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

http://m.hexus.net/tech/reviews/graphics/109078-asus-radeon-rx-vega-64-strix-gaming/

 

An additional 5% performance improvement in games over the reference version of the card. Then by overclocking that card they "added an additional 6.4 per cent performance in a trio of games. The downside of this voltage-induced performance is that power consumption rose to 532W at the wall."! So, that's 1.05 x 1.064 = 1.117, which is 11.7% more performance in games when overclocked in comparison to a stock reference Vega64, MEH really, especially for a 532W power consumption!? Although that is power consumption at the wall for the whole system, and to put it in perspective the stock Vega64 power draw total system wattage was 411W, so 121W extra over stock Vega64 for 11.7% more gaming performance. Stock Vega64 graphic card wattage is 334W (measured by Guru3D), so add the extra 121W, then you're talking that this card when overclocked is using about 450W. I think it's safe to say the limits have been reached! (GTX 1080ti = 280W as measured by Guru3D)

 

So...expect rolling brown-outs when all those Vega mining rigs go online?

 

The power draw is horrendous. It does not seem to be wasted energy as the thermals are not all that bad. Maybe HBM2 is not as good with power as advertised?

 

ASUS Radeon RX Strix Vega 64 preview: faster and quieter especially
https://nl.hardware.info/reviews/7525/asus-radeon-strix-rx-vega-64-preview-sneller-en-vooral-stiller

ASUS ROG STRIX Radeon RX Vega 64 finally gets tested
https://videocardz.com/72128/asus-rog-strix-radeon-rx-vega-64-finally-gets-tested

Asus Strix Vega - The Roggy Vloggy

 

It's not HBM2. The GP100 doesn't consume that much power (still more than GDDR5X). I've been thinking to myself. I got 1200W to 1500W PSUs in my rigs but never did I once think I would actually need that much power unless I was running three or four GPUs. A single OC'ed GPU requiring 350W+ is shocking.

 

We know that stock Vega clocks have way too high voltage to begin with (this is to improve yields).
I wonder what would have happened if Asus released Vega at stock clocks but properly undervolted and possibly overclocked the HBM (which according to GamersNexus improved performance a lot more than overclocking the core did).

HBM by itself should cut power consumption on the VRAM by 60%... and something has been amiss here to see such high power draw on a GPU. We saw that Fury cards can be undervolted and their power consumption drops within about 10% of Nvidia GPU's while improving performance (elimination of thermal throttling).

So, why is AMD still having issues with power draw?
The architecture is new, and we don't know how developer patches could affect performance along with new drivers on Vega (which does have multiple new instruction sets that devs need to optimize for).

Is it the compute hardware?
Is the HBM not working properly?
Could it be that there's simply too much voltage being assigned also to the HBM independently?
We know that Wattman has a separate section for VRAM voltage and frequency control. So, if AMD unlocks those, could people try undervolting the VRAM and see if it helps there too?
Or AMD's design simply could be inferior when it comes to power consumption and even if undervolting is applied, power consumption could drop much closer to nvidia levels, but not necessarily be on par.

 

We'll see when the manufacturer versions start shipping. If I had to guess I'd say your suspicion about the voltage to the HBM is correct, but I'm going to wait and see.

 

AMD Responds To Radeon RX Vega Pricing Outrage, Confirms More $499 Vega 64 Cards Available Soon
https://hothardware.com/news/amd-responds-radeon-rx-vega-pricing-more-499-vega-64-available-soon

"Earlier this week, we were a bit disappointed when we heard rumors that AMD would be dropping its standalone air-cooled Radeon RX Vega 64 SKU, effectively raising its price to $599 for the foreseeable future. If this was indeed the case, it would be the ultimate case of “bait and switch”, since AMD played up the attractive $499 price point for the Vega 64 and reviews recommending the graphics card (including ours) were based on this price point.

Rather than run with the story, which was originally posted by OC3D on Tuesday, we decided to reach out to AMD for further clarification. It took a few days, but the company finally contacted us with an official statement on the matter:

"Radeon RX Vega 64 demand continues to exceed expectations. AMD is working closely with its partners to address this demand. Our initial launch quantities included standalone Radeon RX Vega 64 at SEP of $499, Radeon RX Vega 64 Black Packs at SEP of $599, and Radeon RX Vega 64 Aqua Packs at SEP of $699. We are working with our partners to restock all SKUs of Radeon RX Vega 64 including the standalone cards and Gamer Packs over the next few weeks, and you should expect quantities of Vega to start arriving in the coming days."

Hopefully, this will clear up some of the confusion that has proliferated over the past few days. Yes, Vega 64 is undoubtedly in short supply, and retailers will most certainly try to “adjust” their pricing to take advantage of the incredible initial demand. However, this is not some grand scheme by AMD on its end to kill off the standalone Vega 64 and push for higher prices. Rather, it’s simply a matter of not being able to meet launch demand, which can happen with any new high-profile hardware launch."

