Support NeoGAF

[napata]

This explains the rather bad performance: http://www.pcgameshardware.de/Vega-...adeon-Vega-Frontier-Edition-GPU-Takt-1231795/

Basically the typical clocks for the air-cooled FE are 1.382 mhz according to AMD. I don't know why they claim the 1600mhz clocks then when it almost never seems to reach them.

I wonder what the clocks for the RX variant are going to be and if it can actually run at the clocks AMD will market.

 

[bomblord1]

These are workstation cards what's the point in comparing them to gaming focused cards?

 

[riflen]

These are workstation cards what's the point in comparing them to gaming focused cards?

The point is that there are no other Vega products released. The FE is the same silicon that'll be launched next month as RX Vega. The FE is a prosumer product, designed to handle professional workloads and gaming workloads. It's worth testing it on games so we get some idea about how the silicon and cooling system perform. And yes, we should bear in mind that AMD say RX will be better at games than FE.

FYI, if you compare the Vega FE against pure workstation products, it barely beats 480 Polaris and loses to $875 150W Pascal Quadro cards. The point of the FE is to allow professionals to create and test on the same system, rather than use two PCs at greatly added cost. The closest Nvidia analogue is the Titan, but it's not a perfect comparison because Nvidia do not allow Quadro drivers to be used with the Titan, whereas AMD provide a FE driver that does both pro and gaming workloads with a software toggle.

 

[ethomaz]

I read all that too, but really a 80+ platinum 550W PSU in a mini ITX system and a 4790K should be totally fine. GPUs don't clock down if the PSU is insufficient.
But yes, we'll basically have to wait until Siggraph until all the possible reasons for these results can be brushed aside.

GPUs doesn't boost if the PSU is not enough... so it will throttle at the base clock.

These are workstation cards what's the point in comparing them to gaming focused cards?

It is the same chip with same early gaming driver.

The differences will be probably the cooling system and clocks.

 

[Firenze1]

These are workstation cards what's the point in comparing them to gaming focused cards?

AMD released gaming drivers for it.

 

[riflen]

GPUs doesn't boost if the PSU is not enough... so it will throttle at the base clock.

I'm sorry but this is not true at all. Providing the thermal conditions are suitable, the GPU will try to draw the power it needs. If the PSU cannot provide this consistently and cleanly you will either get an application crash or the PSU will trip an overdraw safety limit and cause the computer to reset.
There is no mechanism that changes the clock of a GPU if the power provided is less than expected.

 

[dgrdsv]

Proper comparison to Quadro cards in the same resolution:

Seems to be slower than P5000 (GTX1080 analogue) and P4000 (GTX1070 analogue) cost only $870 although it's probably quite a bit slower being a single slot one 6-pin power connector card. In any case, I don't think that NV will have any issues countering Vega FE Pro performance with something like P4500.

I wonder what the clocks for the RX variant are going to be and if it can actually run at the clocks AMD will market.

I think that gaming Vega cards will be even slower as they'll have to cost less and thus have worse cooling solutions. Plus the only way I see them being able to push more out of this chip while staying inside the 300W PCIE limit is putting half of HBM2 chips (8GBs) power draw into the chip's clocks - and this will likely be a very minor boost considering that HBM2 isn't very power hungry.

GPUs doesn't boost if the PSU is not enough... so it will throttle at the base clock.

GPU would crash with insufficient power. I've never seen a GPU whos performance was affected by the PSU. It either works or it doesn't.

 

[ethomaz]

I'm sorry but this is not true at all. Providing the thermal conditions are suitable, the GPU will try to draw the power it needs. If the PSU cannot provide this consistently and cleanly you will either get an application crash or the PSU will trip an overdraw safety limit and cause the computer to reset.
There is no mechanism that changes the clock of a GPU if the power provided is less than expected.

That is what happens with RX 480... it throttles below the boost clock and close to base clock when the PSU is not enough.

GPU would crash with insufficient power. I've never seen a GPU whos performance was affected by the PSU. It either works or it doesn't.

