AMD ships over 25,000 Radeon RX Vega graphics cards

[DocSwag]

44 minutes ago, Mr.Meerkat said:

Ermmm...if we look at the trends for the RX480 or 1080 or 1070 or well really any card that has reference then aftermarket, you'd find it takes like at least a month before aftermarket cards start coming out so...not surprised?  

It was different with some others like the 1060 and 1080 ti iirc

[samcool55]

10 hours ago, Jurrunio said:

I guess someone in AMD must have regretted using HBM2 on Vega. Expensive and limited production of this hurts this company which desperately needs more money.

They had to. GDDR5 would make it even more power-hungry and slower.

Vega and Polaris are both limited by VRAM speed and even tho HBM2 is the fastest thing out there, it's still not fast enough.

 

Also AMD hoped HBM2 would be used more by now but atm AMD is still the only company that buys HBM2 in big numbers so if they order a lot of HBM2 the price for it will skyrocket. The only solution is waiting for Nvidia to start selling Tesla P100 cards, they also use HBM2 and that would fix a lot.

[AresKrieger]

16 minutes ago, DocSwag said:

It was different with some others like the 1060 and 1080 ti iirc

Also due to the HBM mess the margins are tight on these cards so expect either markups to continue or selection to be limited on the aftermarket cards unless the HBM situation changes

[AresKrieger]

Eh as already stated that isn't that many and given how they are popular with miners it would seem this launch could be seen as a failure from a gaming market re-entry perspective, they need to improve their designs and it would be wise to avoid pricey hardware that nets little benefit such as HBM 2

[Sauron]

This reminds me of the hawaii mining situation. Hopefully AMD's production catches up and we can get vega for a reasonable price again in a few months.

5 hours ago, leadeater said:

So if AMD sold 3.8 mil units total in Q3 2016 that is 316,666 units per week. 25,000 (10,000 per week) isn't much at all then, 3.15% of total sales assuming same units shipping now as then.

That's true, but vega is only a small part of their lineup (V 56 hasn't even launched yet so for now they are only selling the more expensive 64) and it seems they ran out of stock, so this is still pretty much best case scenario for them.

[dexT]

Not really an issue with all things considered. 

 

Do you think Nvidia sells more Titan Xp than 1050?

[AresKrieger]

16 minutes ago, dexT said:

Not really an issue with all things considered. 

 

Do you think Nvidia sells more Titan Xp than 1050?

No but it should be noted they sold more 970s than anything else and the 56 is in that class level for this gen (though the price of that level is a little elevated compared to the previous gen)

[JediFragger]

A whole 25,000!!!??? Bet nVidia are shitting it

[dexT]

47 minutes ago, AresKrieger said:

No but it should be noted they sold more 970s than anything else and the 56 is in that class level for this gen (though the price of that level is a little elevated compared to the previous gen)

I bought a 970 and sent it back for a 980 after RAMgate as many did lol. I wonder if they deducted 970 sales and added to 980 or not.

[Doobeedoo]

Yeah so shipped not = to sold. Either way, I really did expect it to be higher. But on the other hand I don't even see RX Vega at my retail at all. I know it's 2.5 weeks only.

Also, people may be waiting prices to normalize and custom cards as well.

[MrRavens]

Though AMD needs to rework some stuff here and there, I respect their hustle and grind to keep pushing. 

[Jurrunio]

5 hours ago, samcool55 said:

They had to. GDDR5 would make it even more power-hungry and slower.

Vega and Polaris are both limited by VRAM speed and even tho HBM2 is the fastest thing out there, it's still not fast enough.

 

Also AMD hoped HBM2 would be used more by now but atm AMD is still the only company that buys HBM2 in big numbers so if they order a lot of HBM2 the price for it will skyrocket. The only solution is waiting for Nvidia to start selling Tesla P100 cards, they also use HBM2 and that would fix a lot.

GDDR5 is much cheaper and supply is much greater. Not like Vega is power efficient with HBM2 anyway.

How do you know new AMD architectures are limited by VRAM speed? Even the faster 1060 6gb has lower transfer rate than a rx 470. It should be enough

[ARikozuM]

8 minutes ago, Jurrunio said:

How do you know new AMD architectures are limited by VRAM speed? Even the faster 1060 6gb has lower transfer rate than a rx 470. It should be enough

I believe AMD stated that as their reasoning behind its use. 

[LAwLz]

15 minutes ago, ARikozuM said:

I believe AMD stated that as their reasoning behind its use. 

Are you sure about that? I don't think AMD said it. My guess is that it's something being parroted on Reddit.

[ARikozuM]

1 minute ago, LAwLz said:

My guess is that it's something being parroted on Reddit.