AMD issues official statement regarding Radeon RX Vega 64 pricing
https://videocardz.com/72123/amd-issues-official-statement-regarding-radeon-rx-vega-64-pricing

AMD Responds to Radeon RX Vega 64 Stock Shortage and Rumoured Price Increase
http://www.game-debate.com/news/235...64-stock-shortage-and-rumoured-price-increase

RX Vega 64 Custom Card Benchmarks Surface | AMD Speak On Vega Price | 7960X Vs 1950X Benches

 

RX Vega 56 Hybrid & (Half-Functional) 242% Power Offset Pt2

 

I think the high power draw is associated with the core rather than the HBM, that's my hunch, mainly because HBM is supposed to be so power efficient in comparison to GDDR5, plus historically the vast majority of power draw in a card is associated with the core & not the VRAM. I don't see them improving power efficiency unless they start to undervolt their GPU cores, but AMD are using the current voltage level for a reason - because they think they need that much for stability, they're not gonna be wrong. Not saying people won't be able to undervolt their GPUs, just that on average they're choosing a voltage that is gonna be stable across each & every one of their individual GPUs they sell. I don't think we're gonna start seeing increased power efficiency until they bring about improvements to the silicon manufacturing process which might allow them to reduce voltage.

 

Ryzen 3, Overclocked Gaming Benchmark Guide!

 

Buildzoid tested them. When overclocked, they still drew less than 20W, way less than Nvidia's vram. So, it's not that.

 

In which case, we are looking at two culprits behind larger power draw:
1. Higher than necessary voltage to improve yields on reference gpu's.
2. Manufacturing process not optimized for high clocks but rather for lower clocks (hence why power consumption is so high).

Leading to this: WHY is AMD sticking with a global foundaries manuf. process that's incompatible with higher clocks? And why do both AMD and OEM's then go ahead and push the clocks to such high levels where power efficiency is wrecked?

Let's say that Vega was made on a manuf. process similar to that used by Nvidia... and assuming the voltages were brought under control (optimized from the get go), would we see higher clocked Vega at much lower power draw?

 

You are conflating aspects of an architecture! The IPC is higher, which is why, so long as optimized, Vega can reach the performance of Pascal using less clocks. So yes, they focused on IPC over higher clocks. Because of this, you cannot say something as simple as it being incompatible with higher clocks. HIGHER CLOCKS DO NOT ALWAYS GUARANTEE BETTER PERFORMANCE. Look at Intel during the Athlon XP timeframe and look at AMD FX chips. Stop with that falsehood please.

As to why push high clock levels, they have to to get a competitive product and recover the expense.

As to your question, GloFo has licensed and is using Samsung's process at 14nm at the moment. Samsung's is as energy efficient as TSMC. So it likely is not the process at issue. But, TSMC does have an energy efficient process.

 

Actually, I'm aware that IPC is higher, and that Vega has new instruction sets (which of course won't matter one iota unless developers get off their rear ends and optimize games and software for Vega in the first place).

My point was that given the current manuf. process which AMD uses... it is obvious that it is more suitable for lower clock speeds in order to keep power draw in check, so I don't see the point in clocking Vega higher on core for marginal increases in performance while radically increasing power draw (which only serves as a deterrent to less informed members of the public (whereas HBM oc-ing showed to be apparently more suited to that).
Plus, let's not forget high voltages on the core which AMD needs to find a way to address during production

 

Oh dear, a somewhat damning article from Toms Hardware about Vega64 watercooling & overclocking: http://www.tomshardware.co.uk/radeon-rx-vega-64-water-cooling,review-33986-4.html

Their conclusions, no point overclocking Vega64 (and especially not with AMD's own watercooled version which was deemed inadequate), instead they say/hint to buy a completely different card (as in not Vega64) or keep your old one!

 

IPC is higher because Vega is a bigger and more complex chip than GP104. Many more ALUs, transistors, etc.

The culprit behind Vega's astronomical power draw should be obvious--AMD threw power efficiency out the window in order to ramp up clock speed as high as they could. Having Pascal-esque clock speed was the only way they could compete with Pascal, because as AnandTech noted, Vega's architecture is remarkably similar to Fiji's, just on a smaller process with some minor feature tweaks. GCN's compute-heavy design (lots of ALUs, big memory bus and BW) isn't suited for high clock speeds, we've seen this on every iteration of GCN vs. Kepler Maxwell and Pascal that it simply can't clock as high. AMD has been stripping some non-gaming compute features out of their Radeons over the last few GCN revisions, but it's still not enough to match Nvidia's efficiency.

