AMD Vega MegaThread! FAQ and Resources

Page 16 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
RX Vega will have 3 different variants. We really don't know much. Frontier priced at $999 vs the Titan? I guess would be the comparison. It's not looking good right now, so it's hard to say what the price points might be. The scarce and expensive HBM2 being a factor... Doesn't help the situation.
 
To be fair, Nvidia has always held a similar pricing structure. A new model gets released, then the ti version gets released at a similar price and the rest of the stack moves down. But the 780 ti was released at $700:

http://www.anandtech.com/show/7465/nvidia-announces-geforce-gtx-780770-price-cuts-gtx-780-ti-launch-date

Right as the R9 290X came out. And they dropped the 780 by $150. When the 980 ti came out, it was priced at $650, and the 980 only dropped from $550 to $500 at the time. AMD's response was the much less competitive (compared to the 290x) R9 Fury X. And then of course the Pascal line which we've been talking about, which has already seen significant price cuts while Nvidia has basically been alone at the top end.

Vega's actual gaming performance remains to be seen, as well as exactly how much HBM2 will affect costs.
 
Then it will be a re-hash of Fury. At least, all points to that.

It will be faster than the 1080, albeit not by much and it won't be able to surpass the 1080ti in any relevant metric.

DOA never seemed more fitting.

Cheers!
 
Well if Vega actually launches now the timing couldn't be much better. If you can make a graphics card now you can sell it and there is profit to be made.

And gamers won't be happy with what's on the market and how much it costs for a few months anyway.
 


I have to admit varis you have a point that I foolishly did not consider. Crypto miners will be all over these cards. The sooner the better while there is high demand. Those cards will be selling like hot cakes.
 


Not Frontier, RX Vega I think could have a good chance if rumors are true and the highest ~15% more performance than the 1080. Say $499-550, $399-450 being around 1070 performance, and the 3rd being better than 1060/550 $299-350. GNC raw horse power is okay for the miners if performance and power consumption scale well. And just for the lack of any other available cards in those price ranges. It has a chance.
 


In terms of raw power the 480 was a 5.8 peak Tflop card, while the 1060 6 GB was only at 4.4. But it took a long time for AMD's drivers to unlock that power and they still trade wins depending on the game architecture. AMD has always had that problem, getting the software to take full advantage of their hardware. Mining is much simpler to code, so miners love raw horsepower. Plus it was cheaper than the 1060, probably based on that early gaming performance.

Vega FE is a 13.1 Tflop card, so if the consumer edition matches that then anything below $600 would be very attractive if the ETH craze hasn't finally ended by then.
 
Ethereum Gold Rush Wanes As Price Plummets, Mining Difficulty Skyrockets
by Kevin Carbotte July 12, 2017 at 5:00 AM
"It appears the recent gold rush surrounding Ethereum is coming to an end. The price of Ether dipped below $200, and the mining difficulty level jumped by nearly 20%. If you’re getting into Ethereum mining now, it’s probably too late."
http://www.tomshardware.com/news/ethereum-price-drop-difficulty-increase,34985.html
 
24llu8j.png

Etherum's price is on it's way back up. If it continues to climb this might keep Ethereum mining going.
 


i'm not talking about raw performance. no matter how big the performance on Vega the card was equipped with HBM and not GDDR5. from what i heard eth algo only optimized for GDDR5 memory. that's why some people willing to get used 1070 for $500 when you can get much faster brand new 1080 at the same price because the algo is not optimized for GDDR5X.
 


The Fury X with HBM works. I would think HBM2 would work as well.

edit: "One of the key enhancements of HBM2 is its Pseudo Channel mode, which divides a channel into two individual sub-channels of 64 bit I/O each, providing 128-bit prefetch per memory read and write access for each one. Pseudo channels operate at the same clock-rate, they share row and column command bus as well as CK and CKE inputs. However, they have separated banks, they decode and execute commands individually. SK Hynix says that the Pseudo Channel mode optimizes memory accesses and lowers latency, which results in higher effective bandwidth." Lower latency would be to the plus side for mining. http://www.anandtech.com/show/9969/jedec-publishes-hbm2-specification
 


they work but not as optimized as GDDR5 card. when it comes to mining the "mining efficiency" also have to be taken into account. pure miner will not going to take card that will take too long to repay itself.
 
