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

AMD Community Blog links:

We built the world’s fastest processor for Ultrathin notebooks
https://community.amd.com/community...d-s-fastest-processor-for-ultrathin-notebooks

25x20 On Track: AMD Ryzen With Radeon Vega Graphics
https://community.amd.com/community...-on-track-amd-ryzen-with-radeon-vega-graphics

Now available: Free NVMe RAID upgrade for AMD X399 chipset!
https://community.amd.com/community...e-free-nvme-raid-upgrade-for-amd-x399-chipset

 

Btw, ignore needing custom psu cables. They will come. This should give a better idea of my build.

 

Considering how powerful these APU's are in both CPU and GPU department, Intel will (hopefully) have a very hard time getting people to buy their options.
On the CPU front, the single threaded performance is about 10.5% better for Intel, but that's only because the single core boosts to 4 GhZ vs 3.8 for 2700u and has about 5% better IPC (so 5.25% for IPC and 5.25% higher boost clock).

However, I'm curious as to why is 8550U scoring so low in multi-threaded tasks vs Ryzen in those charts? It seems to lag by about 45%... which is HUGE.
Both are 4c/8th CPU's... so, what gives?
Ryzen is achieving almost 25% higher clocks on base clocks though... would that be enough to tip the scales so much?
I'm hoping it is.

And if this is tested with single-channel RAM... it will be interesting what Vega can do with dual channel, or better yet faster RAM.

 

Intel's HT has already been shown to be inferior to AMD's SMT implementation. Because of that (and there being likely only one CCX on die, which gets rid of the latency of the inter-CCX IF, a sticking point on the ryzen arch), you are seeing a pure shot for shot here. With so little speed and IPC on single core advantage present, you find a situation where the SMT overtakes the IPC+speed advantage. Also, Intel markets the single core boost. You have to compare all core boost. That is likely less for Intel than AMD. If so, that explains all the numbers.

Now, them saying AMD cannot be paired with Nvidia graphics because of no compatibility with the optimus setting I'm calling bull. It makes no sense and is only a switching that is done. It can be worked out if the parties wanted to, or a hardware mux could be put in for bios config.

 

How about AMD RYZEN 5 1600 6-Core 25% cores less for 30% $$ less for 199$! And it also has a cooler. Tempting so much.
w/ promo code EMCBRCG22, ends 11/1

 

Yeah, I don't buy the lack of compatibility with Nvidia cards statement either... however, we do have evidence suggesting that coupling a Ryzen CPU with an AMD GPU yields better DX12 results.
In this instance, it would be great to see more all AMD options when it comes to laptops.

Pairing the said APU with say a discrete Polaris RX 580 or E9550 might come later... and some vendors might try to pair Nvidia GPU's with it as well... though, considering that the Vega iGP seems to be equivalent to GT 1030... I don't think it's particularly necessary.
Vega iGP is now sucessfully replacing low end discrete GPU's... and with Navi around the corner (Ryzen 2 and 7nm) it should be able to keep up in terms of scalability and just get rid of the low end discrete GPU's segment completely.

 

As presently exists there may be issues yet to be resolved. This does not mean down the line they either can be taken care of or with a discrete GPU have the APU disabled.

 

That would be difficult considering that the APU is a CPU and a GPU all in one chip.. and the last time I checked, a functional CPU is needed for a regular computer to function.
Also, given the degree of inter-connectivity between the CPU and GPU in an AMD APU, would it be possible to just disable the iGP?

AMD uses HUMA technology to combine GPU and CPU for specific workloads, so I'm not sure if disabling just the iGP would be viable or if both components need to be active for the APU to function correctly. Plus, there's the Infinity Fabric to consider.
There seems to be just 1 CCX present in Raven Ridge... but Vega does have/uses Infinity Fabric to a degree already in desktop chips, so I'm not sure how this might reflect on the iGP version of Vega... and we don't know if AMD might use IF to connect the GPU and CPU as well or not.

