The 51%/98% GPU usage dilemma

[flash-gordon]

What was that? Can someone explain what is the scenario AMD was trying to show?

Was AMD using different CPUs or FPS target control?

Also, does anyone know what's the GPU usage in AotS for 390 crossfire in DX12?

 

[LTC8K6]

I think the 51% was probably one GPU in a Xfire setup?

 

[Stuka87]

I think it was not explained correctly. I think one GPU was at ~100%, and the other was at ~51%.

That or maybe they had it frame limited, or power limited, or something along those lines.

 

[sm625]

I think AMD had FRTC enabled, and amazingly enough FRTC was actually working correctly. It is the only way to explain how they were able to claim "Better Performance + Efficiency". Because the 1080 would only be drawing 175W during that test. That means the Radeons had to be drawing only about 88W each (or less) during that test. At 51% utilization I can believe 88 watts. It would have been nice if they had explicitly told us what the power consumption for each card was. For all we know it could have been 70 watts. No one benches cards at 50% so I have no idea where that level of utilzation fits onto the power curve.

 

[Bryf50]

Probably just intentionally vague. They wanted to compare against Nvidias latest and greatest but didn't want to put up a graph showing themselves slower than it.

 

[Genx87]

Wouldnt worry about it. In a month we will have reviews.

 

[Bacon1]

The game was CPU bound at that point so it couldn't send the GPU more work, so it was going unutilized.

They didn't want to show single 480 vs 1080 since it would be (slightly?) slower, and people would ignore the huge $200-250 vs $700 price difference.

Kinda like 290 vs 780 Ti all over again, except even bigger price difference.

 

[Hitman928]

The game was power/driver bound at that point so the game couldn't send the GPU more work, so it was going under utilized.

They didn't want to show single 480 vs 1080 since it would be a decent bit slower, and people would ignore the huge $200-250 vs $700 price difference.

Just my take on the situation.

 

[Bacon1]

Just my take on the situation.

I'd doubt its driver bound, its DX12. Yes there are still driver updates that are needed to make sure the API calls are correct, but there isn't the overhead like in previous DX versions.

Ashes is very CPU intensive since its RTS. Same as Total War Warhammer. Even the 1080 was slightly bound since it wasn't @ full 100% usage either.

Considering 51% is just barely over 50% (HALF), half of 2 cards is 1, so one card could be very similar in performance to 1080.

 

[tential]

Just my take on the situation.

If a single rx 480 is slightly slower then a 1080 then its faster than a 1070 and far cheaper. They could have just used a 1070.

If it's a decent bit slower it may be on par with a 1070 or even slightly slower. So why not mask it with a crossfire vs 1080 test. In a game with great crossfire scaling as well. Then take it back a bit with frame rate targeting to get the efficiency win too.


Amd is being vague though as usual. Nvidia's launch was better. By far. Even with the founder's edition pricing mess.

 

[Bacon1]

If a single rx 480 is slightly slower then a 1080 then its faster than a 1070 and far cheaper. They could have just used a 1070.

If it's a decent bit slower it may be on par with a 1070 or even slightly slower. So why not mask it with a crossfire vs 1080 test. In a game with great crossfire scaling as well. Then take it back a bit with frame rate targeting to get the efficiency win too.


Amd is being vague though as usual. Nvidia's launch was better. By far. Even with the founder's edition pricing mess.

They might not have had a 1070 to test against, reviews just came out a few days ago and you don't want to completely change your presentation last minute, especially for people that aren't used to presenting

 

[Stuka87]

Just my take on the situation.

It would be incredibly hard to have a DX12 game be driver bound.

The gaming being CPU bound is quite possible though, it is very CPU intensive. This could definitely result in the GPU's twiddling their thumbs once they hate the frame rate limit.

 

[swilli89]

I read it as the crossfire element (the second card) was at 51% utilization. This puts the 480 used in the comparison at around 75% of a 1080 which would be amazing.

 

[AtenRa]

SFR ???

Pay attention to the last two quotes.

- Frames no longer need to be queued, time between frame completion and user viewing reduced to 2-3x.
- Using the GPUs in parallel to work on one frame allows to multiple GPUs to behave like one much more powerful GPU.

The dual RX480 with only 51% utilization was faster than GTX1080 and perhaps at the same or lower energy consumption.

 

[poofyhairguy]

Can someone explain what is the scenario AMD was trying to show?

A scenario where AMD is still relevant to most GPU buyers despite not having a competitor for the entire top part of Nvidia's lineup for most (all?) of 2016.