AMD RX VEGA XTX, VEGA XT and VEGA XL Variants Confirmed – Add-in-Board Partner’s Custom RX Vegas Ready For Mass Production By Late August
By Usman Pirzada
Jul 12
"Videocardz.com has confirmed details about the various variants of RX Vega. The leak was originally posted on 3DCenter but was confirmed by VCZ later on. At the same time, we also have the release schedule for RX Vega custom variants from 3DCenter forums and it looks like everything is on track for the RX Vega launch on SIGGRAPH Capsaicin on July 30, 2017.
Vega XTX is apparently the Radeon Frontier Edition GPU while as Vega XT will be the gaming centric chip. The only major difference between the two will be TDP differences.

Vega XTX will be a water cooled GPU with 375W of Thermal Board Power while as the GPU itself draws around 300W. The water cooler implies a very fast clock rate which means that we just might be looking at clock rates near or above the 1700-1800 MHz mark. This is something that is also implied by the TDP rating of this particular chip.

Vega XT is going to be a similarly configured chip that will be running an air cooler. The Thermal Board Power of the Vega XT is only 285 Watts with the GPU drawing around 220 Watts. This is quite a big difference in TDP considering the core count is the same so this means we are probably looking at a low clocked Vega GPU over here (probably the one we have seen in demos multiple times before).

Vega XL is apparently a custom only chip that will have a cut down Vega core. While Vega has 4096 SPs, Vega XL will only have 3584 cores. However, the TBP and GPU power draw remains the same. Vega XL will feature an air cooler and will be “AIB-Exclusive”. This means that we are not going to be seeing any reference or ‘Frontier’ cards for this particular chip and will be manufactured directly by AIBs (probably because AMD itself will not be able to handle the large demand of a mainstream chip).
We also have the release schedule of the custom AIB RX Vega variants. From the looks of it, the AIBs are awaiting word from AMD to set a mass production date. Bios and other details will all be taken care of by early August, so mass production can start from any point onwards. Depending on when AMD announces general availability of the GPU on SIGGRAPH later this month, it could be as soon as early September."
http://wccftech.com/amd-rx-vega-xtx-vega-xt-vega-xl-variants-confirmed-add-board-partners-custom-rx-vegas-ready-mass-production-late-august/


 


I agree with you on that! I watched a video yesterday, I should have posted it, describing the power demands reaching limitations of the PCI-E. I'll find it and post it.
"Vega is limited on power, not on thermals. There’s some cross-over in power leakage, clearly, but after we solve for the thermal limitation, we bump against the power limit of about 400W at the PCIe cables (drawing ~33A with overclocks). Our next goal is to increase the power limit in BIOS. We might try this with a Raspberry Pi mod, as ATIFlash doesn’t yet work with Vega: FE."
 
Vega FE Hybrid: Fixing PWR Leakage & OC Boost
Gamers Nexus
Published on Jul 13, 2017
Our results & review video for the Vega: Frontier Edition water cooling mod using our hybrid kit, showing power leakage, thermals, noise levels, FPS, and overclocking.
https://www.youtube.com/watch?v=WbS7c2Een8o

AMD Vega Hybrid Mod: Fixing Power Leakage for Higher OC
By Steve Burke Published July 13, 2017 at 5:12 pm
0120feb98838d333fb173acb9f52435c_XL.jpg

Vega: Frontier Edition Overclocking in SPECviewperf
"Getting into benchmark numbers, this next chart shows the SPECviewperf performance improvement in various workloads with our stock Hybrid and overclocked Hybrid mods. Again, the point isn’t comparison against the Titan Xp – the point is just to look at Vega air vs. water.

