Radeon RX Vega 64 vs Radeon R9 Fury X Clock for Clock @ [H]

If you believe the whitepapers, there's a lot of latent tech buried in Vega. I think it's going to come down to which of two camps is correct: 1) RTG's driver team is going to unlock this latent tech transparently for existing titles allowing 'full' Vega to be experienced, or 2) Game developers are going to have to code explicitly for this latent tech for 'full' Vega to be experienced.

As far as power draw is concerned, not surprising at all. RTG tried to clock it as high as it could so it could trade blows with a 1080. It's no secret that a chip running 'within design parameters' is going to run more efficiently than a chip clocked balls to the wall.
 
Algrim said:
If you believe the whitepapers, there's a lot of latent tech buried in Vega. I think it's going to come down to which of two camps is correct: 1) RTG's driver team is going to unlock this latent tech transparently for existing titles allowing 'full' Vega to be experienced, or 2) Game developers are going to have to code explicitly for this latent tech for 'full' Vega to be experienced.

As far as power draw is concerned, not surprising at all. RTG tried to clock it as high as it could so it could trade blows with a 1080. It's no secret that a chip running 'within design parameters' is going to run more efficiently than a chip clocked balls to the wall.
Click to expand...
It will rely on game Devs to program. Bethesda, so far, has been the only major PC gaming company to go all in on Radeon, so their future games will be the ones to watch to see how much of a difference that makes.
 
Thanks for the review, [H]. This really puts things into a true perspective.



I can't help but think that things went a little something like this...

"What's the status of the Fury rehash?"

"Fury v1.5 is on track for the targeted release date."

"Any significant improvements, other than the higher clocks and more VRAM?"

"Honestly, not really...we've hit a plateau with the performance, so we're looking for ways to maximize clock speeds, while praying that the power draw and heat output don't cause it to nuke itself."

"*Sigh* The marketing team is going to have to work overtime again to sugarcoat it...more of our usual, it seems. Dang, I need to call the Ferrari dealer to get my car in for maintenance."

"Ummmm, OK...so yeah, that's where we are with all this."

"I'll have my assistant call the dealer. What kind of car do you drive?"

"Well, I'm on my fourth Toyota. Big step up from my first car."

"What kind of car was your first?"

"It was on old Chevy Vega. It was rather boring and broken down all the time. I was so happy the day I finally got rid of it!"

"Hold on...that's great! Let's call the new Fury a VEGA!"

"You want to name it after one of GM's shittiest cars ever?"

"Well, yeah, but it sounds cool. Let's get the marketing team on the horn, they're going to love this!"
 
Thank you for your efforts. Very enlightening and kind of what I guessed based on your previous tests.
 
This was a neat article. Any chance of seeing a 1080 vs 980 type comparison, even abridged?
 
asyhvv.jpg
 
Thanks for testing this out, its always interesting to see this kind of comparison.
 
Yeah so without all those "where is it?" features, its just Fiji. doubt we will see any major improvements with those "where is it features" enabled as well, really looks like those features were just a different way to market Vega at this point.

Power consumption seems to have fallen short too, at the same performance level, GF's 14nm should have given them a 60% improvement in perf/watt, that didn't happen, but then again hard to compare w/ cause of the water cooling of Fiji, might have hit that, not sure.
 
I'm quite amazed that given the length of time spent in development, with delay after delay, with AMD marketing promising some kind of "revolution", that this was the result. Granted, the number of stream processors between the two remained the same, but there's something to be said for taking better advantage of the hardware you have to work with. I think it's clear here that AMD, or the game developers, or both, are not doing so. Remarkably disappointing at best.
 
lazz said:
This was a neat article. Any chance of seeing a 1080 vs 980 type comparison, even abridged?
Click to expand...

I'm also curious to see this as well, given what we saw here. Not to derail the topic at hand, but this comparison has my vote. It would be interesting to see given that both marketing teams touted that the change in architecture was a big deal.
 
  • Like
Reactions: lazz
like this
PontiacGTX said:
If the clocks fluctuated like this? how many seconds did it run at 1050MHz vs Fury X?
Click to expand...
I am not sure of what you are asking. The average in that specific game was 1049MHz as noted. We qualified every single game to do this. As noted, the clock is dynamic for Vega.
 
I'd say I somewhat dissapointed by Vega. I mean, it seems AMD should just did a die shrink and slap hbm2.
 
PontiacGTX said:
Those features arent being used properly. HBCC cant shown its purpose in games until memory allocation is different
DSBR was mainly to save power/bandwidth due to ROPs/Rasterizer requirements,Games arent adapted for Primitive shader pipeline, RPM isnt used in any game yet and also async compute isnt being in the way VEGA can do it

If the clocks fluctuated like this? how many seconds did it run at 1050MHz vs Fury X?, using a tool to force all 3D Power states at least 1050MHz wouldnt be more accurate?
1505156973f8us74pr9b_2_1.png
Click to expand...
Even if there were minor variations, it doesn't seem like the end resultd would be different.
 
