• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Radeon RX Vega thread

err my 290x does 14K graphics score @ 24h/7 clocks... yours 10k only?

29% faster? Is your R9 290X clocked somewhere around 1200MHz? I'm just guessing.

That appears to be an R9 390 at 1020MHz, it's a refresh of the R9 290 which has 2560 cores versus the R9 390X and 290X's 2816 cores, at the same clock-speeds those cards should be around 10% faster, coupled with higher clocks I can see why yours can get a 14K graphics score and theirs gets 10.7K.
 
Vega 56 to 64 Bios Mod

so i flashed that 64 bios on my 56 and it' seems to work fine. no new CUs but hopefully power limit should now be at 64s level. HBM seems to be further overclockable also (which is kinda strange...). you can find some quick results here:

56-64resultsula5a.png


repeated that test with fire strike:

firestrikeresults9esxm.png


has to be said, that the 56 results were preformed on the slave bios, as the 64 bios was still on the master. what's the difference? im not quite sure. state 6 & 7 frequencies and voltages seem to be the same.

the overclocked and undervolted results seem to scale a bit better in fire strike.

the main difference between the bioses seems to me, that you can clock the HBM to 1100 MHz on the 64 bios, where it crashes over 960 MHz on the 56 bios. probably something to do with an internal voltage controller that you can't change in wattman or other tools.

as for power comparisons between the two bios'es ... i hope i can preforme those later this week if i can get a grip of a new ampere-meter.


as for the bios tweak. im not sure if it's worth it, as you can get most of the performance gains on the 56 bios while doing an signifcant undervolt and slightly OCing. doing so, you can cap the fan speeds at reasonable levels and won't be much distracted while game sound is running.
 

dr_rus

Member
how a card limited by it's power target to a physical draw of 210W should draw 300-350 W while beeing at MINUS 24% PL and preforming a third less of it's rated clocks (1000MHz) is beyond me. maybe your math is not quite correct? people probably were comparing total system draw from wall to his results..

Ding ding ding, as in power draw at the wall. And his results seem to be just some numbers from HWiNFO64 which generally mean nothing.
 

spyshagg

Should not be allowed to breed
29% faster? Is your R9 290X clocked somewhere around 1200MHz? I'm just guessing.

That appears to be an R9 390 at 1020MHz, it's a refresh of the R9 290 which has 2560 cores versus the R9 390X and 290X's 2816 cores, at the same clock-speeds those cards should be around 10% faster, coupled with higher clocks I can see why yours can get a 14K graphics score and theirs gets 10.7K.

Mine is modded with the 390 bios and tighter ram timings. It does 14K @ around 1150mhz / 1460mhz ram. Its 24/7 settings, but there are guys doing 16K easy for benches only. The 290x was a beast and still manages to really good (new life under the new bios and ram editing)
 
D

Deleted member 17706

Unconfirmed Member
The more reviews I read the more I realize how good is Maxwell/Pascal architecture...

Yeah, I don't think the situation for Vega is going to change even with the aftermarket cards. They are good for the price (assuming MSRP) if you care about FreeSync. Otherwise, you're better off going Nvidia (assuming MSRP).
 

Marmelade

Member
29% faster? Is your R9 290X clocked somewhere around 1200MHz? I'm just guessing.

That appears to be an R9 390 at 1020MHz, it's a refresh of the R9 290 which has 2560 cores versus the R9 390X and 290X's 2816 cores, at the same clock-speeds those cards should be around 10% faster, coupled with higher clocks I can see why yours can get a 14K graphics score and theirs gets 10.7K.

His R9 390 score is on the low side
With my old R9 290 Tri X @1050MHz (just a +50MHz on core, RAM @stock) I could do 12k+
 

MimiMe

Member

Review down.
Yesterday we published the ASUS Radeon RX 64 STRIX review. As shown, it performs awfully similar towards the reference Radeon RX 64. This morning I received a phone call from ASUS, asking us if we’d be willing to take down the article for a few days as they have made a mistake.

The sample we received did not get a final BIOS for its final clock frequencies and fan tweaking. Ergo, the sample we received carries a default reference BIOS.

It’s a colossal mistake, but as such the end-results in the review are not representative enough for the final product. ASUS will get the finalized BIOS over once they have finished (likely a day or two) after which we will re-test the card with that final BIOS and thus republish the review. All this explains why the STRIX card was so incredibly close to Vega 64 performance.

Apologies for the inconvenience, but this mistake was not one coming from us.


Computerbase (German) preview was quite fond of the Strix.
 
I came across a new video from Hardware Unboxed where they undervolt and overclock the RX Vega 56, the performance is quite interesting.