The sad truth about technology is that halo products influence the purchase of non-halo products. The fact that the non reference 980 ti was the fastest card in the world got boiled down by people who are more casual about gaming into "Nvidia cards are the fastest." Heck I bet the 980 ti as a brand sold more GTX 960s than any 960 review every did, people want to be a part of the same crowd as "the best" because they assume the 980 ti's dominance is an example for the entire product line.

AMD knows this, that is why it tried (and failed) to shoot for the top spot with Fiji. This generation they are doing things differently, they are hoping to make OEMs happy (who are making gaming laptops or ready-to-play Oculus boxes or Macs) instead of trying to woo consumers who won't do the research necessary to make rational decisions. It's kinda telling that AMD went from "we aren't a budget brand" to "we want to increase the VR market." It's like the graphics group saw the same writing on the wall that their best isn't good enough to warrant a premium price that had been staring at the CPU side of AMD for almost a decade.

Before anyone takes that as me bashing GTX customers or Nvidia customers in general, please note that this sort of thing is VERY VERY common in technology products. LG makes OLED tvs not because the TVs themselves make a lot of money (actually I think they lose money on each one sold when counting R&D), they make the OLED TVs so they can win every "TV shootout" to influence consumer behavior. The LG OLED TV line is a halo product meant to basically trick low information consumers into buying LG's (very very very crappy) LED TVs that make up the majority of LG's annual TV sales. They want people to read the headline "LG (OLED) tvs are the best" (because the OLEDs are- like comparing a GTX 260 to a 1080 when comparing LED and OLED) and assume that means all LG tvs. And some consumers will do that, the poor bastards.

And it's not just technology where we see the halo effect. Nissan doesn't make the GT-R to make a ton of money. They make it so little boys put posters of it on the wall, and internalize somewhere that the Nissan brand can be "cool." That way when they are 35 years old and they have to buy a minivan to haul the kids around they pick a Quest instead of another competitor because that lame mommy van has the same logo on it that the once desired GT-R has.

We all take mental shortcuts in life when cutting down our consideration set for purchases, and review sites for technology or cars want to write about (and play with) the biggest and the best instead of the products that 90% of people will actually buy. The whole game is rigged against consumer laziness, and poor AMD simply doesn't have the ammo to do the kinds of things it needs to do to make consumers make poor assumptions about their product like Nvidia's products benefit from. So instead they are going to make Apple happy, because Apple is buying based on actual value.

 

[monstercameron]

I dont know man, halo products can be useful for sure...but there has to be a breaking point in the that logic.

Car analogy time!

Its like saying you want a sentra because GTRs (GOJIRA) are cool.

 

[boozzer]

from what raj tried to convey in the stream, what he was trying to said was that there is alot of head room for improvement in the polaris cf vs almost nothing in the 1080.

that is my take on it. no idea if it is true or not, we need benchmarks.

 

[Hitman928]

I'd doubt its driver bound, its DX12. Yes there are still driver updates that are needed to make sure the API calls are correct, but there isn't the overhead like in previous DX versions.

The driver does less work in DX12, but it's still integral to the performance of cards. A bad DX12 driver can still have a very negative impact on performance. With that said, I should have been more clear in that I think it's possible that a user setting within the driver going for max efficiency at a known performance target was being used which was causing less than full load. Immature driver could have also had a hand but I don't think it would be to the extent shown.

If a single rx 480 is slightly slower then a 1080 then its faster than a 1070 and far cheaper. They could have just used a 1070.

If it's a decent bit slower it may be on par with a 1070 or even slightly slower. So why not mask it with a crossfire vs 1080 test. In a game with great crossfire scaling as well. Then take it back a bit with frame rate targeting to get the efficiency win too.

Amd is being vague though as usual.

Agree with this post. Why not just compare it to a 380 or a 960 or even a 980? Performance of a 980+ at less than half the cost, would that not be a pretty good headline to have? Heck that's pretty much exactly what the (presumably) leaked WSJ article said earlier that day. Why not show just that? AMD was purposefully making a direct comparison difficult while trying to show their cards as "winning". They don't want to show the actual performance of their cards in a way that would be easy to compare to yet.

 

[poofyhairguy]

I dont know man, halo products can be useful for sure...

Sure, if you are in the market for that kind of product.

The problem with using halo products as an indicator for the entire product line is consumers have been conditioned for decades to have certain expectations (namely that the company with the best single product will often have the best product lineup), and some companies use that conditioning against consumers to the point where halo products have lost their use as an indicator.

I personally will never ever buy a non-halo product because the halo product from a company is great- I am going to research the product I will actually buy and purchase it based on its merits. The only use I have for halo product reviews when I don't intend to buy that product is to get a general sense of where a market is and where it is going.

