AMD Vega (FE and RX) Benchmarks [Updated Aug 10 - RX Vega 64 Unboxing]

[Despoiler]

Whenever a company says something about their product, they tend to say it in the most positive light aka best case scenario. If RX Vega truly traded blows with a 1080 across the board, AMD would be showing more canned benchmarks with more averages, and not limiting the scope of their comparisons to blind tests, 4-5 games, and minimum frame rates only.

Yes they will try to paint it in the best light. However, in terms of reading comprehension trading blows does not mean

it will lose more than it wins.

You are trying to move the statement from one of equivalence to one of losing.

 

[Det0x]

I'm not in the camp that Nvidia crippled or neglected Kepler; I think Kepler suffered from a combination of aging bad due to design decisions and AMD getting a massive boost during that time frame with the console contracts using GCN coming to fruition.

Which camp are you in regarding Pascal Titan ?

Do you think Nvidia crippled/neglected/gimped GP102 titan in "pro apps" by choice, or did they just now suddenly manage to unlock its real performance in applications for prosumers and creatives ? (ages after GP102 quadro)

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...iver-unlocks-performance-in-pro-apps.2513072/

Probably not a response to Vega FE indeed...

And... Ops more then the mythical 20% performance increase from a driver update O_0

 

[crashtech]

Seems like one would be forced to either believe Nvidia was intentionally crippling cards, or that driver magic really is possible, but only when Nvidia does it.

 

[SinOfLiberty]

^^ This is like saying there is a game X that runs good on Titan,

Your average answer is: Who The Fck Cares...

Stay with more grounded reports, focused on gaming optimized cards.

 

[PeterScott]

Seems like one would be forced to either believe Nvidia was intentionally crippling cards, or that driver magic really is possible, but only when Nvidia does it.

NVidia and AMD both segregate drivers for Consumer and Pro cards, where the drivers on consumer cards are limited in some work stations operations. You can call it "Crippling" if you want, but its the same segregation that both companies have been doing for many years.

AMD released the in between "Frontier Edition" card to make a deadline, and they enabled proper workstation drivers on this card that was neither consumer, nor full pro (no certified drivers).

NVidia essentially followed by doing the same for the Titan which is also a kind of in between card, that doesn't get full pro drivers.


Now there are essentially 3 classes of cards:

1: Full unlocked pro cards with certified drivers
2: In between cards with unlocked workstation performance, but no certified drivers. (Titan and Vega FE)
3: Consumer cards that have weak driver performance on some workstation work loads.

 

[tviceman]

Yes they will try to paint it in the best light. However, in terms of reading comprehension trading blows does not mean You are trying to move the statement from one of equivalence to one of losing.

No, I'm not. One product can win 3 benchmarks and lose 7. That is still trading blows, even if it's 5-10% slower overall.

 

[tviceman]

Which camp are you in regarding Pascal Titan ?

Do you think Nvidia crippled/neglected/gimped GP102 titan in "pro apps" by choice, or did they just now suddenly manage to unlock its real performance in applications for prosumers and creatives ? (ages after GP102 quadro)

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...iver-unlocks-performance-in-pro-apps.2513072/

Probably not a response to Vega FE indeed...

And... Ops more then the mythical 20% performance increase from a driver update O_0

Of course Titan Xp was purposefully crippled in certain pro apps and of course it was unlocked in response to Vega FE. Neither company cripples desktop consumer cards for gaming purposes, so your point is entirely invalid in the scope of this conversation.

 

[Paratus]

As of right now I'm thinking the best Vega card maybe an aftermarket RX Vega 56.
Give it:

• Excellent cooling
• 2 8Pin power connections
• OC ability approaching core and memory speed equivalent to water cooled Vega64
• ~$425-$450
• No mining tax (not likely right now)

If Vegas problems really are thermal throttling and ROP/Bandwith to SPs ratio then the RX 56 is probably better balanced, especially if mem OC to RX64 can be reached. A good aftermarket cooler takes care of thermal throttling and suddenly it should be nipping at the heels of RX64 and the 1080 for an attractive price.

If the problem is front end throughput beefed up power delivery for a core OC should put it in striking distance if the RX64 and 1080.

Out the box the RX56 has 10% more bandwidth per teraflop than the RX64.

If 1.89GT/s is possible from the default 1.6 then RX56 suddenly has 30% more bandwidth per teraflop than the RX64.

Basically dropping 8CUs reduces power and cost at no impact to over all performance if an aftermarket card can allow for similar core and mem clocks..... Maybe.

 

[godihatework]

Which camp are you in regarding Pascal Titan ?

Do you think Nvidia crippled/neglected/gimped GP102 titan in "pro apps" by choice, or did they just now suddenly manage to unlock its real performance in applications for prosumers and creatives ? (ages after GP102 quadro)

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...iver-unlocks-performance-in-pro-apps.2513072/

Probably not a response to Vega FE indeed...

And... Ops more then the mythical 20% performance increase from a driver update O_0

ITT:

People who pretend that they don't know what product segmentation is or how it's achieved...

 

[Veradun]

The one dropping support all over is AMD, already they dropped Win8 support for all of their GPUs (including Polaris & Vega)


https://www.techpowerup.com/235525/amd-releases-the-crimson-relive-edition-17-7-2-whql-drivers

Windows 8 will be EOL in exactly 4 months. Who cares?