3DSMax posts a performance improvement of about 6% when overclocked with the hybrid mod, with equal performance for the stock air and hybrid Vega FE cards. Catia posts an 11% performance gain from the overclock and liquid, with Creo at a 5% gain, followed by the Energy test at a 15% gain, and then the Maya test with a 6.8% gain. The medical benchmark is responsive to the overclock, posting nearly a 19% improvement, with Showcase at 13%, SNX-02 at 9.5%, and SW-03 at an 8% gain from the overclock."
vega-hybrid-spec-results.png

"Only a few of these are particularly exciting, with the rest feeling more like a lot of effort for not a lot of gains. What this does do, though, is set the stage for potential growth from a 100MHz clock boost. The impact on gaming and synthetic tests is more pronounced, despite this card clearly not being gaming-ready at launch – and again, that’s indicative of a rushed launch on AMD’s part. The drivers were not ready, features are rumored to be left disabled, and yet the card was shipped. Even still, despite this being “not a gaming card” and having shipped with incomplete features, we can observe clock scaling headroom in an A/B test versus the air-cooled Vega: FE card ($1000).

We’re not yet talking RX Vega performance and suggest that you refrain from attempting to speculate or extract numbers; just wait for our review to see how RX Vega does in these games."

Vega Hybrid Mod – 3DMark
"With the 1080p FireStrike test, our graphics score on the Hybrid mod posted an overclocked score of 24,998, versus the 21,355 score of the stock air-cooled card – that’s a gain of 17% with the overclocked Hybrid card. With the overclocked air card, we were at 24,554, or a gain of 15% over the stock card. The boost strictly from our extra 40MHz on liquid isn’t tremendous – only a couple percent – but again, this is about thermal and noise performance more than anything."
1080p
vega-hybrid-firestrike-normal.png

Here’s a look at 1440p (Extreme) scaling:
vega-hybrid-firestrike-extreme.png

And 4K (Ultra):
http://media.gamersnexus.net/images/media/2017/GPUs/vega/hybrid-fe/vega-hybrid-firestrike-ultra.png
FireStrike Ultra has our overclocked versus stock air gain at around 17.7%, once again, providing some consistency between 4K and 1080p.
And here’s TimeSpy, just to round it out:
vega-hybrid-timespy.png


Conclusion: Vega Limited by Cooling, then Max Power
"For anyone who skipped the thermals & power page because it’s less “exciting” than gaming, we’d advise going back and reading that page prior to commenting or reading the conclusion. That’s really the bulk of this coverage, and is far more important than the gaming benchmarks; that said, the gaming benchmarks do provide some insight as to clock scalability of Vega in different applications.

This card still should not be bought for gaming, and the card and its drivers still should get polish from AMD before it’s bought in general, really, but the gaming results are interesting. The clock behavior is the most deserving of attention, just be careful of how far you take those numbers, and remember to wait for RX Vega-specific numbers rather than speculate.

All that cleared away, the power leakage reduction under liquid was non-trivial, and granted us our additional ~40MHz of OC headroom on the core (by way of granting ~30W more to the core). Under max consumption and overclock, we measured current at the PSU cables to be ~33.3A at 12V, or just under 400W (+/-2% error). Given that there’s also power going through the PCIe slot, this is one of the most power-hungry overclocks we’ve managed without modding BIOS to increase the power limit. The gains in thermals were tremendous, ultimately affording the reduction in power leakage (and overall increased power available to the core).

Noise is also down, falling to ~43dBA without the VRM direct airflow, or ~50dBA with it (as opposed to 90% GPU fan speeds with our air-cooled overclock). Performance gains range from 6% to 27%, depending on application and how the GPU is used. We noted that HBM2 overclocks, like with the Fury cards, offer an important boost to performance.

Uplift looks good in some titles, but is also at the expense of major power consumption and a lot of effort. As brackets become available for these new cards, it’ll be easier to justify the mod. We’re curious as to what AMD’s liquid-cooled variant offers beyond the CLC, given its ~+$600 price hike. Kind of steep.