Undervolt & Overclock, Radeon RX Vega 56 vs. GeForce GTX 1070 (YouTube video)

Here are some of the performance results:





Mine is modded with the 390 bios and tighter ram timings. It does 14K @ around 1150mhz / 1460mhz ram. Its 24/7 settings, but there are guys doing 16K easy for benches only. The 290x was a beast and still manages to really good (new life under the new bios and ram editing)

Oh I see, that's interesting.
Yeah the R9 290 and 290X have really great overclocking capabilities, I was really impressed to see that people were able to adjust the ram timings and get more performance out of the GPUs!

His R9 390 score is on the low side
With my old R9 290 Tri X @1050MHz (just a +50MHz on core, RAM @stock) I could do 12k+

Oh I see, that's cool!
 

LordCiego

Member
Seeing prices here in Spain and Vega is not interesting right now, the cheapest I found a Vega 56 is at 499€, for 20€ more you can get a Gtx1080 new.
 

thelastword

Banned
I came across a new video from Hardware Unboxed where they undervolt and overclock the RX Vega 56, the performance is quite interesting.

Undervolt & Overclock, Radeon RX Vega 56 vs. GeForce GTX 1070 (YouTube video)

Apart from the Guru 3d Strix review, the hardware Unboxed video is not entirely accurate either..He didn't do the UV properly, so he is looking things over...Only 930MHz on the HBM was a big no no...obviously there are many ways to do a UV, but with Vega, lowering the temps with less voltage and higher fan rpms is good enough to stabilize core clock speeds to at least boost levels...where it won't fluctuate..... Of course, you can give a small uptick to the core as well, if you wish, but generally, that's how it's done.
 
as promised i've done the power testing for the benchmarks from two days ago:

firestrikeresults9esxm.png



the following is system power draw from the wall. as stated before i used a stock ryzen r5 1600 with a TDP of 65 W. idle system power draw was around 110 W.

56-64resultspower62jg6.png


to understand what that actually mean, we need to make some assumptions. we assume a constant efficiency of the PSU of 80% and a constant power draw of the CPU plus other non GPU components of 70 W. this means we get 310 x 0,8 = 248 W on the DC side for the stock 56 result which gives an TDP around 180 W (mind this is the slave bios). for the OCed result with the 64 bios (which means a significant higher power limit) you get 518 x 0,8 = 414 W on the DC side, which should give you a GPU draw of around 340W.


to put that in some context, i divided the firestrike results by the system power draw, which leads to the following

56-64resultsefficienc0qk6b.png


as you can see the maxed 64 bios isn't a good compromise between perf and power (and therefore thermal output).
 
D

Deleted member 17706

Unconfirmed Member
Yay! The latest drivers fixed the issue causing your mouse/machine to lock up for seconds at a time when loading images/video in a browser that has GPU/hardware acceleration enabled. One of the biggest issues I've had with the card since getting it! Very pleased.
 
man he really needs to learn how to make graphs.

also the statement, that direct Input of gpu clocks does not work is wrong. if you change the p6 and p7 clocks the card will still circle thorugh states p1-5. putting in percentages will also change the lower than p6 power stages. that's the difference why he's not expierencing positive perf changes when applying just frequencies. the air cooled card is on p5 most of the time when the PL is set to +50% and OCed. (what doesn't seem to work porperly is min/maxing p6 and p7).

Living the dream, you.

i feel kinda bad about this.
 

TVexperto

Member
Just ordered the Vega 64, hope its a good upgrade from the RX 480 8GB. I wanted a GTX 1080 but my monitor supports freesync so i wasnt sure :(
 

llien

Member
I came across a new video from Hardware Unboxed where they undervolt and overclock the RX Vega 56, the performance is quite interesting.

Something is messed up with undervolting bit.
For starters, I'd like to see undervolted vs stock, which is nowhere to be seen.
In other reviews, undervolted 56 was able to challenge stock 64/1080FE.
 

dragn

Member
Something is messed up with undervolting bit.
For starters, I'd like to see undervolted vs stock, which is nowhere to be seen.
In other reviews, undervolted 56 was able to challenge stock 64/1080FE.

uv really is buggy, i just got 925mv and the gpu draws 10w less than stock (so 0% power target) with 10% more performance. sometimes the voltage doesnt lower itself maybe because of the power states 4-5? dunno

here my steps:

1300mhz 0% pt = 55 fps 165watt from gpu only
1500mhz 50% pt = 62 fps 240watt
1500mhz 50% pt @1000-1050mv = 62 fps 200watt
1450mhz 50% pt @925mv = 60 fps 150watt


the last part with 150 watt doesnt always get active, some times i have to slide the pt target around lol
 
few things you have to know:

1.) your average clocks are higher with the undervolt

2.) hence you gained 10% perf without a rise of 10% in powerdraw but even a reduction

3.) the floor voltage of the 64 is defaulted to 1050mV (950mV for 56) on power states 6 and 7 ...so you effectively won't volt below that point with undervolting
 
it just shows that you are running below 950, but if you measure againt vcore it will min out at 950mV. see buildzoid for further info.
 