I love these type of comparisons from one generation to the next. Reminds me of the old wavy Dave ones at beyond3d.

Was there any reason not to just OC the HBM2 on Vega to 1000 to match. Bandwidths?

Disappointed in not seeing much gain on the architecture for gaming. I don't think AMD indicated there will be much performance gains with future drivers.

Also had problems with 17.9.1 as well.
 
PontiacGTX said:
exactly based on the power saving features used to reduce clock speed in load, wouldnt it change per game basis(and within the game it woudl vary with the load ) and in some cases it wouldnt be always near the target clock speed, losing performance meanwhiel FURY X mostly is running at 1050MHz since doesnt have dynamic such clocks, also it is running underclocked, there should be a variation based on the power limit set on the power tables in BIOS-power states, some older GCN based cards have slighly worse clocks/load if the power was below the power limit target

just wondering if it made a bigger difference on GPU clock
Click to expand...
Did you fully read page 2?
 
PontiacGTX said:
exactly based on the power saving features used to reduce clock speed in load, wouldnt it change per game basis(and within the game it woudl vary with the load ) and in some cases it wouldnt be always near the target clock speed, losing performance meanwhiel FURY X mostly is running at 1050MHz since doesnt have dynamic such clocks, also it is running underclocked, there should be a variation based on the power limit set on the power tables in BIOS-power states, some older GCN based cards have slighly worse clocks/load if the power was below the power limit target

just wondering if it made a bigger difference on GPU clock
Click to expand...
I think you're trying too hard to draw something positive for Vega here.
 
PontiacGTX said:
Those features arent being used properly. HBCC cant shown its purpose in games until memory allocation is different
DSBR was mainly to save power/bandwidth due to ROPs/Rasterizer requirements,Games arent adapted for Primitive shader pipeline, RPM isnt used in any game yet and also async compute isnt being in the way VEGA can do it

If the clocks fluctuated like this? how many seconds did it run at 1050MHz vs Fury X?, using a tool to force all 3D Power states at least 1050MHz wouldnt be more accurate?
1505156973f8us74pr9b_2_1.png
Click to expand...

If you are talking specifically about the above game, that's 9 minutes worth of gameplay data. I think our longest run-through is around 14 minutes. We verified that each game was as close to 1050MHz as possible, a painstaking process I might add.

The 3D Power states don't work like you describe. The only two that matter are the last two, if you are manually overclocking. The others are ignored when you manually set the last two power states.

Both Sniper Elite 4 and Gears of War 4 support Async just fine on Vega, and it works, we had both enabled for testing.
 
Algrim said:
If you believe the whitepapers, there's a lot of latent tech buried in Vega. I think it's going to come down to which of two camps is correct: 1) RTG's driver team is going to unlock this latent tech transparently for existing titles allowing 'full' Vega to be experienced, or 2) Game developers are going to have to code explicitly for this latent tech for 'full' Vega to be experienced.

As far as power draw is concerned, not surprising at all. RTG tried to clock it as high as it could so it could trade blows with a 1080. It's no secret that a chip running 'within design parameters' is going to run more efficiently than a chip clocked balls to the wall.
Click to expand...

For certain Rapid Packed Math has to be implemented by the developer. If games never use it, it'll never be advantageous. I know Far Cry 5 is going to use it, the question is, will other games?
 
noko said:
I love these type of comparisons from one generation to the next. Reminds me of the old wavy Dave ones at beyond3d.

Was there any reason not to just OC the HBM2 on Vega to 1000 to match. Bandwidths?

Disappointed in not seeing much gain on the architecture for gaming. I don't think AMD indicated there will be much performance gains with future drivers.

Also had problems with 17.9.1 as well.
Click to expand...

It was easier to downclock Fury X, than try and overclock Vega 64 that high and introduce instability. I did not need that complication.
 
PontiacGTX said:
exactly based on the power saving features used to reduce clock speed in load, wouldnt it change per game basis(and within the game it woudl vary with the load ) and in some cases it wouldnt be always near the target clock speed, losing performance meanwhiel FURY X mostly is running at 1050MHz since doesnt have dynamic such clocks, also it is running underclocked, there should be a variation based on the power limit set on the power tables in BIOS-power states, some older GCN based cards have slighly worse clocks/load if the power was below the power limit target

just wondering if it made a bigger difference on GPU clock
Click to expand...

Which is exactly why we tested clock speed on a per-game basis, and adjusted as necessary to get us to the target.
 
