AMD Radeon VII 16 GB

[notb]

It seems that this site has all the data anyone here could possibly want on this subject:

http://www.sengpielaudio.com/calculator-levelchange.htm

I think it doesn't (not in the actual physics part, anyway).
But it is bashing hasty usage of psychoacoustics (a lot), so it was a nice read after all. ;-)

 

[sepheronx]

The first sentence is a straw man. I didn't make the rules about logical expression with the English language, which is what my post was about. Logical expression dictates, regardless of your feelings (i.e. the pretentiousness complaint), that a person can only take one position at a time, not two contradictory positions simultaneously.

When a person says a card is loud and that they can't stand noise, that person is clearly saying noise is unacceptable to them and that the card is loud. This means that the card is included in the realm of things the person cannot tolerate. This means that, for that person, the card is not good (also known as unacceptable).

Performance encompasses all aspects of how a product performs, including the noise it produces. Furthermore, regardless of how many FPS it delivers, you already declared that the product belongs to the category of unacceptable products due to its noise output and your lack of tolerance for noise. Draw a Venn diagram.

Finally, the accusation about pretentiousness is an example of concern trolling (tone complaint). It's a distraction from the point and an example of the ad hominem fallacy.


It seems that this site has all the data anyone here could possibly want on this subject:

http://www.sengpielaudio.com/calculator-levelchange.htm

So reviewers are wrong in saying that pro being performance and con being noise should instantly mean the card is bad, even if they say it isn't a bad card? I can't stand certain noise but can tolerate it regarding other considerations to the card. Still seems rather odd to me that you can negate every other consideration.

If it may help: I can't stand noise. But depending on situation and other criteria, I can look past it.

Is that better?

You are nitpicking for pretty much no reason. You decided to single someone out for what? I would rather you spend your time worrying about other things than cherry picking a sentence cause you seemed to have misunderstood the point. Other than derailung the thread, why not talk about the card? If you have nothing else to say other than being sentence police, might I suggest ignoring me from now on please?

 

[Super XP]

Without factoring in the noise, power draw (Despite being 7nm) and high $$ tag, the Radeon VII isn't a bad card after closely looking at several reviews. W1zzards wonderful review already gave me this same conclusion. But had to venture out and check another 10-12 sites. The majority all have similar results as W1zzards review. All on par with what this card is capable of.

 

[notb]

Without factoring in the noise, power draw (Despite being 7nm) and high $$ tag, the Radeon VII isn't a bad card after closely looking at several reviews. W1zzards wonderful review already gave me this same conclusion. But had to venture out and check another 10-12 sites. The majority all have similar results as W1zzards review. All on par with what this card is capable of.

Without factoring the noise, fuel consumption and high price, Focus RS is a perfect everyday car.

Honestly... with this kind of logic you can justify anything.

 

[R0H1T]

The part about high power draw is slightly exaggerated, nearly all Vegas can be undervolted & in case of VII the power savings are substantially higher. Depending on workload & Si lottery, the max (system) power draw can be reduced anywhere between 50W to nearly 100W - granted this ought to have been done by AMD or board partners themselves, but it's not as big an issue as it was on the original Vega.

 

[londiste]

The part about high power draw is slightly exaggerated, nearly all Vegas can be undervolted & in case of VII the power savings are substantially higher. Depending on workload & Si lottery, the max (system) power draw can be reduced anywhere between 50W to nearly 100W - granted this ought to have been done by AMD or board partners themselves, but it's not as big an issue as it was on the original Vega.

Silicon lottery seems to be a thing, if I remember correctly it was Guru3D who wrote that undervolting gave them 10, maybe 15 watts.
My Vega64 undervolting resulted in a very modest boost. If my memory serves right, about +40MHz at the same power consumption. It still hit the power limit and did not reach boost clocks.

By the way, have you seen any reviews doing serious undervolting for RTX2080 for comparison?