And before you belabor undervolting yet again, have you stopped to consider that maybe there is baseline voltage needed to guarantee Vega's current clock speeds across chips of different quality, and that finding an undervolt that would work across all chips would require further testing and validation, which would thus incur further cost and delay on AMD's part for a product that is already 1+ years late and barely price competitive with Nvidia? As we speak, AMD is probably already doing so for Vega Nano. Also 14nm isn't new anymore, so further improvements from an optimized manufacturing process are probably limited.

 

The Witcher 3 AMD RX Vega 64 Stock Vs GTX 980 TI OC Frame Rate Comparison


RX VEGA 64 Ethereum mining Hashrate with AMD Crimson Blockchain Compute Driver | GPU Power usage

 

It is a better miner than the 1080TI but not a better gamer. too bad it was not better on power because then the miners might leave the other cards alone on the market for us.

 

MSI X399 Gaming Pro Carbon AC: Threadripper is Here!

GPU Passthrough for Virtualization with Ryzen: Now Working

 

AMD’s Official Statement On RX Vega Pricing Issue – Side Steps Around The Problem

 

That report again is 8/15. Further information seems to have come to light since then.

 

I posted due the clock time it was posted. If it's old news, oh'well

 

When I made my earlier statements, it was without knowing they did a Point of Sale (POS) rebate (covered by LTT in the WAN show). So, they literally had those rebates end a week after release.

 

But that rebate ending rumor predates the more recent announcement from AMD that they are keeping the individual card prices and pack prices the same, so LTT's info is out of date - I thought it was an embarrassment that LTT went with old news, and left it posted.

As far as AMD has said themselves, the only source that matters, the release prices and sku's remain the same, and they will be restocking all sku's.

So there's no deception and no bait and switch by AMD, just a lot of clickbait noise from hungry media outlets

 

If you go to the link he uses as the material to base this on, which is his own post, it is dated 8/15. I would not trust this guy at all.

http://wccftech.com/amds-rx-vega-64s-499-price-tag-was-a-launch-only-introductory-offer/

 

Searching for a case that will let me expand my water cooling while keeping the appearance of wires to a minimal is difficult. I really do like the TT Core W100, W200, and the X9. I am leaning towards the W100, especially since my "local" (80+ miles away) Microcenter is selling it for $179.99 whereas every retailer is selling it for $289.99 or higher.

 

Now available;
https://www.newegg.com/Product/Prod...H&Description=enermax+aio+tr4&N=-1&isNodeId=1

 

I have the core X9. Easy to fit 3x480s in it. As to cables, get a modular psu and you will need to get clips and do clean runs for the best look, but not a bad case for so cheap. If looking at the others, also consider caselabs offerings. More expensive but extremely customizable.

Aio is sold out.

 

Enermax TR4 AIO is sold out at NewEgg already.

 

Rage Red - Threadripper
1950X -
Asus Zenith Extreme X399
EK Custom CPU Loop - Red
32GB of TridentZ RGB
1080 SLI - Depending on the card(s) I finally settle on, it will be added to the loop.
TT Core X9
White/Red LED case fans.

The EK Threadripper CPU block is on backorder and I won't receive that until the end of August it seems. I may use an AIO in the meantime. Going to be a fun build.

 

https://segmentnext.com/2017/08/21/amd-raven-ridge-apu-confirmed/

 

Too bad it is so late. Would have been nice to be out for the back to school shopping season.

 

https://www.hardwareluxx.de/index.p...vega-56-und-vega-64-im-undervolting-test.html
Feels like Polaris all over again.

 

If you've got the room for it, the wider W200 should give you a lot more options for coolant loops and cable routing, that would be my choice if you're looking to hide components.

And I'm a little envious, sometimes you just want to walk into a brick and mortar place for parts, and I'd love to have a Micro Center that close to me. (Closest currently is 200+miles from me).

 

I have to drive a couple hours each way to go. Shame they no longer have a train between here and there, or I'd do that. We are/were an area considered for the Musk hyper loop. I'd do that in a heart beat if they put one in.

 

I wouldn't be so sore about it except all the local B&M electronics places have turned into glorified phone stores.

 

Hyperloop = HyperSplat?

Can our bodies handle the hyperloop?
http://www.pbs.org/newshour/updates/is-traveling-on-hyperloop-a-ticket-to-puke-city/

Thunderf00t Critical Video's
https://www.youtube.com/user/Thunderf00t/search?query=Hyperloop

I really love Elon's ability to fund his dreams, all very expensive to government and individuals, very risky, but it certainly would be a "boring" time without them, very entertaining overall.