I think Intel and AMD have a bit different approach when it comes to inter-connectivity between a CPU and IGP... they are done differently.
Switching technologies hadn't had a particularly good track record to be honest, what with games defaulting to the IGP instead of the discrete GPU, etc.

 

Wow, all in all that looks like a dogs dinner, never seen such a spaghetti junction of cables (not just cables though), maybe that's what you meant by custom cables yet to arrive. Mind you my rig doesn't look that great after I've cut out the fan grills hastily and added a cardboard baffle to direct airflow!

 

AMD Announces Ryzen Mobile With Vega Graphics
http://www.guru3d.com/news-story/amd-announces-ryzen-mobile-with-vega-graphics.html

An overview of some of the more interesting slides from the AMD media slide-deck (download).




http://forum.notebookreview.com/thr...ga-polaris-gpus.799348/page-360#post-10622116
http://forum.notebookreview.com/thr...ga-polaris-gpus.799348/page-360#post-10622389

 

Yup.... it looks as if Raven Ridge is using infinity Fabric to connect the CPU to the iGP and everything else.


I don't think disabling the IGP alone would be feasible if you just want to use a dedicated GPU... making it go idle perhaps or support the dGPU for minimum stable/high framerates and taking some pressure off the dGPU.

But yes, I can see how there might be an issue with supporting an Nvidia GPU in there if OEM's decided to stick Raven Ridge with GTX 1060 or 1070.
Honestly, I would just prefer they use RX 580 (or E9550) in MXM factor and just have 1 set of drivers doing everything.


Now if we can get HUMA support inside software... that would be amazing.
Even previous generation of APU's kicked Intel's rear end when using HUMA option... I think we need a widespread adoption of this.
Wasn't there something about Android doing this?

 

Most of those aren't as visible with the case door on, but I'm thinking of getting longer fan extenders and running half down and around the front. Only reason I ran them on the outside instead of the center is to maintain and better access if needed. The second ram fan is suspended by the loop itself. But those aren't the perfect length I need (a little long). Down at the PSU, all the cables are too long except for one running the pumps on the sata power for the BDXL burner. I have 3x480s in there with 16 fans, but the right toward the front are too fat as a bundle to run fully inside the edge. It's a work in progress, but considering the vrm blocks, and the pain dealing with that part of the loop...

When I reseat the cpu block (and this time test it outside of the case), I'll take a picture. Also, get Velcro to mount the water temp sensor. Mind you, that sensor is set to measure water temp after the CPU and VRM blocks, but before the first rad.

I'm more function over form, but do, for the first time, want to have it cleaner. I'll post a pic of my old build in the same case.

 

We already knew the AMD TR4 socket was a bear to screw down, but Jay managed to screw it up - dead. So, be carefuller with TR4 than Jay

There is a lot of speculation that one socket retension version Foxconn vs Lotes (labeled on metal) is better than the other, and others say no, so who knows.

It seems to me it might be related to screwing down each screw completely and torquing it down before moving on in sequence, 1,2,3 - instead I'd catch each and then do a 1/2 or 1/4 turn per screw going in sequence so that each screw gets an even bite and pressure to affix it evenly.

As I was watching Jay I cringed as he torqued down the 1 screw and the other end with the 2 other screws had it's butt in the air... sigh, classic mistake, I would have thought he knew better.

Did a bad TR4 socket kill my AMD 1920X??

 

The old build looks more sensible. I don't know if I'm a fan of RAM fans - I can imagine them contributing to noise & not being necessary if you've got good case airflow in general. Your RAM fans in your current build look too close to the radiator fans, I can imagine them competing each other for air, and I can also imagine those RAM fans recirculating quite a lot of hot air. The fan sitting on top of the PSU & I guess directing air at the motherboard - that doesn't look very efficient - looks starved for air & again looks like all it would do is recirculate hot air. Airflow in the case looks a bit haphazard & innefficient from what I can see - I can only see what's in the pics, but that's how it looks to me. I'd imagine you could get a significantly quieter and cooler build by improving layout & case airflow & perhaps ditching the RAM fans.