 

[R0H1T]

Any reviews that show 2080 (or 2080Ti) being undervolted & clocking higher/better like in case of VII - no I haven't. Also btw Nvidia's x80 have been the most efficient GPU in their lineups since Kepler.

Silicon lottery seems to be a thing, if I remember correctly it was Guru3D who wrote that undervolting gave them 10, maybe 15 watts.

They didn't specify whether they undervolted whilst OCing, also you have a couple of German sites (earlier in the thread) testing this under more workloads/games & the results were impressive.

From legit reviews - https://www.legitreviews.com/amd-radeon-vii-16gb-video-card-review_210489/14

3DMark Firestrike

• Stock 1069mv, GPU 1801MHz, HBM2 1000MHz, 0% Power – 342 Watts in GT1 (overall 21,724 / GPU 26,941)
• Undervolted to 934mV, GPU 1801MHz, HBM2 1000MHz, +20% Power – 291 Watts in GT1 (overall 21,658 / GPU 27,240)
• Undervolted to 1000mV, GPU 1854MHz, HBM2 1100MHz, +20% Power – 301Watts in GT1 (overall 21,447 / GPU 26,433)
• Undervolted to 934mV, GPU 1854MHz, HBM2 1100MHz, +20% Power – 274 Watts in GT1 (overall 21,249 / GPU 26,397)
• Undervolted to 1000mV, GPU 1854MHz, HBM2 1200MHz, +20% Power – 306 Watts in GT1 (overall 21,513 / GPU 26,559)
• Overvolted to 1100mV, GPU 1854MHz, HBM2 1200MHz, +20% Power – 349 Watts in GT1 (overall 21,312 / GPU 26,146)

So, with roughly a 50MHz increase on the GPU and 200MHz increase on the HMB2 memory we noticed no performance improvement on 3D applications. What we did learn is that dropping the GPU voltage dramatically lowers the power on this card. Dropping the power from 1069mV at stock settings down to 934mV took the power from 342 Watts down to just 274 Watts. That is a 68 Watt reduction or about 20% lower power consumption. It of course also lowered temperatures and noise levels!

It feels like AMD was overly generous on GPU voltages, so you’ll be able to play the silicon lottery game with Radeon VII cards and see if you got a good card for undervolting. So, overclocking was a no go on our test sample, but undervolting was amazing and worth looking into.

 

[londiste]

934mV is a very good result. Computerbase was at ~950mV. Most sites tend to and up at around 980mV. Both voltage and clocks are test-specific though and each site seems to use a different test for OC and UV
Undervolting to under 1V range does put Radeon VII efficiency to around RTX 2080 level.

The thing to keep in mind about undervolting that it is by no means a guaranteed thing. It has a lot in common with overclocking in that regard. From forum and reddit posts from Radeon VII owners there are games that will need more voltage to be stable, Battlefield V and Shadow of Tomb Raider are usually brought out as examples.

 

[R0H1T]

I can bet my anonymous internet reputation that every VII can be undervolted, given the stock voltages we're seeing, then again the increased efficiency will depend on a lot of things including workload & the rest of the system.

 

[EarthDog]

One has to wonder why, if these perform so much better and use so much less power undervolting (and many can/do) why did AMD put this version of the VII in the wild??? Feels like its shooting themselves in the foot, no?

Why wouldnt they want to be closer to the 2080 in performance and power? Theres a reason for it...

 

[R0H1T]

I'm not saying it's as good as 2080 in terms of performance or efficiency, it is however better than Pascal in both areas. Also wrt why AMD did this - money? Apart from Ryzen & TR you'd be hard-pressed to remember any spectacular launches from the red camp in well over half decade, if not more. This doesn't justify the outcome but possibly could explain why AMD do this repeatedly.

 

[londiste]

Also wrt why AMD did this - money?