Could be. Some [on this forum] have said AMD, others have said der8auer. 

[samcool55]

54 minutes ago, Jurrunio said:

GDDR5 is much cheaper and supply is much greater. Not like Vega is power efficient with HBM2 anyway.

How do you know new AMD architectures are limited by VRAM speed? Even the faster 1060 6gb has lower transfer rate than a rx 470. It should be enough

It's thanks to HBM2 the gpu has an extra 30-40W of headroom. The memory controller is much more efficient and the VRAM itself is also more efficient.

And we know Polaris and Vega are limited by VRAM speed because when you overclock the GPU on polaris or Vega, performance barely goes up.

However if you oc the VRAM performance improves a LOT, it's almost linear. (10% higher VRAM clocks = almost 10% more performance)

 

I mean there's a reason why vega in power efficient mode and turbo mode barely improves performance wise (2-3%) even tho the GPU eats up a lot more power. HBM clocks stay the same in all modes btw.

 

No idea why Nvidia performs that well with "little" VRAM speed tho. There's probably an architectural related explaination but i'm not smart enough for that.

 

edit @LAwLz you should give this a read too

[Jurrunio]

2 minutes ago, samcool55 said:

It's thanks to HBM2 the gpu has an extra 30-40W of headroom. The memory controller is much more efficient and the VRAM itself is also more efficient.

And we know Polaris and Vega are limited by VRAM speed because when you overclock the GPU on polaris or Vega, performance barely goes up.

However if you oc the VRAM performance improves a LOT, it's almost linear. (10% higher VRAM clocks = almost 10% more performance)

 

I mean there's a reason why vega in power efficient mode and turbo mode barely improves performance wise (2-3%) even tho the GPU eats up a lot more power. HBM clocks stay the same in all modes btw.

 

No idea why Nvidia performs that well with "little" VRAM speed tho. There's probably an architectural related explaination but i'm not smart enough for that.

I did a overclock test with my RX 470 back when I got it because I want to know whether I should let it run overclocked all the time (why bother with online reviews when I can do it myself). Results are disappointing with the Witcher 3, Watch Dogs 2 and Sniper Elite 4 though. Bumped the core clock by 100MHz to 1330MHz, about 5% increase in FPS (I turned graphics settings all the way up so my CPU wont bottleneck). Removed the core overclock, bumped memory clock by 200MHz to 2400MHz (7200MHz effective), next to no increase in FPS (I'm not even sure whether the increase really exists, or I just hope it exists). At the same time, power draw reported from Afterburner increased from just over 90W to over 120W. Eventually, I underclocked the core to 1050MHz instead since its reported power draw is only 70-80W but I only lost about 3-5% of fps in games.

 

I dont know how power efficient mode and turbo mode works in Vega, but its well-known that Vega is clocked way over its efficient range.

[leadeater]

1 hour ago, samcool55 said:

No idea why Nvidia performs that well with "little" VRAM speed tho. There's probably an architectural related explaination but i'm not smart enough for that.

Nvidia has extremely good memory compression so effective bandwidth is much higher than what AMD has, Anandtech has an article showing the difference if you're interested.

[leadeater]

6 hours ago, JediFragger said:

A whole 25,000!!!??? Bet nVidia are shitting it

Plot twist Nvidia brought all 25,000 of them to 'help' them out 

[AlTech]

15 hours ago, VanayadGaming said:

in 2.5 weeks guys, I find this pretty ok. Also, Vega for gaming while bad, for rendering is amazing. They even have those SSG cards for 8k timeline scrubbing.

Nobody said VEGA was bad for gaming. You misunderstand VEGA's problem.

 

VEGA is not bad for gaming. It's actually good for gaming. People did not expected it to be just good for gaming. They expected unrealistic things out of VEGA and thus they were disappointed.

 

At MSRP, it's a good GPU. Above MSRP, you're money would be better spent elsewhere.

14 hours ago, Windspeed36 said:

Keep in mind that doesn't actually mean they've sold them; just that they've been shipped to end points (retailers & system integrators) or have simply been sold to distributors. 

From the sounds of things, not a lot of people have been buying VEGA in the Radeon bundle packs. They haven't been selling out. Only the individual cards (sold as is) are selling out. So much for trying to stop miners snatching up all the cards.

[LAwLz]

7 hours ago, samcool55 said:

It's thanks to HBM2 the gpu has an extra 30-40W of headroom. The memory controller is much more efficient and the VRAM itself is also more efficient.

And we know Polaris and Vega are limited by VRAM speed because when you overclock the GPU on polaris or Vega, performance barely goes up.