TVexperto

Member
I have problems with my Vega 64, I dont understand what the red button means. Does it mean its not getting enough power? Which is weird considering I have 650W....


IwKOgid.jpg


I9Pp1OF.jpg
 

Renekton

Member
https://www.pcper.com/news/Graphics-Cards/AMD-Radeon-Technologies-Group-Raja-Koduri-Goes-Sabbatical

Radeon Technologies Group SVP and Chief Architect Raja Koduri is taking sabbatical from AMD as of today, with a target return date in December. As first reported by our friends at Fudzilla (and also Tweaktown), and that I was able to confirm through AMD this evening, one of our favorite people in the graphics industry will be stepping aside for the time being. AMD CEO Lisa Su will be taking over the Radeon Technologies Group in the interim.
 

The article: Vega 64 vs. Vega 56 Clock-for-Clock Shader Comparison (Gaming)

This is very odd, if their testing is to be believed it looks like the Vega 64 is running into some memory bandwidth limitations or something. In the three games they tested the Vega 56 is offering pretty much identical performance to the Vega 64 which has roughly 14.28% more cores than the Vega 56, 4096 cores versus 3584 cores.

 

joesiv

Member
yeah that is very odd. Some kind of bottleneck I suppose. I wonder what the lower limit is on shader cores to see the performance delta we'd expect.
 

kuYuri

Member
Aren't the Vega AIB cards supposed to show up around now? I'd read the Asus Strix Vega 64 OC was supposed to be early September and we're already in the middle of the month.

Have the other manufacturers revealed their custom designs yet?
 

Taggen86

Member
Just saw the DF review and Vega 64 really is a massive failure. Vega 64 really is a massive failure. How can a 47 percent increase in core clocks and a 40 percent increase in transistors over fury x only result in a 25 percent uplift in performance in most games? Can anyone explain this outcome? Given how inefficient their architecture is, one gets very worried about the next gen consoles and their performance leap over current gen. PS5 and Xbox two will most likely use the follow up to RX480 on a 7nm process (navi) which should result in largely the same flops and performance as vega 56 or vega 64, not very impressive considering what we could expect from next gen GTX 1160 or 1260 (volta or beyond). I really hope AMD switch architecture before Navi. Otherwise, AMD's law of diminishing returns will destroy next gen console performance.

I also want to question the driver argument. based on history and previous AMD cards, can we really expect a dramatic increase in performance over time? Seems like wishful thinking to me. Given the flops of vega 64, we should expect GTX 1080 ti performance, but we are not even close and I doubt drivers will change that.
 

Harlequin

Member
Whether they are going to fix the drivers and whether the drivers are at fault for the lower-than-expected performance increase are two separate questions, though.
 
Just saw the DF review and Vega 64 really is a massive failure. Vega 64 really is a massive failure. How can a 47 percent increase in core clocks and a 40 percent increase in transistors over fury x only result in a 25 percent uplift in performance in most games? Can anyone explain this outcome? Given how inefficient their architecture is, one gets very worried about the next gen consoles and their performance leap over current gen. Next gen will probably use the follow up to RX480 on a 7nm process (navi) which should result in largely the same flops and performance as vega 56 or vega 64, not very impressive considering what we could expect from next gen GTX 1160 or 1260 (volta or beyond). I really hope AMD switch architecture before Navi.

I also want to question the driver argument. based on history and previous AMD cards, can we really expect a dramatic increase in performance over time? Seems like wishful thinking to me. Given the flops of vega 64, we should expect GTX 1080 ti performance, but we are not even close and I doubt drivers will change that.
Sony and MS should try to get a good Nvidia contract for next consoles.How are they going to get 4K games with better than current graphics with a Vega 56 equivalent in PS5 and XB2?.
 

thelastword

Banned
Aren't the Vega AIB cards supposed to show up around now? I'd read the Asus Strix Vega 64 OC was supposed to be early September and we're already in the middle of the month.

Have the other manufacturers revealed their custom designs yet?
There was a Strix review posted around September 5th on Guru3D, it was taken down because they did not have the proper bios wiith the updated clockspeeds and improvements.....Guru3D will have to retest everything when they get that new bios.
 

TarNaru33

Banned
Sony and MS should try to get a good Nvidia contract for next consoles.How are they going to get 4K games with better than current graphics with a Vega 56 equivalent in PS5 and XB2?.

The issue is with the bold, Nvidia is going to demand a higher margin than AMD will, not to mention they do not have an equivalent APU like AMD.

Unless you want really expensive consoles for next gen, it isn't going to happen.
 
The issue is with the bold, Nvidia is going to demand a higher margin than AMD will, not to mention they do not have an equivalent APU like AMD.

Unless you want really expensive consoles for next gen, it isn't going to happen.

Doesn't the Xbox One X have a Polaris equivalent running 4k already? (to some degree). Zen + Vega APU sounds entirely plausible.
 
Top Bottom