 

Actually, using the ram fans and water cooling the VRM dropped the ram heat under load by 15C, whereas the VRM are about the same amount cooler. As to the distance, the old AIO was half as thick as the current rad. The one pointed to the back of the MB is to use the air blown through the third rad toward the back of the MB and isn't to be a major cooling, although I may examine taller offset solutions in the future. The front bottom of the case, not pictured, is a 200mm fan helping to move the air from front to back. Due to the nature of two top facing rads, it is negative pressure, not positive pressure, meaning it is all going to be sucked out without worry of hot spots forming, considering 12 of the 16 120mm fans are venting the top, with 4 120mm in and the 200mm in.

But considering I run my ram around 1.5V, I'd rather have it than not, as this reduces the temps to where it would be off not overclocked. This may be more of an issue after ram VRM are included in the DDR5 Dimms. Also, 8xNoctua iPPC3000s in push/pull versus two corsair ram fans with 2x80mm fans each set (single ram fan unit) with a max of 26CFM. I'm pretty sure, if anything, the ram fans are the one's struggling.

Edit: also, there is open grilling at the back of the case above the psu, so the air is not stagnant there, it gets pushed out through the cross wind.

 

Fair enough, it's hard to visualise all of the airflow just from the pics you showed, and I can picture just some of what you're saying there. Doesn't look particularly airflow optimised to me, but I don't have the rig, so I can't really comment on it's effectiveness or not.

 

I hear you. With 3x480s in it, it is hard to optimize for anyways, especially with the thickness of the rads (55mm). Here are some images to help with the visualization:


So, I have the grid panel on the GPU side with the third rad. There are no fan mount areas for the top door, but that means much of the air is being pulled slightly from the bottom rad, but primarily from the door for the rad above the GPU/PCIe slot side, which that rad is in pull config. The plate that used to separate the two areas the PSU could go in was removed from between them, but the hard drives are hidden on the other side of the PSU at the bottom, which does make that cooling capacity less due to air flow (the third rad doesn't due as much work as the first two). The PSU sucks air from under the case and pushes straight out the back, so not really an issue. The one on top of the PSU is to add some cooling to the back-mounted components to the VRM, which is better than just stagnant air, which it would be less there if not for that fan. As you can see, the bottoms of both sides are ventilated like the side panel, so air also comes in from there, especially with the push/pull being on the side with the glass, attempting to move more air to the side with the CPU components. The front intake fan is meant to help start the backward momentum of the air pulled in from the third rad that is being blown across the bottom of the case, this way to allow it to move toward the back four 120mm fans on Rad1, moving over the ram as it goes. As I said, it is a negative pressure case, so you will have the most movement moving up to expel the heated air out of the top. Does that make a bit more sense?

Edit: Just thought about moving the HDD to on their edge above the rad 3 toward the front. It would allow more air flow from that rad, making the 140mm above the PSU more efficient. Other than that is getting custom PSU cables for the rig, which would improve airflow overall in the bottom of the case.

 

That's exactly how Intel/NV switching works as well. The iGPU is never disabled, just not stressed. You can always run an OpenCL app on the iGPU while playing games on the NV.

On muxless Intel/NV Optimus systems you need the iGPU running for the internal display to work. Even on muxed systems, the Windows shell and whatever graphical application you opened before switching to NV would have their graphical contexts staying on the iGPU. Windows copies the output to the NV for display.

 

When I put mine together the 3 screws were a pain. I had to get them in order to grab 1/4 of a turn. It was a pain but once all three were started I then in the proper order torqued them down. I never checked for the contacts as one torque down was enough and in the end it posted fine.

 

Yup, that is the recommended way, so perhaps Jay's current MB has a faulty mechanism, as that end with screw 2 / 3 is waving around too much in the wind for my comfort

 

AMD Introduces New Ryzen Mobile Processors, the World’s Fastest Processor for Ultrathin Notebooks(1)New Ryzen 7 2700U and Ryzen 5 2500U Mobile Processors with Radeon Vega Graphics Deliver Blazingly

 

I think one of the things that threw me was the horizontally mounted motherboard, which intuitively looks like a poor airflow option to me. As you say it will help airflow a bit once you clean up the cables. I have negative case air pressure too, it was the best cooling configuration for my case too, I tested loads of different fan configurations and that was the best option.