Both RX 480 and GTX 10xx didn't reach the max boost clock if the PSU is not enough... it throttles near the base clock. That happened a lot with RX 480 at launch when it needed over 150W while the PSU couldn't supply that.

Perhaps it crash if there is no power to sustain the base clock.

 

[JohnnyFootball]

3DMark FireStrike graphics scores
Graphics Score
GTX 1080 Ti AMP Extreme 29613
GTX 1080 Ti Founders 28340

GTX 1080 AORUS 11 Gbps 23197
Vega Frontier "gaming mode" 22916 <--
GTX 1080 AORUS Xtreme 22805
GTX 1080 Founders 21905

GTX 1070 Gaming X 19103
GTX 1070 Founders 18255

Fury X 16081

RX 580 Strix 1411 MHz 14624
GTX 1060 Gaming X 9 Gbps 13544
RX 480 Gaming X 1303 MHz 13411
GTX 1060 Founders 13156
RX 480 Reference 1266 MHz 12195



so yeah, it's basically a GF 1080.

Unless AMD price this well, it will be dead on arrival.

Why is 1080 speeds considered dead? That's not too bad.

 

[ethomaz]

Why is 1080 speeds considered dead? That's not too bad.

Because GTX 1080 is already dated in terms of performance for high-end (it was launched in May 2016) and Volta is near launch.

It was not a bad result if launched last year with Polaris.

 

[digitalrelic]

Why is 1080 speeds considered dead? That's not too bad.

Well for starters, Vega FE is $999. AMD will have to split that price in half if their RX line has similar performance. I don't know how they're going to do that. Also, the 1080 is over a year old. AMD not being able to match nVidia's flagship product from over a year ago is not a good look.

 

[napata]

Why is 1080 speeds considered dead? That's not too bad.

a 300+ watt ~500 mm² card being on par with a year old 180 watt ~300 mm² card is really bad.

 

[Wag]

Because GTX 1080 is already dated in terms of performance for high-end (it was launched in May 2016) and Volta is near launch.

It was not a bad result if launched last year with Polaris.

How close is Volta to launch? I just purchased a 1080 Ti, and was planning on getting a 2nd (I game @ 4k). If it's that close I might hold off on the 2nd.

 

[Firenze1]

High Bandwidth Memory is really disappointing. AMD should ditch it for navi.

 

[ISee]

How close is Volta to launch? I just purchased a 1080 Ti, and was planning on getting a 2nd (I game @ 4k). If it's that close I might hold off on the 2nd.

Probably Fall 2017.

 

[Sqrt_minus_one]

a 300+ watt ~500 mm² card being on par with a year old 180 watt ~300 mm² card is really bad.

GP104 is about 410mm2

But aside from that it's not looking great for Vega to debut a year afterwards, higher power draw and be more expensive to produce for the same performance.

Nevertheless I am sure AMD will price it appropriately

^ will be pleasantly surprised if we see consumer Volta this year. But I would not bet on it

 

[atpbx]

How close is Volta to launch? I just purchased a 1080 Ti, and was planning on getting a 2nd (I game @ 4k). If it's that close I might hold off on the 2nd.

I've just bought one and a 4K monitor to go with it.

It's the age old thing of you can keep waiting for the next best thing and you'll never upgrade.

 

[napata]

GP104 is about 410mm2

But aside from that it's not looking great for Vega to debut a year afterwards, higher power draw and be more expensive to produce for the same performance.

Nevertheless I am sure AMD will price it appropriately

^ will be pleasantly surprised if we see consumer Volta this year. But I would not bet on it

I thought it was around 300 mm² and google seems to agree with me. Where do you get the 410 mm² from?

 

[FutureProtagonist]

GP104 is about 410mm2

https://www.techpowerup.com/gpudb/2839/geforce-gtx-1080

It's 314mm^2.

 

[dgrdsv]

Both RX 480 and GTX 10xx didn't reach the max boost clock if the PSU is not enough... it throttles near the base clock. That happened a lot with RX 480 at launch when it needed over 150W while the PSU couldn't supply that.

