IIRC, Pascal was actually added to the nVidia roadmap between Maxwell and Volta. Volta definitely needs a lot of baking, and nVidia was correct in pushing out Pascal because of the sheer insanity in efficiencies.
-
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
- Thread starter PFD
- Start date
IIRC, Pascal was actually added to the nVidia roadmap between Maxwell and Volta. Volta definitely needs a lot of baking, and nVidia was correct in pushing out Pascal because of the sheer insanity in efficiencies.
What are you gaming at? If you're going for higher resolutions, then an upgrade would probably be the best... but I wouldn't bother at 1080p levels since the 480 already performs well at that level.
dr_rus
Member
Pascal was added mostly because they had to launch Maxwell on 28nm while the plan was for Maxwell to be on 20nm and Volta (or whatever is planned next for gaming) on 16nm. With Maxwell launching on 28nm after 20nm failed in GPU space there was a hole in the lineup which they filled with a "Maxwell 16nm shrink" which is what Pascal is.
Yeah, I play at 1080p. Haven't had any problems so far. Though, I've been considering downsampling from 1440p. Also, I might be able to flip the rx 480 for a profit because of the mining craze.
Hardware Unboxed's take on the Vega pricing thingy: https://www.reddit.com/r/Amd/comments/6v5f92/interesting_discussion_from_steve_on_hardware/
Ooops yeah, I forgot about the whole 28nm thing.
If you're flipping it for a profit, then it probably wouldn't hurt to upgrade... if you can get the 56 at a decent price.
Ooops yeah, I forgot about the whole 28nm thing.
If you're flipping it for a profit, then it probably wouldn't hurt to upgrade... if you can get the 56 at a decent price.
Dat Titanium X
scoobs
Member
Volta needs baking? They have production cards out there already in datacenters. Its already a thing.
Unknown Soldier
Member
The current Volta GV100 will bear no resemblance to the consumer Voltas
Coulomb_Barrier
Member
MSRP Increase looks like another made up story by Gibbo:
'JayTwoCents' is still hurt. I think it's more likely someone at Intel/Nvidia slipped him a cheque so he started recommending their hardware instead. The guy is a fool. 'Jay no sense' as usual.
kirankara78
Member
He seems to fail to take into account what retailers are saying though, that the rebate which was offered to retailers has expired, meaning if they want to make money on Vega 64, they have to put up prices. So even though AMD hasn't changed MSRP, they have changed pricing in an indirect way.
kirankara78
Member
Not according to gamer nexus it isnt.
People were thinking they were getting anywhere up to 1900mhz to 2ghz.
That's a 400mhz overclock lol
llien
Member
Welp:
"I've spoken with local retailers and they say their(sic) free to allocate all stock to standalone cards if they want and they say they will do this with shipment that's coming later this week"
Sounds to me like gibbo was trying to justify price hike.
And you seem to be ignoring AMD's deflection/PR i.e why did a rep on overclock.net couldn't make a statement about pricing if it was truly as AMD had presented. Or why AMD needed days to craft PR to address said issue and again why Reps can't give an answer to pricing if things were clear as they want it to seem.
As pointed out by Gamersnexus -
Coulomb_Barrier
Member
Conspiracy!
Locuza
Member
Sebastian Aaltonen is back from vacation with some interesting information:
Yesterday AMD presented Vega at Hot Chips, one slide (probably from Hot Chips) was posted from an Golem employee about the DSBR performance and perf/watt improvements:
https://www.forum-3dcenter.org/vbulletin/showthread.php?p=11468534#post11468534
So for his game there are 27% less flushes on GCN Gen 5 than on 4.
Yesterday AMD presented Vega at Hot Chips, one slide (probably from Hot Chips) was posted from an Golem employee about the DSBR performance and perf/watt improvements:
https://www.forum-3dcenter.org/vbulletin/showthread.php?p=11468534#post11468534
kirankara78
Member
Gamernexus reached out to AMD and got a very garbled unclear pr friendly response.
