AMD Vega Frontier Edition Unboxed and Benchmark (updated)

Published by

Click here to post a comment for AMD Vega Frontier Edition Unboxed and Benchmark (updated) on our message forum
#5447255
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
"Rushed"? If AMD wait until drivers are fine, Volta will be here (drivers included). The hardware comes late and his drivers are not going to make a miracle bump in gaming performance. Too much HYPE= Big expectation= Big disappointment
I don't think Volta is going to release consumer variants until next year. I could be wrong, but my opinions are based on that. If AMD spent another 1-2 months cooking good drivers for a proper release, I don't think it would have bit them as hard as releasing the cards in this state. Especially after the majority of the AMD community spent the last 6 months hyping the fact that Vega should be the first GPU released with "proper" drivers.
#5447257
https://forums.guru3d.com/data/avatars/m/250/250418.jpg
Surely nobody will buy less for more(gaming wise)
These are not gaming cards... Plus drivers are not optimized: when Polaris came out it had crap drivers too. One year later and the drivers made the card gain allot of performance.
Im pretty sure that Vega Frontier is targeted at machine learning workloads, so complaining about its gaming performance is kinda silly.
Plus complaining about it being a heater...
I could repeat the mantra that I'm actually certain is true, that the drivers aren't even close to ready, but then why even release them with a driver like that AMD? :S
I'm sure they only marketed this as a workstation card, so no gaming drivers ready.
To all above, card isn't for gaming. Full stop.
Agree.
You're failing to address the main issue which is that this GPU was released with gaming drivers.
If it wasn't, people would bash it for that... You can't please everyone it seems.
"Rushed"? If AMD wait until drivers are fine, Volta will be here (drivers included). The hardware comes late and his drivers are not going to make a miracle bump in gaming performance.
AMD drivers do make big bumps in performance. Not saying Vega will have a big bump though.
#5447258
data/avatar/default/avatar35.webp
If that was the case, why release with gaming drivers at all? Me too, to be honest, but I would prefer it to be a bit more delayed than bad.
What are gaming drivers? Just because one IHV calls "game ready" it's driver, "gaming drivers" means nothing. Of course professional GPUs run DirectX graphics, as they run OpenGL and Vulkan, but this does not those GPUs have some tweaks like surface surface format substitution or some floating point optimizations enabled, but all this doesn't prevent you to run a "gaming" application on those GPUs.If there is something to blame on this card, is that they did not bring back FP64 top performance. It would be nice to see a 4->2->1 configuration for FP16-FP32-FP64.
#5447261
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
What are gaming drivers? Just because one IHV calls "game ready" it's driver, "gaming drivers" means nothing. Of course professional GPUs run DirectX graphics, as they run OpenGL and Vulkan, but this does not those GPUs have some tweaks like surface surface format substitution or some floating point optimizations enabled.
This isn't quantum physics man, it's marketing 101. You don't tack on a mode called "gaming mode" and then expect people not to take the results at face value in gaming. You also don't compare/market the card to a Titan Xp, which is regarded as a "gaming card hybrid thing" then when it fails at beating it go "oh well it wasn't intended to compete in gaming!" despite the fact that their own marketing used the Titan Xp as a comparison in a workload the Titan Xp also wasn't intended for. I don't think either me or PrMinister are complaining about the drivers or the card itself - it's more of AMD's marketing team once again dropping the ball. The amd subreddit is filled with people complaining about the performance. The fact that we are having this discussion right now means that the marketing failed. Which is my main complaint. Better choice would have been to forego the gaming mode and brand the card WX and seed it to review sites focused on workstations/deep learning. No one would have complained. No one would have cared about it's gaming performance.
#5447263
https://forums.guru3d.com/data/avatars/m/259/259654.jpg
I'm sure they only marketed this as a workstation card, so no gaming drivers ready.
There are gaming drivers ready for it. It even has a (buggy) software switch for that. If it was just a professional card (which it should have been), they shouldn't have included the gaming driver.
AMD drivers do make big bumps in performance. Not saying Vega will have a big bump though.
I actually believe it will. That doesn't make their decisions up to now unquestionable.