However if you oc the VRAM performance improves a LOT, it's almost linear. (10% higher VRAM clocks = almost 10% more performance)

Would like to see some benchmarks for this. I haven't read many Vega reviews but the ones I did could barely overclock Vega at all, so I haven't seen anyone do a comparison of the difference between overclocking only the memory, vs only the core.

 

I don't agree that there isn't any difference between the different modes though. According to TechPowerUp's tests (which are based on 20 different games) the difference between at 4K between the 2nd BIOS in Power Save and the 2nd BIOS in Turbo is 9.8%.

That's the test where there is the biggest difference.

 

At 1440p the difference is 8.7% and at 1080p the difference is 5.3%.

Assuming what you said is true (that the memory is kept the same when choosing the different modes), then it doesn't really make sense to say that Vega is memory starved since a faster core clock helps more at higher resolutions than at lower resolutions.

 

 

7 hours ago, samcool55 said:

No idea why Nvidia performs that well with "little" VRAM speed tho. There's probably an architectural related explaination but i'm not smart enough for that.

Like leadeatersaid, Nvidia is way ahead of AMD in terms of compression.

 

 

5 hours ago, AluminiumTech said:

VEGA is not bad for gaming. It's actually good for gaming. People did not expected it to be just good for gaming. They expected unrealistic things out of VEGA and thus they were disappointed.

 

At MSRP, it's a good GPU. Above MSRP, you're money would be better spent elsewhere.

I don't agree.

Even at MSRP the price:performance would be somewhere around the 1080, which is quite old at this point. And even if it was the same the Vega would be more noisy and draw hell of a lot more power. That does not seem like a "good for gaming" card.

[Kamina]

25,000 seems rather low.

[Agosto]

9 hours ago, AluminiumTech said:

From the sounds of things, not a lot of people have been buying VEGA in the Radeon bundle packs. They haven't been selling out. Only the individual cards (sold as is) are selling out. So much for trying to stop miners snatching up all the cards.

You know, they're much more expensive than MSRP and verey few people will buy Vega+1700X/1800X+selected mobo+2 games only to get those games for free, while actually paying a bit more for the expensive motherboards. Not counting the freesync monitor discount, since it's limited so a few countries and looks like it's valid only for one model by Samsung. For example there are no packs at all in Italy, even those without monitor, and only 2 standalone V64 cards for €680.

4 hours ago, LAwLz said:

Assuming what you said is true (that the memory is kept the same when choosing the different modes), then it doesn't really make sense to say that Vega is memory starved since a faster core clock helps more at higher resolutions than at lower resolutions.

Core clocks on Vega obviously help but HBM overclock has an high impact on performance and much less on power consumption

[LAwLz]

14 minutes ago, Agost said:

Core clocks on Vega obviously help but HBM overclock has an high impact on performance and much less on power consumption

Well, the person I responded to, whom specifically tagged me and told me to read the post, said that there is a 2-3 performance difference when there is clearly more.

 

Both you and that person has since said that there is a big difference when the memory is overclocked but so far I have seen no evidence of this. Like I said before, the only reviews I have seen struggled to overclock the card at all, and I have not seen any review compare only memory overclock vs only core overclock so I would like to see that before I buy into those arguments.

 

Not saying that the arguments are wrong, but I won't believe them without evidence. Especially not when the person who told me that is quite a bit off regarding the performance difference between power profiles.

[leadeater]

17 minutes ago, LAwLz said:

Both you and that person has since said that there is a big difference when the memory is overclocked but so far I have seen no evidence of this. Like I said before, the only reviews I have seen struggled to overclock the card at all, and I have not seen any review compare only memory overclock vs only core overclock so I would like to see that before I buy into those arguments.

If you haven't seen it Gamers Nexus does the best job of all the reviewers on covering overclocking of Vega, 56 only last I looked. They did a lot of combinations so it's hard to say if they specifically did pure memory overclocking but the data should be there, I skipped through it a lot since the video is EXTREMELY long. They did show memory overclocking does give a decent performance increase but they were also overclocking the core clock at different amounts too, not sure if there is a stock core + memory OC result in there but I'd be surprised if there isn't.

 

TL;DR You can overclock Vega 56 quite well but only through a lot of effort due to broken utilities, might be fixed in 17.8.2, and undervolting a fair amount while also increasing the maximum power limit. Vega 56 mostly hits the imposed power limit of the firmware though but that kinda isn't a good thing if you know what I mean, so it could OC higher.

 

Edit:

Oh and Vega 56 memory OC limit is between 980 and 1000 Mhz, the 1000Mhz is only if you're extremely lucky and was more alluded to as possible in the Gamers Nexus video.