PontiacGTX said:
Those features arent being used properly. HBCC cant shown its purpose in games until memory allocation is different
DSBR was mainly to save power/bandwidth due to ROPs/Rasterizer requirements,Games arent adapted for Primitive shader pipeline, RPM isnt used in any game yet and also async compute isnt being in the way VEGA can do it
Click to expand...


I really think its marketing BS dude everything that separates Vega from prior GCN uarch is turned off or not usable with current games or close to be released games, drivers aren't made that way, drivers are done before the chip is taped out and if those features are turned off, there is a reason why they are turned off.

Async compute yeah don't even need to go there, we know it needs to be tailored per generation, per GPU within each gen too. So dev's need to do the work.

RPM is dev controlled again.

This is AMD for you, let everyone else other themselves do the work to make their products look good.

Its the same thing they did with LL API's, the mess with async compute (concurrent workloads), more work for dev's (although yeah its good in the long run, just not right now), now you have Vega.

That is not how things go in the industry and that is also why adoption rates suffer, if its more work, it a slower adoption rate automatically.
 
Last edited:
I think we'd all LOVE it if AMD pulled a rabbit from a hat with a driver release and the Vega WTFPWND the 1080s. I know I would.

Just doesn't seem super likely. <shrug>
 
Brent_Justice said:
For certain Rapid Packed Math has to be implemented by the developer. If games never use it, it'll never be advantageous. I know Far Cry 5 is going to use it, the question is, will other games?
Click to expand...


It also depends on what its used for. RPM so far from what AMD has shown so far, and what has been talked about, can help it with vertex set up, lessening the burden on AMD chips, and helping with polygon throughput. Higher level shaders that require higher precision, like lighting, not sure if it can be used for that at this point. So performance increase with such a thing will only show up when that is the limiting factor. (lighting, AO, GI, and other techniques use the vast majority of the processing power)
 
"This is fun, right?"

Yes it is ! Great article ... some fine, and from another point of view, in-depth investigative [H] journalism.
There seem to be more of these interesting articles lately with a different take (like comparison Vega56/64 to Fury/X, Threadripper installation, ...), is this a trend ? (or just because it are exiting times hardware wise, and my memory doesn't go back more than a couple of months)
Some time ago, if I remember correctly, you mentioned the passion was somehow getting lower ? But it seems it's back, certainly when you guys do a review "just because we wanted to know" ... that is what it's all about :)
(and to put these results in perspective, a similar comparison with competing Geforce cards would be great indeed)
 
I'm still scratching my head on why they used HBM2 but at half the bus width. So they had to delay the card due to delayed HBM2 just so they could get less memory bandwidth? I don't understand it at all, and from the somewhat limited memory oc stuff I've seen it seems to be crippling the card to a degree.
 
Thanks for the article guys! It sounds like getting everything working just so was a PITA.

It makes me wonder WTF AMD was up to during the entire development of Vega. Were they simply adding pipelining registers to the logic to allow it to clock higher? You guys pretty much summed it up in this quote:

When you look at all the above architecture changes and benefits over previous high-end AMD GPUs you have to ask yourself where are those and why aren’t they making a bigger difference? Either the features are broken, turned off, not working, or its advantages were simply over-marketed. As it is right now, the end result, is that the clock speed advantage of Vega is what is holding it up, but actually holding it up quite well. Clock speed has long been king in the world of gaming and that seems to still be an incredibly important factor today.
Click to expand...
 
bobzdar said:
I'm still scratching my head on why they used HBM2 but at half the bus width. So they had to delay the card due to delayed HBM2 just so they could get less memory bandwidth? I don't understand it at all, and from the somewhat limited memory oc stuff I've seen it seems to be crippling the card to a degree.
Click to expand...

All about power on that one, they simply needed the power savings of HBM2 to run the GPU clock so high, otherwise board power would be through the roof.
 
Brent / Kyle, thanks for the clock for clock review. Did you by chance also monitor how much power each card was using at the Fury-X clocks? I am assuming the Vega pulls less?
 
Thank you for writing up this review Brent. Same clock analysis was exactly what I asked for in the thread for the lasy fury vega comparison which oddly people sort of laughed off at the time. Cutting through the generic and offering variation to the world of measuring fps will continue to differentiate hardocp from less apt sites. I believe your conclusion of brute force is a rather apt summary of the Vega launch. The relatively identical clock to clock numbers really underlines the lack of measurable innovation in the architechture that has been counted on for so long. With node shrinks being more engineering intensive, I can see AMD continuing to tow the line of brute force and delayed gpu releases. I'm not sure if this should be a celebration of how successful Vega is going to be, which essentially any mid to upper end gpu would be successful in this current environment, but more of a kudos to the Fury architecture and how much more potential was left on the table.
 
You must log in or register to reply here.
Back
Top