AMD needs to be in the picture. It is a forced move. Vega64 could no longer cut it as a viable product high enough in the ladder. Leaving pricing aside, RTX 2080Ti, RTX 2080, RTX 2070 and GTX 1080Ti from last generation were all offering higher performance. At the same time GTX 1080 and RTX 2060 are uncomfortably close. Plus all the Titans.

Radeon VII would be efficient as hell with voltage and frequency pulled just a little bit back. Unfortunately, that would put it in a really uncomfortable place with regards to performance and I doubt it could be priced much lower than the $699 MSRP.

Edit:
In fact, Computerbase tested it at 240W power limit (-20%, the lowest it would go). It apparently lost a mere 3% performance overall and 6% in the worst game:
https://www.computerbase.de/2019-02..._vs_rx_vega_64_bei_gleicher_leistungsaufnahme

 

[EarthDog]

I'm not saying it's as good as 2080 in terms of performance or efficiency, it is however better than Pascal in both areas. Also wrt why AMD did this - money? Apart from Ryzen & TR you'd be hard-pressed to remember any spectacular launches from the red camp in well over half decade, if not more. This doesn't justify the outcome but possibly could explain why AMD do this repeatedly.

I'm not saying you did, nor was I talking about Pascal.

I'm not asking why they released the card in general... (its b/c of market presence, not money.. the rumor is they are losing their arse on these cards...remember these are $1500 cards with a different bios and drivers in the Instinct cards)...


I wondering why if "all" the cards undervolt and show performance improvements, why didnt they release it like that? It would be more attractive in the market..

 

[TheGuruStud]

Looky what I found https://techreport.com/review/34461/revisiting-the-radeon-vii-and-rtx-2080-at-2560x1440/3

Manual testing reveals canned benchmarks can be a total joke. Surprise.
This has been going on for a while.

 

[trog100]

I'm not saying you did, nor was I talking about Pascal.

I'm not asking why they released the card in general... (its b/c of market presence, not money.. the rumor is they are losing their arse on these cards...remember these are $1500 cards with a different bios and drivers in the Instinct cards)...


I wondering why if "all" the cards undervolt and show performance improvements, why didnt they release it like that? It would be more attractive in the market..

i think you are right about high end market presence.. they have a desperate need for that.. worth losing a bit of money for a limited edition (gaming) run i doubt these things will ever be available in real numbers..

trog

 

[os2wiz]

We all know AMD needs a complete GPU Re-Design. Until this happens, this is what they can offer. Though I do believe the price is at least $150 too high, despite the 3 x AAA Games included.

Hi Super XP. I ordered Radeon VII 2 days ago on AMD's website. it will be delivered Friday. I placed the order when I heard the new driver was released and that the FP64 compute function is better than on any other consumer graphics cards. Some strategy games definitely will take advantage of that capability plus Adobe Premier. I expect more games will eventually be programmed to leverage FP64 compute. Strategy games are really giant data bases which obviously use a lot of compute function per turn.

Like many, you have misconceptions of how games/game engines work.
It is exceedingly rare that games have specific optimizations for hardware, and even making such optimizations would be nearly pointless, since they would be tied to specific iterations of GPU architectures, not vendors.

The kind of bias that exists in some games are usually not intentional, but simply a consequence of the game being developed for and tested on one set of hardware. Such bias is in most cases favoring AMD, as many AAA titles are developed exclusively for AMD based consoles, while "no" games are developed exclusively for Nvidia.

And I want to emphasize; just because card A scales better than card B in a game, doesn't mean there is bias in the game engine, it can simply be a consequence of better balance of resources on that card. So a card performing better in some games than other is not proof of bias, but often a mismatch between people's expectations and the reality.

W hat you are saying is only a half-truth at best. Optimizations have made immense performance differences when AMD has collaborated with software vendors. This goes on with Nvidia which pays development costs for far more games than AMD does. So the developers automatically use Nvidia hardware . Yes different hardware has different theoretical capabilities . Often times these capabilities are never reached. Then sometimes the cost of AMD to fully enable a technology in Vega has proven too time consuming and costly to pursue.We know thatat least 2 capabilities in Vega were never exploited as each developer would have to invest too much time and money to employ it in their games so no API was ever written for it.