Radeon RX Vega64 demand continues to exceed expectations. AMD is working closely with its partners to address this demand. Our initial launch quantities included standalone Radeon RX Vega64 at SEP of $499, Radeon RX Vega64 Black Packs at SEP of $599, and Radeon RX Vega64 Aqua Packs at SEP of $699. We are working with our partners to restock all SKUs of Radeon RX Vega64 including the standalone cards and Gamer Packs over the next few weeks, and you should expect quantities of Vega to start arriving in the coming days.
Reading this PR line more carefully, its clear that this is really three separate statements: (1) Demand was higher than expected at the initial launch price, (2) stock is being refilled, (3) Vega64 is one word, not two. Never does the comment indicate a restocking at the launch prices, so we inquired further:
That helps -- thanks for not leaving press & consumers in the dark.
I have a point of clarification on the official statement: The statement notes initial launch quantities and attaches prices to those initial quantities. What is not clear -- to me, at least -- is whether the RX Vega 64 cards will be restocked at a specific price. The statement works around that. Can you confirm for print whether RX Vega 64 single cards (non-bundles) will be restocked at $500?
AMD responded:
Because we cant control pricing, I cant say that.
We inquired again:
A follow-up, then: To what does AMD attribute the price increase found on retailers? What does AMD think caused the instant price hike at retailers?
I suppose this is what I'm getting at: If the defense is we can't control the pricing, and yet the launch price is clearly MSRP and has later spiked $100, then it seems as if there is some level of control somewhere. If that level of control is exercised through MDF, then that certainly seems an important part of the story. I am curious as to AMD's knowledge or speculation on what caused the price spike.
This is where it was clear that no further ground would be gained, and that AMD is not providing information beyond the above statement: AMDs representative was not authorized to further discuss the pricing questions.
We are still looking into this matter, but it seems as if the true price of standalone RX Vega cards (wherein standalone indicates gamer pack) is about $100 higher than expected. AMDs timing is interesting: Launching in the middle of the mining market, theres plausible deniability that mining demand although the card mines far worse than we expected is influencing retailer pricing. This defense is employed to a point of nearly accusing retailers of gouging based on demand, levying the we cant control that against a rising tide of confused consumers. It's certainly a possible cause, but the stories from retailers, AIB partners, and AMD do not align. We were hoping for a response with some substance from AMD, but the PR line is all were getting for now.
dr_rus
Member
I mean, it's pretty obvious that it needs "baking" solely because NV decided so. If they would have decided to launch something new this year then we would have Volta lineup in GeForces already. But with Vega and mining there's zero reason for them to move off Pascal for now.
llien
Member
GamerNexus was with gibbo on bogus mining claim, so, shrug.
Undervolting test, undervolted Vega 56 basically on par with Pascal on per/watt (and faster than stock Vega 64):
https://www.hardwareluxx.de/index.p...vega-56-und-vega-64-im-undervolting-test.html
LelouchZero
Member
GamerNexus was with gibbo on bogus mining claim, so, shrug.
Undervolting test, undervolted Vega 56 basically on par with Pascal on per/watt:
https://www.hardwareluxx.de/index.p...vega-56-und-vega-64-im-undervolting-test.html
Am I seeing that right? It's performing better after being undervolted? I'm guessing that it's able to hold higher clock speeds more consistently.
Spartancarver
Member
It doesn't manage to keep max boost clock all the time cause of power throttling, lower voltage means less power usage which means it can reach higher boost clock.
icecold1983
Member
What are poly rates in pascal?
1 tri every .5 clocks per sm?
1 tri every .3 clocks per sm?
1 tri every .3/.5 clocks per gpc?
Vega is 1 tri per clock per shader engine right?
dr_rus
Member
0.5 triangle per SM per clock. But the whole geometry pipeline can easily be limited by something else than peak triangle rate so there's no point in comparing these peak values.
The fact that 1070 is basically on the same level as Titan Pascal in PCGH's benchmark hints at peak triangle rate not being the limiting factor for Pascal there.
Time will tell with regards to Vega pricing. But, except for maybe the 1080 Ti, none of the higher GPUs are available at anything like MSRP. I'm not sure what the big fuss is?