Report: Tesla is bleeding talent from its Autopilot division
http://forum.notebookreview.com/thr...ng-talent-from-its-autopilot-division.808387/

 

AMD RX Vega 56 and 64 Review - Massive Rant

 

Vega56 actually performing astoundingly with an undervolt & overclock combined - really good, beating the GTX 1080 founders with lower power consumption than the GTX 1080 too! Problem is they describe the undervolting process as very complicated, and it's a bugged process at the moment (more specifics in the article). I suppose we also have to realise that every single Vega56 is not gonna be able to undervolt & overclock like the one in that review, could be they got very lucky with the silicon lottery! If you're lucky though, and you put the work in to undervolt & overclock the Vega56 it actually seems like it's gonna be a card worth getting as long as prices don't become inflated massively beyond MRSP.

 

I hear you. I love Microcenter. It is bigger and better than our old computer parts chain, CompUSA. The best thing is that they can actually compete on price with etailers like Newegg/Amazon. The price of that computer case is just an example. However, sometimes after a long week of work, I don't feel like traveling the 81 miles to go and "visit". My gf doesn't like me going, I end up spending a lot. Hahaha.

 

Piece by piece, arriving in the mail. The EK X360 extreme kit will be here on Friday, along with the PSU, RAM and a crap load of fans.

 

In my case I get around the lady's reservations about making that trip by reminding her that there's an IKEA there too. We just make a day of it.

 

Even with the small risk, it's a relatively straight shot, although mountains are mountains. Other than that, not really concerned and I'd use that regularly.

 

Why do people assume the worst when talking about technology they know little or nothing about?
Obviously, the human passengers would be well insulated to not even feel that kind of acceleration in the hyperloop.
Also, the hyperloop concept is outdated.
Jacque Fresco proposed a far superior system that we could have built back in 1974 on interview with Larry King:

And then there's ET3 (on which Hyperloop is actually based on) that's far superior :


Technologically, it's been doable for decades... the only thing preventing us from actually implementing all of this in practice is the outdated socio-economic system which always worries about how much things 'cost' - whereas the real question should be : 'do we have the resources to make it happen with the least possible impact to the environment in a short time frame?'
Answer: yes... resources can be found on landfills that can we can easily convert into usable materials of massive amounts (with most of it that can go back into the Earth in most usable form to repair the damage we caused), and construction can be carried out by automation (mobile 3d printers with drones, robots, etc.).
People focusing on manual labor in this day and age (or actually over the past century) are living in the dark ages when it comes to technological understanding, but that's understandable because they live in a system that doesn't inform them properly on scientific and technological understandings.

Elon Musk is hardly a daydreamer... his main issue is he's not really going far enough... heck, most of his proposals are technologically outdated compared to what can be done, and people STILL think of it as impossible (mainly because they have not studied technological advances in multiple fields throughout history to know that many systems already exist which can be brought together to create a better one - instead they focus on specializing in a select field of choice).

Alas, how did the thread go into discussing Hyperloop?
I only checked it 2 days ago and it wasn't there yet.

 

Spoiler: OT

@ajc9988

Here you go, "Trans-Atlantic Tunnel", SciFi movie from 1935:


And yes, I've watched it

I think the first serious tunnel I read about was from 1802, in 1802, French mining engineer Albert Mathieu- Favier put forward the first ever design for a cross-Channel fixed link based on the principle of a bored two-level tunnel: the top one, paved and lit by oil lamps, to be used by horse-drawn stagecoaches; the bottom one would be used for groundwater flows.

And, I'm sure they all would have specified AMD hardware for all their computing needs

Threadripper Large vs. Small Coldplate A/B Test (Noctua NH-U14S)

 

Who is crazy enough to wants to buy Alienware desktop? Even with Ryzen, you have no guarantee that Dell will release new firmware for the next Ryzen CPU. Aw laptops is bad, but damn... Desktop from Dellienware is on same level. Bro @Cass-Olé have had bad experience with ALIENWARE desktops as well. Stay away!!

 

His mention of the Enermax cooler brings up the topic I commented about six pages back. Does these new "TR 100% coverage" CPU block designs for Enermax and even the EK Threadripper block, cool better than an AIO designed for smaller chips? As far as I can tell, none of these companies have expanded the fins in the pump, only the cold plate.

I am going to test the results of an AIO water cooler versus my custom water cooler setup with the EK Threadripper block. I am not going to be happy if my custom loop is within 5 degrees of an AIO. On a positive note, I still have over a week before my EK TR4 block officially ships and I hope someone that has one, posts their results.

 

To be fair to Dell, you should already know what you're getting into when you buy a laptop or desktop from Dell. It is no secret that Dell uses proprietary mobos, PSU, power adaptors and more in their products. That is one of the many ways they can sell some computers with "higher end" components at such cheap prices compared to boutique shops.

I am condoning what they do and the products they sell, but enthusiasts like us, for the most part design and build our own desktops. It has been a long time since Dell/Alienware has cared what we like and/or want in a product they produce, since we were never their intended customer demographic in the first place.