Perhaps it crash if there is no power to sustain the base clock.

First time I'm hearing this and this definitely not how NV cards operate. Could you provide some links on this maybe?

How close is Volta to launch? I just purchased a 1080 Ti, and was planning on getting a 2nd (I game @ 4k). If it's that close I might hold off on the 2nd.

There will always be something better coming soon. Tbh, right now I don't think that Volta will even launch in gaming markets. Vega hardly being able to beat a mid range year old competitor's GPU certainly puts no pressure on NV to launch anything new soon.

 

[ethomaz]

How close is Volta to launch? I just purchased a 1080 Ti, and was planning on getting a 2nd (I game @ 4k). If it's that close I might hold off on the 2nd.

Some months... probably late this year.

 

[Papacheeks]

The point is that there are no other Vega products released. The FE is the same silicon that'll be launched next month as RX Vega. The FE is a prosumer product, designed to handle professional workloads and gaming workloads. It's worth testing it on games so we get some idea about how the silicon and cooling system perform. And yes, we should bear in mind that AMD say RX will be better at games than FE.

FYI, if you compare the Vega FE against pure workstation products, it barely beats 480 Polaris and loses to $875 150W Pascal Quadro cards. The point of the FE is to allow professionals to create and test on the same system, rather than use two PCs at greatly added cost. The closest Nvidia analogue is the Titan, but it's not a perfect comparison because Nvidia do not allow Quadro drivers to be used with the Titan, whereas AMD provide a FE driver that does both pro and gaming workloads with a software toggle.

I'll wait for specific drivers and api enhancements for developers before grabbing the pitchforks. From what they showed off a while back on vega is with support from developers for api, and engine development drivers for pushing the card will get much better.

 

[Sqrt_minus_one]

I thought it was around 300 mm² and google seems to agree with me. Where do you get the 410 mm² from?

https://www.techpowerup.com/gpudb/2839/geforce-gtx-1080

It's 314mm^2.

My bad I had 413mm2 stuck in my head for the longest time I checked it ages ago.

My apologies you guys are right

 

[PFD]

a 300+ watt ~500 mm² card being on par with a year old 180 watt ~300 mm² card is really bad.

Man when you put it that way, this card is sounding more and more like a train wreck

 

[michaelius]

Both RX 480 and GTX 10xx didn't reach the max boost clock if the PSU is not enough... it throttles near the base clock. That happened a lot with RX 480 at launch when it needed over 150W while the PSU couldn't supply that.

Perhaps it crash if there is no power to sustain the base clock.

That was because radeon 480 was going above power that could be delivered through PCI-Express slot and 6 pin connector - not because of PSU.

 

[dgrdsv]

Videocardz

Guru3D

Same source btw.

One thing to remember here is that the WC version will probably be able to clock higher than the ~1400MHz of the AC version. But I think that even the WC version won't be able to reach 1080Ti/TitanXp.

 

[riflen]

The 550W PSU complaints are shown to be nonsense anyway, because since then another person has received a Vega FE and benched it in lots of games and applications. His results:

• Time Spy Graphics = 6785 pts
• Fire Strike Ultra Graphics = 5091 pts
• Doom Vulkan 4K Ultra @ 55-65 FPS
• The Witcher 3 @ 28-35 FPS at 4K Hairworks on / 41-42 FPS with Hairworks off (80 C running the game)
• Troubles OCing past 1650 MHz with the blower
• Tried to OC HBM, starts at 945 MHz, managed 960 MHz stable and at 980 MHz it got too hot
• Tester says you won't be able to OC without additional cooling, thermal throttling (80-85 C)
• Tester put 2x 80 mm fans to help (open air case) and it was down to 75 C
• Tester doesn't think it will touch the 1080 Ti performance-wise, only 1070-1080s
• Mining: 30-35 MH/s
• Cinebench R15 OpenGL: 97.39 FPS
• Initial testing done using Pro drivers, then switched to the Gaming drivers and performance didn't change
• Card initially operate at 1348-1528 MHz in gaming mode running The Witcher 3 (37-42 FPS), then after retesting it was back to 1650 MHz max (and delivered the same 42 FPS as Pro mode)
• >300W for the card
• EVGA P2 1200W Platinum Power Supply + Ryzen R7 1800X CPU