Regardless, the big take-away is clock behavior and performance scaling with increased clock-rates on Vega: FE. It’ll be interesting to see how RX Vega manages and if its clock is boosted, but we’ll wait until RX Vega’s launch for more coverage of that aspect."
http://www.gamersnexus.net/guides/2986-amd-vega-hybrid-mod-results-overclocking-liquid-vs-air
 
Vega: FE Water Cooled Internals & Break-Down
Gamers Nexus
Published on Jul 15, 2017
Our break-down of the AMD Vega: Frontier Edition watercooled video card internals & cooling solution.
https://www.youtube.com/watch?v=oSZz6xoooAw

Vega: Frontier Edition Water Cooled Internals & Tear-Down
By Steve Burke Published July 15, 2017 at 2:54 am
amd-vega-fe-water-1.jpg

"The card’s inlet is on the left side of the pump block, at which point liquid is fed in and cycles through the usual copper microfins and coldplate channels. This is all handled by Cooler Master, just like AMD’s previous Fury X CLC and Radeon Pro Duo CLC cards. Although some aspects look similar to the Radeon Pro Duo, likely no components were reusable.

Anyway, once the liquid feeds in through the inlet, it eventually feeds out of the outlet and hits the VRM components. This includes chokes, FETs, and capacitors – everything, basically. The VRM components are hit with liquid after the core and HBM, meaning that the water will be pre-heated, but the VRM can handle far higher heat than the GPU + HBM2. This is the ideal way to route the coolant."

http://www.gamersnexus.net/guides/2988-vega-fe-watercooled-tear-down-and-internals
 
looking at what we got so far i think the AIO variant will be clock at 1600mhz. just that the AIO version will be more stable to keep that 1600mhz all the time. it might be possible to see 1700mhz edition from board partner but i don't know how AMD want to handle this. after all there is no custom Fury X exist. even if they can OC the card to 1800mhz mark the gain probably will be smaller than polaris going from 1200mhz+ to 1500mhz. one thing for sure though. AMD is having hard time to nail power efficiency on their current architecture once they try to go for higher clock. it is the sign that i also saw with polaris.

when we look at "gaming" pascal we thought that it was merely maxwell clocked higher. but most of us never really think about how nvidia able to do so without making the power consumption going crazy.
 
Fixing Vega FE: Undervolting to Improve Performance & Power Draw
By Steve Burke Published July 17, 2017 at 5:15 pm
"This feature benchmark dives into one of the top requests we received from our Patreon backers: Undervolt Vega: Frontier Edition and determine its peak power/performance configuration. The test roped us in immediately, yielding performance uplift largely across the board from preliminary settings tuning. As we dug deeper, once past all the anomalous software issues, we managed to improve Vega: FE Air’s power available to the core, reduce power consumption relative to this, and improve performance in non-trivial ways."

Vega: Frontier Edition Undervolting Power Consumption
vega-undervolt-current_3.png


Undervolting Impact on Thermals – Vega: Frontier Edition
vega-undervolt-thermals_3.png


Vega FE’s Struggle to Maintain Frequency
vega-undervolt-frequency_3.png


FireStrike Ultra – Vega Undervolted Benchmark
vega-undervolt-firestrike-ultra.png

"FireStrike Ultra starts us out. The Vega FE Air card when completely stock ran a graphics score of 4906, with our 50% power offset cards both operating at around 5370 graphics score. This includes the undervolted card, which manages about a 9-10% performance uplift over the stock card. Here’s the crazy thing: Again, we’re not overclocking to achieve this. All we’re doing is making more power available while reducing the voltage, which nets a marginal power consumption spike at the trade of more consistent and faster frametimes. That’s a pretty good trade for 15W, and is far better than the 87W of the power offset without undervolting.

For point of reference, our Hybrid FE overclock performed at 5774, which is 7% faster than the undervolted card. Kind of puts into perspective just how far undervolting and over-powering will get you."

vega-undervolt-timespy.png

"TimeSpy gives us a gain of about 7.6% from the undervolted card over the stock card, with our Hybrid OC gaining another 9.6% on top of that – though drawing significantly more power at around 33A.