In fact the only personal use for halo products (that I don't want to buy) I can think of is they are a good way to explain relative value down the road. For example "the GTX 1070 is as fast as a 980 ti" is a much more convincing argument than "the GTX 1070 is ___% faster than the GTX 970" because it gets across the point that YOU Mr. non-halo buying average person gets to taste the same fruit that the 1% once tasted and bragged about. Heck I will admit I bought a GTX 970 I really didn't need because "it was as fast as a Titan OG," even though the only reason I was enamored with the Titan OG was because Nvidia stuck that stupid $1000 price on it. We are programmed as humans to want what we can't have, even if what we can't have isn't that great to begin with.

 

[swilli89]

SFR ???

Pay attention to the last two quotes.

- Frames no longer need to be queued, time between frame completion and user viewing reduced to 2-3x.
- Using the GPUs in parallel to work on one frame allows to multiple GPUs to behave like one much more powerful GPU.

The dual RX480 with only 51% utilization was faster than GTX1080 and perhaps at the same or lower energy consumption.

So its possible they were using SFR, but the FPS doesn't make sense.. its obviously not locked to 60fps so what is going on?

 

[Despoiler]

I dont know man, halo products can be useful for sure...but there has to be a breaking point in the that logic.

Car analogy time!

Its like saying you want a sentra because GTRs (GOJIRA) are cool.

I can't even count how many times I've seen a Nissan Sentra where the owners stuck a red GTR badge on the back. I still think it's a poor analogy though. Comparatively fewer people dream of high end GPUs like they dream of high end cars.

 

[monstercameron]

I can't even count how many times I've seen a Nissan Sentra where the owners stuck a red GTR badge on the back. I still think it's a poor analogy though. Comparatively fewer people dream of high end GPUs like they dream of high end cars.

Ill tell ya, a good car analogy is hard to make

 

[tential]

Sure, if you are in the market for that kind of product.

The problem with using halo products as an indicator for the entire product line is consumers have been conditioned for decades to have certain expectations (namely that the company with the best single product will often have the best product lineup), and some companies use that conditioning against consumers to the point where halo products have lost their use as an indicator.

I personally will never ever buy a non-halo product because the halo product from a company is great- I am going to research the product I will actually buy and purchase it based on its merits. The only use I have for halo product reviews when I don't intend to buy that product is to get a general sense of where a market is and where it is going.

In fact the only personal use for halo products (that I don't want to buy) I can think of is they are a good way to explain relative value down the road. For example "the GTX 1070 is as fast as a 980 ti" is a much more convincing argument than "the GTX 1070 is ___% faster than the GTX 970" because it gets across the point that YOU Mr. non-halo buying average person gets to taste the same fruit that the 1% once tasted and bragged about. Heck I will admit I bought a GTX 970 I really didn't need because "it was as fast as a Titan OG," even though the only reason I was enamored with the Titan OG was because Nvidia stuck that stupid $1000 price on it. We are programmed as humans to want what we can't have, even if what we can't have isn't that great to begin with.

This is what I've been saying for god knows how long. Nvidia doesn't do this by accidently, they pick those stock clocks on the 1070 on purpose to be at the 980Ti level. They could have put it above and made it faster, or slightly slower, but they put it EXACTLY at the 980Ti level, so that you too can get a taste.

Nvidia
-We've known for AWHILE now that a 1070 is as fast as a 980Ti.

AMD
-We know an RX 480 is as fast as a.... GTX 1080 but isn't because if it was it would be literally the most insane deal ever but if that's the case then why is it matching a GTX 1080 in crossfire mode, but it's 50% utilization, so it's as fast as a GTX 1080? but wait, then why jump through these hoops, maybe they are using some type of framerate targeting? That also explains the efficiency... blah blah balh.

This is why AMD fails. Even after a presentation about their GPU launching it, they can't deliver a solid performance expectation.... Worst marketing department ever. Just use a magic 8 ball seriously.

 

[Olecki]

Ill tell ya, a good car analogy is hard to make

Much better car analogy is motorsport in Europe. Companies pumping A LOT of money into WRC series, where models have same name as retail model but they share maybe few body parts. Volkswagen Polo R WRC isn't really same product as VW Polo, but success in motorsport is huge advertisement.

 

[biffosaur]

Ill tell ya, a good car analogy is hard to make

I would have gone with the Hellcat.

Dodge sells a lot more v6 challengers than it does Hellcats (or before that, SRT8's) because people say "hey, it looks a lot like a Hellcat so it's cool to drive even if it's not that fast"