What are gaming drivers? Just because one IHV calls "game ready" it's driver, "gaming drivers" means nothing. Of course professional GPUs run DirectX graphics, as they run OpenGL and Vulkan, but this does not those GPUs have some tweaks like surface surface format substitution or some floating point optimizations enabled, but all this doesn't prevent you to run a "gaming" application on those GPUs.If there is something to blame on this card, is that they did not bring back FP64 top performance. It would be nice to see a 4->2->1 configuration for FP16-FP32-FP64.
There is no IHV games here. AMD does ship this card with a software switch between "gaming" and "professional" drivers. That alone is confusing. They should have shipped it with only the pro driver until the other part was ready. On the other hand, the person testing this doesn't seem to have an adequate PSU, and judging by the rest of his comments he's not exactly a system optimizer. A bit of patience, more of these will popup. Not that AMD's list of bugs for the current drivers seems non-alarmist.
#5447276
data/avatar/default/avatar25.webp
There is no IHV games here. AMD does ship this card with a software switch between "gaming" and "professional" drivers. That alone is confusing. They should have shipped it with only the pro driver until the other part was ready.
That's not how a graphics driver works. There is no thing like "gaming driver" except for single application profiles. Everything else is just fancy useless software like "Wattman" "Relive" "Chill" or whatever are named (the same happen with the other IHV). EDIT: to be clear, of course there could be (I bet there are) different driver branches for that card. On one branch could work one team (focusing on professional software optimizations) and another branch where another team is working on general APIs support and optimizations, and another branch where another team is just working in the control panel GUI features. But this doesn't mean AMD is working on 2 or more set of different drivers, this is just the common procedure when you work on project that isn't just a school homework. Different branches for different software pieces are meant to synchronize with each other after a piece of work is down and then start another job, all this in loop.
#5447286
data/avatar/default/avatar03.webp
It performs worst than GTX 1080 and it costs at least double from GTX 1080, wonder who will ever buy it?
Because you do not run Solidworks on a GTX 1080 at workplace.... at least you want to be getting fired.
#5447293
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
Because you do not run Solidworks on a GTX 1080 at workplace.... at least you want to be getting fired.
Yep and that's why Nvidia doesn't include a "workstation mode" on the GTX1080...
#5447298
data/avatar/default/avatar02.webp
And if they do not release a Vega in June, the wiseguys will shout "They still don't have any Vega cards but in July-August".
And if they do release in June, it's mission accomplished? Like who the hell gives a FF about the card which can be used for gaming? Was that anyone's idea of "Vega" H1 release? - $1000-1500 card which in gaming gets beaten by $700 card. I would say fine if at least this was a compute monster excelling in pro applications, but if that was the case they wouldn't be selling it for 1500 peanuts. Already down from reported MSRP price. Another question, if gaming Vega turns all-rightish, lets say close to 1080Ti performance for $600, what's stopping those niche Vega FE potential buyers (all 5 of them) to just get Vega RX instead and use pro drivers?
#5447299
https://forums.guru3d.com/data/avatars/m/164/164033.jpg
I think in the end the drivers will shake out and the performance will be relatively fine. I don't think these results are indicative of the actual hardware. It's just a really bad way to release a new series, especially one as hyped as Vega is.
Agree with you on all points. After rather good release that was Ryzen, this is horrific.
#5447313
data/avatar/default/avatar20.webp
Anyway, on topic, it certainly doesn't look good but apparently, the card is throttling because it's being starved of power (Onna states in his thread that the PSU used is 550W can anyone confirm this?).
Radeon Pro WX 9100(same Vega chip, true pro-built card) is clocked at 1200MHz. This thing has Typical Clock of 1382MHz and the Peak Clock of 1600MHz. http://pro.radeon.com/en-us/product/radeon-vega-frontier-edition/ So throttling is a feature, not a bug. One thing which seems to be lost is the bandwidth issue. HBM2 benefits seems to be used entirely to keep the perf/Watt in check, no spec. benefits over GDDR5X whatsoever: Vega FE comes with less bandwidth than the Fury X(!) 1080 Ti/Titan Xp improve on 980 Ti/Titan X bandwidth by 43% and 63% respectively. 1080Ti comes with 25% more bandwidth than 2x GTX1060 Vega FE has 6% less bandwidth than the 2 Polaris cards combined
#5447315
https://forums.guru3d.com/data/avatars/m/259/259654.jpg
That's not how a graphics driver works. There is no thing like "gaming driver" except for single application profiles. Everything else is just fancy useless software like "Wattman" "Relive" "Chill" or whatever are named (the same happen with the other IHV). EDIT: to be clear, of course there could be (I bet there are) different driver branches for that card. On one branch could work one team (focusing on professional software optimizations) and another branch where another team is working on general APIs support and optimizations, and another branch where another team is just working in the control panel GUI features. But this doesn't mean AMD is working on 2 or more set of different drivers, this is just the common procedure when you work on project that isn't just a school homework. Different branches for different software pieces are meant to synchronize with each other after a piece of work is down and then start another job, all this in loop.