Here’s FireStrike Extreme:"
vega-undervolt-firestrike-extreme.png

"As for games, some experienced instability at 1090mv and had to be moved up to 1100mv; For Honor was particularly unstable, and required a core voltage of about 1120mv."

vega-undervolt-grw-4k.png

"At 4K and with VH settings, the undervolted AMD Vega Air card performed at 41FPS AVG, with lows close by at 37 and 36. The stock card with no modifications operated at 37.7FPS AVG, resulting in a performance uplift of 8.8% from the stock card. This uplift is because the stock card cannot maintain 1600MHz without a power offset – but again, a power offset without overvoltage increases your power consumption by 80-90W, thereby increasing thermals that the card deals with. This undervolting and over-powering appears to be the best approach to extracting more performance."
vega-undervolt-doom-4k.png

"With DOOM using Vulkan, Async Compute, and rendered at 4K, the Vega undervolted card operates at 71.6FPS AVG, with low-end frametimes also improved over the stock card. Our AVG FPS improvement is about 11.5% in DOOM, following the trend of DOOM being a somewhat best-case scenario for AMD on a routine basis. The performance uplift is tremendous when considering our minimal power consumption increase and better overall control on the card."

Conclusion: Overpower, Undervoltage Far Better than Stock Config
But not without their caveats.

"The trouble with this solution is that it is imperfect by nature. First, every chip is not made the same; ours may undervolt better or worse than others out there, and that means there’s no easy “use these numbers” method. You’ll ultimately have to guess and check at stability to find the numbers that work, which means more work is involved in getting this solution to be rock-steady. That’s not to say it’s difficult work, but it’s certainly not as easy as plugging a card in and using it. We found that some games required 1120mv to remain stable, while others were fine at 1090mv. Ideally, you’d make a profile for each application – but that’s a bit annoying, and becomes difficult to maintain. The next option would be to choose the lowest stable voltage for all applications (in our case, that might be 1120mv). You lose some of the efficiency argument when doing this, as the bottom-end is cut off, but still gain overall.

A straight +50% overpower configuration is a huge waste of power down the PCIe cables, which results in running hotter than necessary and thereby louder.

Software is also buggy and frustrating. No, not everyone sees the same issues – that’s the nature of buggy software. It is difficult to precisely pinpoint the issue causing HBM2’s brutal downclocking of -445MHz, but we have seen it happen routinely and on multiple systems with multiple environments. We think that this has to do with manually configuring all 7 DPM states and their corresponding voltage states; when we only configured DPMs 4-7, the downclocking issue did not occur. Fan speed curves are also inaccurate, and report about 200RPM higher than what the user requests. Wattool has the same bugs as WattMan, and Afterburner can’t adjust voltage (yet). The point isn’t to say that it’s impossible to undervolt like we did, it’s just to say that you should really be aware of all the different variables when tweaking. It’s possible to inadvertently hinder performance (in major ways) if HBM2 underclocks without the user’s knowledge. Keep an eye on it.

As for the task at hand, it seems the best possible configuration is to overpower the core (+50%), undervolt the core (roughly -110 to -90mv), and run a fan RPM that keeps temperatures at or below 80C. That’ll depend on your cooling solution, case, and case/room ambient temperatures.

This yields a decent boost to application performance (professional and gaming) without costing the insane +90W draw of a straight +50% overpower configuration."

http://www.gamersnexus.net/guides/2990-vega-frontier-edition-undervolt-benchmarks-improve-performance

Edit: The video https://www.youtube.com/watch?v=OZkfYDWwo7s
 
Not much faith still with RX Vega...

I think a potential watercooled version of it at 1080ti price will not be as nice as the Fury X next to the 980ti was back then because the performance delta now is bigger. And I will be *very* surprised if they reach parity.

Cheers!
 
to me something still doesn't add up. 232mm2 polaris is very competitive with 200mm2 GP106. so i would imagine something like 330-350mm2 polaris will be competitive with 314mm2 GTX1080. i believe Raja already said that Vega is 484mm2? GP102 is 471mm2. so looking at die size alone if Vega is build upon polaris (and further improved) then it should be not that behind vs fully enabled GP102. actually it might put Vega directly on the same level as 1080ti. unless Vega is simply Fiji on steroid.......
 
Not necessarily. It might just be that for gaming workloads GCN doesn't scale as well as nVidia's uArch, or there's hardware in there (which I think it is the case) not being utilized fully for current code paths. This last bit is probable, due to some DX12 extensions not being correctly used in graphical engines yet, but with the initial batch of games we have that sport the "DX12" logo, Vega will be a repeat of Fury it seems.

Cheers!
 
Status
Not open for further replies.