 

[os2wiz]

Based on the drasticallly improved Valentines Day driver from AMD another review is in order. Undervolting now works flawlessly. The easily over clock to 2000 mhz and above. The HBM2 can overclock to 1150 mhz. . The card has significant room for overclocking. I am positive if this review is repeated with the latest driver performance will be be at least 5 per cent better. Fanspeed and noise is much easier to regulate. Night and day difference.

 

[londiste]

I placed the order when I heard the new driver was released and that the FP64 compute function is better than on any other consumer graphics cards. Some strategy games definitely will take advantage of that capability plus Adobe Premier. I expect more games will eventually be programmed to leverage FP64 compute. Strategy games are really giant data bases which obviously use a lot of compute function per turn.

No, games will not take advantage of FP64. On one hand, there are not enough GPUs around with sufficient FP64 performance on the other hand games do not need that level of precision. As the best case scenario for FP64 is 1/2 the performance of FP32, that hit is huge. Right now, the drive is exactly the opposite - as both Vega and Turing can do FP16 at 2 times FP32, games are trying to use less precision wherever they can to get more performance.

Often times these capabilities are never reached. Then sometimes the cost of AMD to fully enable a technology in Vega has proven too time consuming and costly to pursue.We know thatat least 2 capabilities in Vega were never exploited as each developer would have to invest too much time and money to employ it in their games so no API was ever written for it.

These were not pursued because AMD was not able to get them to work properly. The capabilities need to be implemented and exposed in APIs and/or drivers before developers can start using them.

 

[os2wiz]

No, games will not take advantage of FP64. On one hand, there are not enough GPUs around with sufficient FP64 performance on the other hand games do not need that level of precision. As the best case scenario for FP64 is 1/2 the performance of FP32, that hit is huge. Right now, the drive is exactly the opposite - as both Vega and Turing can do FP16 at 2 times FP32, games are trying to use less precision wherever they can to get more performance.
These were not pursued because AMD was not able to get them to work properly. The capabilities need to be implemented and exposed in APIs and/or drivers before developers can start using them.

That those features did not work was not a hardware failure. It was the amount of investment required to develop the API and train software developers to implement it. At that time AMD was just starting to emerge from the financial funk they had fallen into. The time required to fix the features was too long to make a difference with Vegas success. AMD now should revisit those capabilities . They now have considerably more resources.
But no one has responded to my post on the latest drivers from AMD that now unlock the full overclocking and undervolting capabilities of the gpu and memory. Coul amount to greater than 10^ performance improvement and eliminates the loud fan issue.

 

[evolucion8]

That those features did not work was not a hardware failure. It was the amount of investment required to develop the API and train software developers to implement it. At that time AMD was just starting to emerge from the financial funk they had fallen into. The time required to fix the features was too long to make a difference with Vegas success. AMD now should revisit those capabilities . They now have considerably more resources.
But no one has responded to my post on the latest drivers from AMD that now unlock the full overclocking and undervolting capabilities of the gpu and memory. Coul amount to greater than 10^ performance improvement and eliminates the loud fan issue.

Yeah, it is not the first time AMDs old technology gets revisited and then implemented on a newer API. Tessellation is one of them, Geometry Instancing is another one, along with other small features like the ones on DX10.1 and DX11.2 that got implemented later on even on DX12 which even has other features grabbed from Mantle which evolved in Vulkan. So I think that 16-Bit precision will make a good comeback without resorting to proprietary stuff.

 

[os2wiz]

Another dissapointment. 1TB/s bandwidth and it's not even better than my 1080Ti (stock vs stock) at my resolution (3440x1440)....
Excellent review as always!