Yeah the GPU market right now is terrible. Demand has skewed prices something awful. So glad I built a system back in May when prices were sane.
After looking at all of my options, I'm waiting a month or two to see where aftermarket-cooled Vega cards land and then going that route. It's more sensible for me to sell my 980 Ti and current monitor then go to Vega + FreeSync than it is to get a GSYNC monitor... GSYNC market is a mess.
Yeah the GPU market right now is terrible. Demand has skewed prices something awful. So glad I built a system back in May when prices were sane.
After looking at all of my options, I'm waiting a month or two to see where aftermarket-cooled Vega cards land and then going that route. It's more sensible for me to sell my 980 Ti and current monitor then go to Vega + FreeSync than it is to get a GSYNC monitor... GSYNC market is a mess.
Leonidas
Member
GamerNexus was with gibbo on bogus mining claim, so, shrug.
Undervolting test, undervolted Vega 56 basically on par with Pascal on per/watt (and faster than stock Vega 64):
https://www.hardwareluxx.de/index.p...vega-56-und-vega-64-im-undervolting-test.html
Now I'm wanting a Vega 56 again. Hopefully they have a decent amount of stock so they can actually sell for $399 on release date....
Well, Maxwell actually does do a 400mhz OC, and so does pascal if you're lucky at least over the officially listed reference boost clocks. My 1070s will hit 2100mhz on occasion while gaming which is more than 400mhz over the official boost clock. This is of course one hybrid card and one remarkably cool Evga acx card in sli. Amazed by that Evga card. Only hits 60 degrees or so max load in sli which is only 10 degrees hotter than the hybrid.
Also is 1900mhz out of the question for an undervolted watercooled Vega? I don't think so.
GamerNexus was with gibbo on bogus mining claim, so, shrug.
Undervolting test, undervolted Vega 56 basically on par with Pascal on per/watt (and faster than stock Vega 64):
https://www.hardwareluxx.de/index.p...vega-56-und-vega-64-im-undervolting-test.html
Hmmm, I actually could see that amd fine wine surpassing the 1080 Ti in at least some titles. Of course talking about amd with an OC and Nvidia without, but still, Vega looks more impressive than initially thought.
Yeah, we've been hearing about Volta for a very long time now. GV100, especially with an increased reticule size, should in no way be compared to consumer level cards.
On another note, "AMD RX Vega's pricing was ”not just for launch, but ongoing"" from an AMD rep: https://www.pcgamesn.com/amd/amd-rx-vega-launch-price-reduction
This is a fair point too.
Good
How long until after Volta launch will amd have Navi ready I wonder? Do we know if adopting gddr6 over hbm will help them on turnaround? Do we even know if they are abandoning hbm?
icecold1983
Member
Nah the gap is too big. It will be clearly better than a 1080 after at least a year tho imo
Based on synthetics and games the 64 is already better than a 1080 founders anyway. 500 point 3dmark gap vs 800 gap 1080ti has over Vega 64. Games I guess it's kind of a wash right now, and then power consumption, but yeah the 64 is at least pretty much neck and neck with a 1080, and the undervolted 56 actually is too for less money. Not bad actually. Yeah it may not ever completely close gap but amd launch drivers are notoriously bad. I think they will get close just given the pure size and brute force of this chip, it should come close to the Ti and I think eventually it will. Just gonna take a while.
Also, does anyone know why the undervolting here is improving performance so much and why does AMD seemingly have their cards running too much voltage for their own good out of the box?
DemonCleaner
Member
yeah doesn't run into the 200W power limit all the time that way. hence higher avg clocks.
really stunning that it even surpases the stock vega64 this way
icecold1983
Member
What 3d mark results are you referring to? The most relevent one(timespy) a 1080ti is far ahead
Amd runs higher voltages for stability. Just like nvidia does, but to a lesser degree
The undervolted firestrike extreme above. My math or my eyes suck today though as it's actually a 900 point gap, which is effectively 20%. So yeah Vega probably never going to catch the Ti, but I think it may get within 10% of the FE with driver improvements.
icecold1983
Member
1900ish at .95 on 1080/1080ti/titanxp isnt uncommon. Havent yet seen how much it reduces power draw tho
Does it improve benches though? What's surprising is performance increasing with power draw also going down, which is generally not how things work.