 

I hear you. Reason I chose this case was it was one of the cheaper options for the amount of radiator options available. But, once I started building the loop, it felt so small and like there was very little space for many things I wanted to do. The first rad, above the CPU side, is about 105mm thick with the push/pull config, which really impedes getting my hand in there and seeing it from the side makes it look much messier. I cannot do push/pull on the second or third because of the graphics card inlets sitting so high and because of the space in the bottom. Also, having the place for cables to connect in the middle of the case leaves it for seeing cables go up right there, which isn't ideal if trying to run them to hid them. Also, I moved the filters on the top of the case to the outside (I have cats that jump on the case once in awhile, so need the filter for what their little paws may track up there with them). That increased airflow leaving the case significantly cooler, dropping the idle water temp at the hottest point in the loop by 3-4C. You can now greatly feel the air being pushed out the top, the fans are running slower/quieter, and hopefully the air will deter the cats from getting up there.

So I'll still be playing with it more as time goes on to figure out the optimizations, along with adding DEMCIfilters for the rads just so they don't get clogged and easier to clean. But until I reseat the CPU block, I can't have near as much fun! As I said prior, my 980 Ti Classy doesn't even get passed 33C under load when OCed, meaning it is having no throttling when it eventually hits its breaking point and crashes. Now I just need that on the CPU side!

But this is also my first watercooling build, so headed in the right direction slowly.

 

Yep, learning steps. I've only ever built one PC, the one I own in my sig (air cooled), and I've learned a lot from the process, all-in-all it worked out really well after incremental tweaks: optimising airflow/fan configuration, cutting out exhaust grills, replacing unused PCIe slots near GPU with fine high airlfow filter medium, delidded CPU combined with low RPM quiet CPU fans, Noctua case fans, cardboard baffle to isolate front air intake routing air direct to GPU first, blocking off areas around front intake fans so that only cold air is sucked into case (no reciruculating of hot case air back through intake fans), GPU backplate cooling mod along with liquid metal on the GPU core - it's a quiet & high clocked high performing system as it stands, but that's really a result of all the incremental gains. My next build should be much easier, but that's some time in the future! I'd imagine that if you ever do another water-cooled build you'll be in a better position to choose the 'perfect' case & components as well as the build process & configuring/testing.

 

Definitely. I've built my own PCs since the early 90s (it was a hobby that me and my dad could do together). He didn't teach me soldering, although my first soldering experience was to take apart my walkman after my sister purposely knocked it to the ground and a wire became disconnected. Still don't have the skills he has on soldering, though. He also has me beat on networking skills by a longshot.

But I'm turning the 6700K desktop build in my sig into a router/firewall (reconditioned server quad port NICs with 1Gbps), with that 980 Ti in my TR build. So not a bad firewall, although I need to build that in the next 2-3 weeks before I get to disassemble this rig to reseat the CPU block.

Next time, if money allows, definitely going Caselabs. Much more room with which to work. I am using Thermal Grizzly Kryonaut and TG 8 thermal pads on everything, currently. I know I could do better with LM, which I have two tubes of CLU, but it violates the 1 year warranty from SL on the CPU and with the orientation of the GPU and the temps I am already getting, there is no need for it. But I used it on my laptop GPU and am currently using Indigo Xtreme on the CPU, so...

 

I am happy with my choices so far. With the bios and fans set up it is silent unless under load. Under full load it can get a bit loud but temps are relatively low even with the 4 GHz. Just an awesome system overall.

I have built many a system back in the 1990's and mid 2000's. It has become much easier with time.

 

Wanted to give the pic of the die for the new APUs! You clearly see the single CCX and GPU cores...

 

Hm... I wonder if AMD makes Raven Ridge on 7nm (using Ryzen 2 of course)... will it will include 8c/16th and Navi in a similar power envelope?