 

[Malogeek]

Windows 8 will be EOL in exactly 4 months. Who cares?

Agreed. Stupid waste of limited resources developing and validating for an OS that no one uses. Anyone who had W8 would have upgraded to W10 during the free period. Anyone too stubborn or hated W8 will have stayed on W7 and it doesn't apply to them either.

 

[CatMerc]

http://m.imgur.com/9VZDxXi?r

Looks like RX Vega won't be pegged to 300W.

 

[Bouowmx]

AMD can do odd things with respect to GPU power: "GPU power consumption" may mean "GPU-only" and not total (including memory and board).

 

[CatMerc]

AMD can do odd things with respect to GPU power: "GPU power consumption" may mean "GPU-only" and not total (including memory and board).

Considering in Overwatch it's reaching 285W, I'd guess it's total.

The GPU package (including memory) power limit is supposed to be 220W.

 

[Stuka87]

AMD can do odd things with respect to GPU power: "GPU power consumption" may mean "GPU-only" and not total (including memory and board).

I have never seen them produce numbers of a GPU that only had numbers of the chip. They have talked about chip power consumption, but never in reference to testing a GPU (GPU being entire card). Which is what they are doing here. They are talking about a specific card model.

 

[AtenRa]

http://m.imgur.com/9VZDxXi?r

Looks like RX Vega won't be pegged to 300W.

What is the performance with Chill enabled ?? that is the big question.

 

[Magic Hate Ball]

What is the performance with Chill enabled ?? that is the big question.

Probably locked to 60hz v-sync?

 

[mohit9206]

As of right now I'm thinking the best Vega card maybe an aftermarket RX Vega 56.
Give it:

• Excellent cooling
• 2 8Pin power connections
• OC ability approaching core and memory speed equivalent to water cooled Vega64
• ~$425-$450
• No mining tax (not likely right now)

If Vegas problems really are thermal throttling and ROP/Bandwith to SPs ratio then the RX 56 is probably better balanced, especially if mem OC to RX64 can be reached. A good aftermarket cooler takes care of thermal throttling and suddenly it should be nipping at the heels of RX64 and the 1080 for an attractive price.

If the problem is front end throughput beefed up power delivery for a core OC should put it in striking distance if the RX64 and 1080.

Out the box the RX56 has 10% more bandwidth per teraflop than the RX64.

If 1.89GT/s is possible from the default 1.6 then RX56 suddenly has 30% more bandwidth per teraflop than the RX64.

Basically dropping 8CUs reduces power and cost at no impact to over all performance if an aftermarket card can allow for similar core and mem clocks..... Maybe.

I applaud your optimism. But Polaris has already been proven to be a terrible overclocker and vega looks to be no different. These are already operating at max capacity with hardly anymore left in the tank. Amd already squeezed the maximum clocks they could out of it.

 

[Krteq]

Probably locked to 60hz v-sync?

Chill isn't a framerate limiter.

 

[Veradun]

Chill isn't a framerate limiter.

And that slide says "Radeon Chill disabled" anyway

 

[Malogeek]

And that slide says "Radeon Chill disabled" anyway

Read the sentence after the highlighted part. It's a comparison with Chill off and on.

 

[AtenRa]

And that slide says "Radeon Chill disabled" anyway

It has numbers for both enabled and disabled

 

[FatherMurphy]

I think it is important to note that the tests were run at 1080p. Thus, the test shows the average power consumption at that resolution in those games. Vega was built with 4k in mind (or so we are told) and, therefore, should have the rasterization hardware for that resolution. Is it not a reasonable conclusion that power consumption of the GPU would increase not insubstantially if the GPU must make use of all of its hardware?

 

[Snarf Snarf]

Those are large drops in power, I recently undervolted my 980ti in Overwatch to keep power down so I don't bake alive in my room while the sun is up. Medium settings locked at 120 fps -25% power target got me down about 40W and roughly 8-9 C on core temp with absolutely no drop in performance. I had long suspected that the FPS limiter in Overwatch doesn't actually effect power consumption very much because my card was drawing the same power at 60/120/300 fps with the settings the same. Maybe it just culls the extra frames but still renders them? That would explain the massive drop in power consumption if Chill just does what the frame limiter is supposed to do. Without knowing the performance numbers its hard to draw a conclusion, but those are some pretty good looking numbers

 

[railven]

People hyped Async compute to the moon and beyond! 3 years later, it gives like what? 10-15% in best case scenarios, in truth it averages to about 5~10%. Now async is gone and primitive shaders are supposed to give 70% more performance! HYPE!

This is what ultimately bothers me. The console contracts, hardware Async Compute, and at the end of the day, AMD either didn't get the gains these people were expecting or the functions went unused because the devs didn't seem to bother.

And we're back in the circle with the same predictions, and yada yada yada. What makes these claims even harder to stomach is the constant "proof of the contrary to what I say" yet they're whole angle and prediction is based on marketing statements and probabilities. But the one time, that one dev, that one game that uses these functions properly in the sea of future titles and these people feel vindicated.

Welps, perhaps with more Ryzen money AMD can secure more devs. I'll believe it when I see it.