremely important! But that is how I took my laptop outside in -4C weather (around 24F) and got a 4790K to benchmark at 5GHz. Hotwell was a hard one to get that high, but I bought a great CPU (voltage of [email protected], was using under 1.3 at 4.8GHz, and under 1.4 for 5GHz if memory serves).

 

I'm hopeful, seems like AMD would enter in the mid range first instead of going straight for the extreme cards. But time will tell.

 

Well, let me put it like this: if your overclock stops at [email protected], and you cannot squeeze any more out of it, but your temps are at 70C under full load, lapping won't help. In that regard, all you are doing is lowering temp to help with degradation, which will be sped up by that voltage. With Ryzen, it seems you will hit a wall, past which you cannot get around without a chiller, a single stage, DICE, a cascade, and LN2. Not only that, some of the IMCs are picky with the cold on top of that, making it even more of a pain. Also, once you find voltage doesn't help, it is good to back off voltage until instability, that way to find the maximal voltage.

I couldn't find the exact amount, but am starting to think I posted it in the P750ZM, as toward the end, that's where most of the action was going on. (I only made it moving from the end to where I flashed the SPD to load 2400 on my Kingston DDR3 so I didn't have to reload it each time and didn't have to put in an XMP for the ram; CL11 13-14-28-228 CR-2T http://forum.notebookreview.com/thr...to-the-batcave.767105/page-1008#post-10276953). I dropped about 20C+ with the lapping, the screw mod (going to straight screws instead of springs and using pressure contact film to verify contact pressure across the chip), CLU in both die/IHS and IHS/heatsink.

On the GPU - AMD did a good job either way, but I'll wait. Plus, on the graphics side, AMD always sucks at release, with them reaching the top speeds about 6 months after release with driver tweaks. So I am hopeful as well. Also, with the HBCC and HBM, I'm wondering if game optimization is needed to better utilize the cards as well. But, as you said, time will tell.

 

In all honesty we also don't know if those 'leaks' are accurate - though they weren't much off on Polaris either.
Plus, it is surprising that RX Vega would score as much as 1080 in gaming whereas Vega FE (which supposedly has same hardware) in various pro software (Cinebench OpenGL included) outpaces even the Quadro P6000 (which should be based on the 1080ti) - though we know that compute doesn't always translate to gaming.

Plus the power draw can likely be addressed via undervolting and stabilize the GPU from throttling down to 1440MhZ (and keep it at 1600 MhZ).

Some people on REED forums also speculated that a PSU of lower quality can slow down GPU's.
It supposedly happened with 290x and throttled down the card by a lot.
Perhaps something similar might be happening with FE (coupled with overvoltage?).

Then again, we are basing all of that off Vega FE, which was tested in games with relatively unstable drivers and stock voltages (which are simply too high and induce massive throttling).
The readings/gaming performance (if I remember correctly) were/was all over the place.
We don't know (yet) how RX Vega will perform in games, and we had previous 3d mark scores that were 'leaked' which turned out to be fakes.

P.S. I'm not trying to justify anything btw. I accept the possibility that RX Vega simply might not be able to compete against 1080ti despite how much drivers mature and could easily be at the level of 1080 or slightly above it (though it would be nice for AMD to be competitive with Nvidia 1080ti in this regard as they need it after all the promotional stuff they put out for Vega).

 

I have to say this could have been essentially one reason for the delays. No matter how hard they try they can not compete against the 1080TI. No matter what the hype or press. In the end Vega may not do much for strengthening their place in the video card market, at least not yet.

 

If all this is true, then I'll likely get two Vega or 1080s to start and wait for 7nm. Both companies will get a large jump at 7nm.

Sent from my SM-G900P using Tapatalk

 

Myself, I am looking at a 1080 TI and TR for now.

 

Same. My initial build will have two 1080's in SLI but I will settle on a single 1080 TI eventually.

 

Only reason I'm looking at two cards is it will need to play games over the network and be able to support two games played at the same time. One Ti would give better performance overall and less troubleshooting, but it couldn't push two games going with the settings I want at the same time.



Sent from my SM-G900P using Tapatalk

 

Like the RX480 / 580, it wasn't the fastest, matching the 1060, but it sold so many units along with Ryzen revenue to bring profitability back to AMD and bring their stock up too.

The RX(1080) will be a good seller too, based on price performance, and will be a good match for those that want an AMD GPU to go with their Intel and AMD CPU's.

A few performance points ahead isn't any reason to buy Nvidia or Intel, give AMD some $ encouragement to keep fighting, it's not gonna kill you to have AMD for a few generations while they are just off the top performance, saving a bunch of money (FreeSync too).

At this point few games are going to *require* a 1080ti

AMD is doing great for coming in with so many products, coming from behind and bringing themselves into contention.

It's still a few days away from release, and months from peak performance.

 

I game @4K@60, so wildlands is even pushing a Ti on ultra. But love your points here!

Sent from my SM-G900P using Tapatalk

 

The reason I would build it with a 1080 TI is whatever I do put in it will probably never get replaced/upgraded. Also water cooled at 375w if it were a 1080ti killer I might do, but not for 1080 only performance.

 

I wonder how many 1080ti systems get built vs 1070 / 1080?

The price of a water cooled 1080ti over the water cooled RX(*1080*) should be about 2x+

Not sure if the performance difference is worth paying 2x.

Seems like the same story as AMD CPU vs Intel CPU

 

Here's my problem. It is 18 months, give or take, before 7nm, which should be a huge jump. For the cost of one killer 1080 Ti, I can get two 1080s and blocks (maybe a little more, but not much). So, my thought is do that, get by, and save up for the large jump with 7nm. I may be doing a TR CPU upgrade around the same time, but will lower certain other costs in the interim. So, that is my thought (still building a firewall, squid, etc.).

Sent from my SM-G900P using Tapatalk