The way it sounds to me (switch with delay), sounds like they do switch between two different drivers.
#5447317
data/avatar/default/avatar35.webp
Because you do not run Solidworks on a GTX 1080 at workplace.... at least you want to be getting fired.
That's exactly the reason you would want to run a Quadro (P2000, P4000, P5000 or P6000) instead of a TitanXp in Solidworks. 🙂
#5447318
https://forums.guru3d.com/data/avatars/m/250/250418.jpg
There is a chance, that it was a problem with the testing [spoiler]https://image.prntscr.com/image/U2caFii5QiKnl4tzgk-fVw.png[/spoiler]
This is a case of a guy with too much money and lack of knowledge. Using a 550W PSU with a 300W TDP GPU... The card is either throttling due to heat or not enough power.
Drivers can partially alleviate a lack of hardware performance. AMD drivers made a big jump in DX11 performance in the first half of 2015 but even that was insufficient to reach Nvidia DX11 drivers performance.
When RX480 came out, the 1060 beat the crap out of it. How about today?
There are gaming drivers ready for it. It even has a (buggy) software switch for that. If it was just a professional card (which it should have been), they shouldn't have included the gaming driver. I actually believe it will. That doesn't make their decisions up to now unquestionable.
All in all, we should wait for a competent review to be made.
That's exactly the reason you would want to run a Quadro (P2000, P4000, P5000 or P6000) instead of a TitanXp in Solidworks. 🙂
The price of a Quadro though...
#5447319
https://forums.guru3d.com/data/avatars/m/231/231931.jpg
When RX480 came out, the 1060 beat the crap out of it. How about today?
On average still slower. RX580 is like 2-3% faster on average than 1060 Games with poor multi-threading in DX9-11 AMD drivers are performing poor compared to pascal
#5447320
data/avatar/default/avatar10.webp
the price of a quadro though...
p4000 $850, p5000 $2000. FE's performance is about P4000 level.
#5447324
https://forums.guru3d.com/data/avatars/m/115/115710.jpg
It's a workstation card, ppl are just embarrassing themselves at this point.
Is it? AMD has workstation graphics line out. Also they compared this to Titan Xp instad of Quadro.
#5447325
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
This is a case of a guy with too much money and lack of knowledge. Using a 550W PSU with a 300W TDP GPU... The card is either throttling due to heat or not enough power.
It's throttling due to heat. There is a reason why the 375w WC edition of the card has the exact same specifications and clock speed. Also:
Typical Engine Clock (MHz) 1382
Which is why the guy couldn't get it to run at 1600mhz stable. AMD has no problem advertising the tflops at the 1600 though - I find that amusing. http://pro.radeon.com/en-us/product/radeon-vega-frontier-edition/
When RX480 came out, the 1060 beat the crap out of it. How about today?
https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1060/26.html 10% is hardly "beat the crap" - and now it's 3%. https://www.techpowerup.com/reviews/Sapphire/RX_580_Nitro_Plus/30.html
All in all, we should wait for a competent review to be made.
Which is pretty much the entire argument. That they didn't seed reviewer cards, which created this mess.
The price of a Quadro though...
Is cheaper? Uses half the power? Has application certifications? Doesn't have driver issues with professional apps? http://www.nextwarehouse.com/item/?2715581_g10e
#5447327
data/avatar/default/avatar14.webp
The price of a Quadro though...
Just ECC memory slone is a valid reason to buy a Quadro.. Plus optimized drivers.

Previous Page

Next Page