 

On 7nm, I think 8c/16th Ryzen 2 and Navi APU might be doable in a same TDP as Raven Ridge here.
Also, the manuf. process will allow much higher clock speeds on Ryzen 2 (5GhZ for the desktop)... so I'd imagine that if a current reduction from desktop format in frequency is preserved (roughly 1 GhZ lower for u series across all cores), then Raven Ridge 2 could be clocked to about 3.2 GhZ baseline, and single core boosting to 3.7/4 GhZ (pretty much like current Ryzen 1700).
The Navi chip clocks might be same or higher in comparison... but given that Navi will have Infinity Fabric connecting multiple GPU chips ... and architectural enhancements... I don't know.
Will be rather interesting to see what 7nm brings all round (hmm... might be fascinating to side-grade in that case and replace the 1700 with say low power Raven Ridge 2 with Navi - or 35W version of both... uuh... that would be better).

I can only speculate on what Asus needs to do in terms of coding to make Raven Ridge compatible... but if they pull it off (and probably will)... YOWZA!!

 

Here are some benchmark results of my Asus ROG Strix GL702ZC:



Seems to get similar score as mobile GTX 1060 3GB in the Graphics Score.

 

No, more like 18-20% lower graphics score vs. Mobile 1060 3GB.

 

Source?
I keep seeing barely higher scores than mine for mobile 1060
Vs desktop, yes I would agree there would be about 20% difference... but not for the mobile

 

why not trust my words?

Edit. Oh’well, I’m sure the Msibook on NBC should run stock clocks.

 

Because I like to corroborate information with actual evidence and I was comparing mobile to mobile. Not mobile to desktop.
For example, someone tested their mobile 1060 (3GB) with i7 6700k and these were their results:


I also found multiple GTX 1060 (desktop version) that scored just above 36 000 in graphical score in SkyDive... so practically 10% difference.

 

See my Edit 40.000 in Graphics score stock should be within reach. With stock clocks. Oc’d more like +-44.000

 

And preferably the SkyDive benchmark itself?
Where is the said benchmark for that laptop?
I can't see it on that Notebookcheck website.
And if the score is accurate, was it done under 720p resolution or 1080p?
Mine is under 1080p (native) res.
I find is 'suspicious' that Notebookcheck is getting gpu scores within a few % of 1060 and surpassing 1060 in TimeSpy... and I'd be getting what... 20% lower scores?
Something isn't right... though, I could be getting different scores since I'm not using drivers specifically tailored to my GPU or laptop in general ... I'm using desktop drivers and I lost FreeSync as a result.
Asus still hadn't put up the original or new gpu drivers for this laptop so I can't switch.

 

Btw... here's my TimeSpy score.
I'm scoring slightly higher than Notebookcheck scored for both the 1060 and RX 580 laptops they tested:

 

I'm sure they use 1080p. Why should they change between the different benchmarks as 1080p is the goal? See pict 5


See also the same from Acer Predator Helios 300 (7700HQ, GTX 1060, Full HD)


For more info, take a look into 3dmark.com/1060 (Notebook)

 

From same reviews as above. And NBC haven't optimized tested systems as the owners probably do. + the tested machines contain also the low end Core i7 BGA Filth. Only i3-i5 BGA is worse

MSIBOOK


ACERBOOK

 

Off topic but anyone watching AMD stock prices is getting an eyefull as it has taken quite a beating, going from bullish to now bearish. If AMD keeps this up it ccould be in major trouble as the prices have taken quite a beating and AMD seems to be doing nothing about it either.

 

I haven't heard of any functional reasons for the drop, it's likely related to investor protocols.

It could be long time holders wanting to exchange positions, it should be over now that the prices have dropped.

 

It would be nice, but there seems to be no floor. I think the problem is with the 7980xe and Coffee Lake AMD did not respond timely with even lower prices of the Ryzen/TR chips to combat them and now the maket does not see sustainability. Once the prices went over to the Bearish side the maket just further collapsed for them. I think they were hoping the mobile anouncement would sustain them for a bit but that just is no where near enough.