Poor selection of games . I believe only 3 were DX12. Why were there no Stardock games?? Star Control Origins is a great game that is DX 12 The skewed game selection made the results more lopsided than they should have been. Now that the February 14 drivers from AMD are available the overclocking works completely (manual and auto) both of memory and gpu. The undervolting works as well. Fan noise now completely contollable without performance loss. I ask Tech Powerup to rerun the review the differnce in results will be surprising.

So reviewers are wrong in saying that pro being performance and con being noise should instantly mean the card is bad, even if they say it isn't a bad card? I can't stand certain noise but can tolerate it regarding other considerations to the card. Still seems rather odd to me that you can negate every other consideration.

If it may help: I can't stand noise. But depending on situation and other criteria, I can look past it.

Is that better?

You are nitpicking for pretty much no reason. You decided to single someone out for what? I would rather you spend your time worrying about other things than cherry picking a sentence cause you seemed to have misunderstood the point. Other than derailung the thread, why not talk about the card? If you have nothing else to say other than being sentence police, might I suggest ignoring me from now on please?

The noise issue is dead. Yesterdays driver release allows for undervolting which will reduce fan speed and noise with almost no impact on performance.

 

[BaneSilvermoon]

When adding the washers, did you tighten down the screws or leave them a little loose? I tried some Nylon washers on mine yesterday and saw no change whatsoever. My over-clocked card still reaches 114c.

Would love to see those temps come down and be able to happily maintain the OC, or possibly push it higher. Have gotten +191mhz on the gpu and +200mhz on the vram. Significant performance boost.

 

[arbiter]

Poor selection of games . I believe only 3 were DX12. Why were there no Stardock games?? Star Control Origins is a great game that is DX 12 The skewed game seklection made the results more lopsided than they should have been. Now that the February 14 drivers from AMD are available the overclocking works completely (manual and auto) both of memory and gpu. The undervolting works as well. Fan noise now completely contollable without performance loss. I ask Tech Powerup to rerun the review the differnce in results will be surprising.

Saying its poor selection of games is just making excuses for why it has poor performance. The games used are high end games that a lot of people would play either now or when they were out. Star control origins only had 1 month of over 1000 peak players and as for their other game "Ashes of the Singularity" was for most part a glorified tech demo made in to a game. Using games that never really had much of a player base is a waste of time.

 

[os2wiz]

Without factoring in the noise, power draw (Despite being 7nm) and high $$ tag, the Radeon VII isn't a bad card after closely looking at several reviews. W1zzards wonderful review already gave me this same conclusion. But had to venture out and check another 10-12 sites. The majority all have similar results as W1zzards review. All on par with what this card is capable of.

I hope you have seen Not an Apple Fan 's Youtube review of the Radeon 7 that he reexamined after thelatest Adrenalin driver release on February 14. All issues that had previously been reported have been resolved : fan noise by applying auto undervolt, maual overclocking can now bring gpu frequency at least to 2000mhz with a reports of close to 2100mhz, Memory can be overclocked consistently to at least 1150mhz with reports of up to 1200mhz. He changed his conclusions which were negative in his initial review to neutral now. He says he no longer regrets purchasing the card, only that it costs $700 . But he also acknowledges that with 16GB of HBM2 it could not cost less than $700.

Saying its poor selection of games is just making excuses for why it has poor performance. The games used are high end games that a lot of people would play either now or when they were out. Star control origins only had 1 month of over 1000 peak players and as for their other game "Ashes of the Singularity" was for most part a glorified tech demo made in to a game. Using games that never really had much of a player base is a waste of time.

Star Control Origins by Stardock is a top 50 game. It should have been in the mix. Are you going to stretch credibility and say very few of the DX12 games are popular? That would be a complete lie.

 

[dalekdukesboy]

Hey Wizzard or anybody, you overclock the card and run Heaven benchmark...but you don't list the settings or resolution?