 

[Firenze1]

The 550W PSU complaints are shown to be nonsense anyway, because since then another person has received a Vega FE and benched it in lots of games and applications. His results:

• Time Spy Graphics = 6785 pts
• Fire Strike Ultra Graphics = 5091 pts
• Doom Vulkan 4K Ultra @ 55-65 FPS
• The Witcher 3 @ 28-35 FPS at 4K Hairworks on / 41-42 FPS with Hairworks off (80 C running the game)
• Troubles OCing past 1650 MHz with the blower
• Tried to OC HBM, starts at 945 MHz, managed 960 MHz stable and at 980 MHz it got too hot
• Tester says you won't be able to OC without additional cooling, thermal throttling (80-85 C)
• Tester put 2x 80 mm fans to help (open air case) and it was down to 75 C
• Tester doesn't think it will touch the 1080 Ti performance-wise, only 1070-1080s
• Mining: 30-35 MH/s
• Cinebench R15 OpenGL: 97.39 FPS
• Initial testing done using Pro drivers, then switched to the Gaming drivers and performance didn't change
• Card initially operate at 1348-1528 MHz in gaming mode running The Witcher 3 (37-42 FPS), then after retesting it was back to 1650 MHz max (and delivered the same 42 FPS as Pro mode)
• >300W for the card
• EVGA P2 1200W Platinum Power Supply + Ryzen R7 1800X CPU

really meh benchmarks. Let's hope Gaming VEGA does a little better atleast.

I have no reason to replace my Fury X if it is on 1070 level. Volta it is.

 

[ethomaz]

Anybody knows if Vega is using TSMC or GF?

I have no reason to replace my Fury X if it is on 1070 level. Volta it is.

It is probably on 1080 level but yeap no reason to jump.

 

[dgrdsv]

• Time Spy Graphics = 6785 pts
• Fire Strike Ultra Graphics = 5091 pts
• Doom Vulkan 4K Ultra @ 55-65 FPS
• The Witcher 3 @ 28-35 FPS at 4K Hairworks on / 41-42 FPS with Hairworks off (80 C running the game)

For reference:

TW3 HW are off there. FS Ultra is a couple of posts above.

Doom Vulkan result is really weird. If it's correct then Vega's Vulkan drivers are likely to be in a bad shape for now.

 

[Firenze1]

For reference:



TW3 HW are off there. FS Ultra is a couple of posts above.

wait what? Fury X is on the same level in doom 4k? wtf is AMD doing?

 

[LelouchZero]

The alleged performance leaks so far paint a lackluster picture, hopefully by the time the gaming variant releases the performance is in a better shape.

 

[Firenze1]

The alleged performance leaks so far paint a lackluster picture, hopefully by the time the gaming variant releases the performance is in a better shape.

I am a big AMD fan (own ryzen and fury x), but there is no surprise that AMD didnt send review samples of VEGA FE.

 

[riflen]

The results are so underwhelming that there absolutely has to be at least 10% more performance coming with RX Vega. Is it possible that RX silicon is going to be binned and able to clock higher than 1650 Mhz? That just seems so unlikely given the price points of FE and RX products.
Given this is AMD and knowing what we know about GF and clocks, I'm more inclined to think that FE was pushed out the door to meet 1H 2017 and that drivers will absolutely be better next month.

EDIT: Response from RTG employee when asked if the current driver for FE is gimped somehow in games.

It's not gimped (that would be completely ridiculous), but it is older.

 

[longdi]

Vega FE had 16gb of hbm2 but the bandwidth is no better than 1080ti n Txp, and hbm1 has damn poor overclocking. Seems like a bad decision by AMD

 

[Durante]

How the hell does this thing not consistently outperform Fury X at least? (Look at e.g. the Doom results)

There has to be something wrong still on the software side. I expect improvements of ~25% at least in that benchmark. Because even a 25% improved result would still be disappointing.