People ove4rlook at with this quarter how much debt they've overcome with the sell offs andn warents issued fron Dubai etc. But things will be as they are.

 

I don't think investors have that level of sensitivity, or move that quickly due to such estimations, besides the play on the Epyc Datacenter is in play, as are ThreadRipper, Ryzen 7,5,3, and Ryzen mobile.

AMD is profitable 2 Q's in a row, for the first time in a long time, those leaving their positions will be replaced by those looking to profit in the short and long term, hopefully AMD will continue to be profitable and the stock will recover over time.

 

Something happened on the beginning of 25-th October. Could be huge investor decided to cash out on the base of cofee lake competition and then collapse occured on bad expectation.

 

Agreed and they tore the stock up. Right from Bullish to bearish. Hurt quite a few long term people I am sure, myself included.

This may not be helping either;
https://www.cpubenchmark.net/market_share.html

 

RX Vega Aftermarket Cooler = Success?? Alphacool Eiswolf 120 GPX Pro ATI RX Vega M01

Wolfenstein II | Rx Vega 64 LC vs 1080 TI vs 1080 | 1440p Mein Leben Benchmark Comparison | 7700k

 

It looks like newegg has a bunch of RX Vega 56 / 64 units in stock...

https://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100007709 601301447 601302833&IsNodeId=1&bop=And&Order=PRICE&PageSize=36

SAPPHIRE Radeon RX VEGA 64 DirectX 12 100410L 8GB 2048-Bit HBM2 HDMI / TRIPLE DP (UEFI) PCI Express 3.0 CrossFireX Support ATX Video Card - Including both Black/Silver RX 64 GPU's @ $499
https://www.newegg.com/Product/Product.aspx?item=N82E16814202300

PowerColor Radeon RX Vega 56 DirectX 12 AXRX VEGA 56 8GBHBM2-3DH 8GB 2048-Bit HBM2 PCI Express 3.0 CrossFireX Support ATX Video Card @ $419
https://www.newegg.com/Product/Product.aspx?Item=N82E16814131730

XFX Radeon RX Vega 56 RX-VEGMLBFX6 8GB 2048-Bit HBM2 PCI Express 3.0 CrossFireX Support Video Card - Black Design @ $419
https://www.newegg.com/Product/Product.aspx?Item=N82E16814150810

 

AMD => Where to Buy Ryzen Pro Commercial Systems
http://www.amd.com/en/where-to-buy/commercial-systems
Ryzen Pro
http://www.amd.com/en/ryzen-pro

 

Nvidia's 1070ti Reference and AIB cards are all shipped at the same clocks, no advantage out of the box for AIB cards other than better cooling to hold those clocks, but the user can OC the same (and more, see video) as any other OC'able Nvidia GPU

NVIDIA GTX 1070Ti - Can it REALLY be overclocked?


ccmaster86 2 hours ago
Consumer: "Nvidia why did you launch another 10 series product just to beat Vega 56 instead of just dropping prices?" Nvidia: "Because "fargo" you that's why." Consumer: "Ok cool. Here's all our money." Nvidia: "That's right "female dogs"."

GTX 1070 Ti Review.. Oh and Also WTF NVIDIA??

Nvidia GTX 1070 Ti Founders Edition Review

Nvidia GTX 1070 Ti Founders Edition Review
https://overclock3d.net/reviews/gpu_displays/nvidia_gtx_1070_ti_founders_edition_review/1

 

So, updated benchmarks at [email protected]


@hmscott @Robbo99999 @Papusan @TANWare @Deks @Rage Set

Almost got you score TANWare at 50MHz slower. Just finished optimizing my ram (TR is picky as hell, and just going to the last stable timings doesn't always work). But I found out adding 100MHz only increased the heat output by 2C in these benches. I went to 1.25V, going up 25mV from 3.9, just because I didn't want to fuss with intermediate instability. Might go for 4GHz+ later today. But [email protected] is pretty good! Also, my ram running 3600@CL14 is SWEET!!!! I'm guessing that is where the extra performance over the 50MHz difference came from (and using a single purpose install of Win 10 Ent. FCU).