It's because it doesn't thermal throttle...
Is the hybrid 64 thermal throttling at all? Does it get pretty big gains by increasing voltage and power if it is still under thermal throttle point?
What results in better performance here essentially? Unvervolting or keeping the card as cool as possible with water while simultaneously jacking up voltage and power.
I'd like to see someone come out with a hybrid 56 and see how she does.
papersleeves
Member
Those results from hardwareluxx.de look rather suspicious. The Vega 56 beating the 1080 in Witcher 3 @1440p?
DF says otherwise:
Edit: I see. It's undervolted. It's really strange though. I'd like Gamersnexus to test this.
DF says otherwise:
Edit: I see. It's undervolted. It's really strange though. I'd like Gamersnexus to test this.
Higher clocks are causing the performance boost. The 56 undervolted is also running at 1600+ mhz stable versus hovering between 1300-1470 at stock voltage. It's only natural that overclocking causes a performance boost.
Of course people need to keep in mind that this might be an excellent chip and that not every Vega 56 card will be able to achieve this. Stock voltages are chosen for a reason. It's to increase yields and not because AMD can't figure out optimal voltages.
Basically, this.
To the original question: the general trend for overclocking is that higher voltages equal higher clocks, with the hard limits being heat and overall 'stability'. The recent trend is now to reduce voltage while overclocking to push the efficiency envelope as high as possible -- and to avoid thermal limits that may have prevented overclocks in the first place.
The only problem is that not every chip can undervolt the same way. It's much easier to push a chip with higher voltages than to undervolt.
edit: To understand the true extent of this undervolting, the community really has to band together to test a large amount of cards. Otherwise, we'll have no idea as to the statistical distribution of what to expect.
michaelius
Banned
thelastword
Banned
This is only part of what I've been alluding to.....A vega 56 performing better than a GTX 1080 through undervolting, even at stock and through small overclocks it beats the 1080 in a few games as it is.....Undervolted it's doing much better as has been shown through the collaboration of Gamersnexus and Buildzoid and of course now this site, but there are still limitations and restrictions on the VRM/HBM and bios, if these are unlocked I can see even more performance out of Vega....We've already seen some recent tests with DSBR and HBCC on and it's very promising as a further uptick on perf.GamerNexus was with gibbo on bogus mining claim, so, shrug.
Undervolting test, undervolted Vega 56 basically on par with Pascal on per/watt (and faster than stock Vega 64):
https://www.hardwareluxx.de/index.p...vega-56-und-vega-64-im-undervolting-test.html
Drivers will mature though, but something is still wrong with Vega 64 perf even with undervolting, it should perform much better...Even at stock between the cards, 56 and 64, tech of tomorrow highlights that 56 beats 64 in some games with MSAA on, as low as 2-4xMSAA, there is defintely something wrong with Vega 64 perf on a whole and that alone indicates that something is just not right here. Again, it's an AMD launch and not everything is firing on alll cylinders yet, even polaris had troubling issues at launch, farless for Vega which they said they wanted to push within the quarter.
I think AMD should allow overclockers.net to help them by opening up the bios and restrictions on the VRM and HBM tbh....Apparently, Vega is slighly, just slighlty bandwidth starved, so pushing more core speed without pushing the memory just as hard in tandem is going to limit your perf returns. The other problem is that pushing the core too hard is going to increase wattage considerably, so good coolers and undervolting should prove very effective....It will be even better if we could push HBM beyond it's 20 watt limit as well.....
FairXchange
Member
Just picked up a Vega 64. Feels good to finally be playing PUBG in 1440p at a high frame rate. Very impressed with the card so far.
icecold1983
Member
nvidia runs that particular game much faster
FairXchange
Member
Why are you in a Vega thread with this? If it's a joke I apologize, otherwise no need for that. I didn't ask which gpu runs it better.