 

[Firenze1]

Vega FE had 16gb of hbm2 but the bandwidth is no better than 1080ti n Txp, and hbm1 has damn poor overclocking. Seems like a bad decision by AMD

Hmm what was the selling point of using HBM again?

 

[What A Time To Be Alive]

How the hell does this thing not consistently outperform Fury X at least? (Look at e.g. the Doom results)

There has to be something wrong still on the software side. I expect improvements of ~25% at least in that benchmark. Because even a 25% improved result would still be disappointing.

Thats what I'm saying. This thing is litterally a Fury X.

There has to be something going on from a software perspective.

 

[lordfuzzybutt]

How the hell does this thing not consistently outperform Fury X at least? (Look at e.g. the Doom results)

There has to be something wrong still on the software side. I expect improvements of ~25% at least in that benchmark. Because even a 25% improved result would still be disappointing.

Yep. I mean it is understandable for AMD, or any vendor, to drop the ball. However, this is equivalent to them taking the ball and smash it on the ground.

 

[riflen]

How the hell does this thing not consistently outperform Fury X at least? (Look at e.g. the Doom results)

There has to be something wrong still on the software side. I expect improvements of ~25% at least in that benchmark. Because even a 25% improved result would still be disappointing.

In the case of DOOM, which is a really weird result, it's possible the Vulkan driver is not great yet, as dr_rus speculated.

 

[longdi]

Hmm what was the selling point of using HBM again?

Strangely Vega has lower memory bandwidth than fury x.. either that or gpuz is reading only half of its actual value.

Another thing is vega has the same number of cores as fury x, 4096. So how more efficient was polaris to Hawaii?

 

[Khaz]

Are we expecting a full range of Vega cards, or just a couple of high end products?

 

[dgrdsv]

How the hell does this thing not consistently outperform Fury X at least? (Look at e.g. the Doom results)

There has to be something wrong still on the software side. I expect improvements of ~25% at least in that benchmark. Because even a 25% improved result would still be disappointing.

Doom's result is easy to explain with early Vulkan drivers, the card seems to perform more or less according to its specs in DX11 and DX12. Also worth noting that game results from games with no built-in benchmarks can heavily depend on which level/area/scene was tested by different benchmarkers.

What's not so great though is the apparent lack of any boosts from architectural improvements of GCN5 when compared to Fiji's GCN3. If we consider Fury X a rough equal of 980Ti then linear scaling of the same GCN3 architecture should result in GTX1080 performance on ~11,5TF mark - which gives us ~1400MHz for a design of the same width as Fiji which is awfully close to the stated typical clocks of Vega FE AC.

So either GCN5 is a dud or some features are disabled / not fully optimized yet in the drivers for now.

 

[DPB]

Are we expecting a full range of Vega cards, or just a couple of high end products?

Just high-end. The mid range is already there in the form of Polaris rebrands.

 

[Firenze1]

I hope AMD price gaming vega 499. Only way i am getting one.

 

[Paganmoon]

I hope AMD price gaming vega 499. Only way i am getting one.

That'd be a bad price if it's performance stays like this really. You can get GTX 1080's for that price. They'll not get marketshare, nor mindshare with that price.

 

[Firenze1]

That'd be a bad price if it's performance stays like this really. You can get GTX 1080's for that price. They'll not get them marketshare, nor mindshare with that price.

Ok 399 then

 

[michaelius]

How the hell does this thing not consistently outperform Fury X at least? (Look at e.g. the Doom results)

There has to be something wrong still on the software side. I expect improvements of ~25% at least in that benchmark. Because even a 25% improved result would still be disappointing.

There are two different sources for those benchmarks - considering this isn't canned timedemo this makes it impossible to make any comparison

 

[Drakhoran]

PC Perspective have gotten their hand on a Vega FE and will be streaming some benchmarking.

 

[RootCause]

So whats the expected price?