AMD Vega MegaThread! FAQ and Resources

Page 14 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.

As Vega is a different architecture than previous GCN iterations and these cards are certified for pro applications, I'm not so sure it's that easy.
We should be able to see how Vega scales in comparison with other AMD chips through Vulkan, as it's both much simpler driver-wise and there's no pro app requiring a certified Vulkan driver. DX12 games should do the trick, too.
 


there will be one for reviewer just that it is not this FE but true RX Vega.
 
@mitch
Excellent stuff will keep an eye out for doom scores there actually in that video.. but he was running card in pro mode.. there gaming mode in the driver but It does look like they are a way's off on the gaming driver itself.. hopefully we could see a fairly big improvements in performance over the next while..

But people are just probobly gonna overreact to bits again over slightest thing anyway, if it's not savage straight out the gate.. if Ryzen was anything to go by.. are we in for a repeat of that hopefully not.. They don't have that much time either with Vega to sort it. It's already late an in few months Nvidia will release another Pascal card I think I heard.. not even Volta. Is Volta jus not ready yet or do they just not feel threatened enough to bother to release it yet ?
 


hopefully anandtech will get to come out with their review on day one. i'm interested to know how drastic the change will be compared to GCN. will it a big change like how it was Cayman VLIW5 to first gen GCN?
 
I'm not saying results weren't below expectations it came in at 1070 scores a lot of time..

But, he should of increased fan speed a bit an left it, once he did thermal throttling stopped..
All tests were done at 1400mhz..
AMD limit power draw to 280w max by default. But the card is rated at 300w

So testing could of been better would like to see it with better fan speed an higher clock speed pus 300w max power draw.
This would give us an indication of what the RX will do right !!

Don't get me wrong poeple probably gonna flip out over this but it's not a direct reflection of what the RX will be hitting an certainly could have been made more similar quite easily but wasn't.

An yeah driver update and gaming optimizations is bound to make a diff also..

I guess they have a month to sort it.. I would like to see some results at 1650 with full fan an 300w max power !

J

Edit:
Wasn't HBM 2 supposed to be faster too.. they couldn't get the faster stuff because someone block booked it all at samsung.. this would make a diff too.. if or when skhynix ramp up to 2.0 gbps.. I guess that ship has kinda sailed though eh..

Edit Again again...
It's really nice to start getting some bench's though eh thank god about time !!!
 
Considering those are "professional" drivers and have a lot of known issues, I'm not surprised.

On the "con" side, so far seems to be only ~40% faster than Fiji, even with the added extra hertz. I don't know what they changed in GCN for Vega, but it doesn't seem to have helped that much and looks like they put all the effort into the memory subsystem.

We'll have to see how they tweak Vega for the consumer side, because I have to say I am *not* amused nor impressed.

Cheers!
 


what PCPer do is testing the card's "out of the box" performance so it can be compared to 1080/1080ti/Titan Xp stock performance. in the past toms's hardware was hammered by reader when they compare stock 5850 vs factory overclocked GTX460 (it was EVGA GTX460 that clocked at 850mhz). in TH defense they said pretty much all GTX460 on the market can hit such clock but not many people buy their card and overclocked them. and the performance difference is quite noticeable especially with reference model which only clocked at 675mhz.

i'm sure given time PCPer will do many test including overclocking but what PCPer intend to show us with the live video is just the glimpse of what the performance we will get out of the box without any kind of tinkering done to the card.

about HBM AMD co-develop HBM development with SK Hynix. and back then we even heard that AMD will get the first priority to get SK Hynix HBM supply if there is other customer that also get their HBM supply from SK Hynix. but in the end most of SK Hynix supply probably used up by AMD so i think that's why Nvidia work with samsung instead because they most likely did not want to fight for scarce supply from SK Hynix with AMD. from people's discussion i heard so far i always have impression that SK Hynix HBM 2 is always faster than Samsung HBM2. and if you look at technical spec for Vega FE and nvidia tesla V100 that seems to be the case. for Vega the HBM2 used on the card was rated at 1.89Gbps:
http://www.anandtech.com/show/11583/amds-radeon-vega-frontier-edition-formally-launches-air-cooled-for-999-liquid-cooled-for-1499
now for nvidia tesla V100 the HBM2 is rated at 1.75Gbps:
http://www.anandtech.com/show/11559/nvidia-formally-announces-pcie-tesla-v100-available-later-this-year

 
@renz496

Please don't get me wrong it was good review if your curious about performance of the FE out of the box. But I think most are just dying to know what the RX will do.

This is the question we are all waiting to be answered... an well it still hasn't been unfortunately.

But we are getting closer I guess..

Also Samsung have the faster HBM2 AFAIK but no one else can buy it unfortunately. An the better Fab process it seems. I just don't understand how they have zero available for sale ? It seems like it could be a strategy from Nvidia to buy it all up. I mean if it is it worked they had to release the card with slower HBM in the end.
 


even if the card was overclocked and did not show the result people want to see people will still not going to be satisfied anyway. under normal circumstances the pro driver should not hold the card performance too much when running game related benchmark. GTX1080 for example is only 7.4% faster than Quadro P5000 in 3DMark Time Spy as tested by hothardware in their review for Quadro P6000 and P5000.

https://hothardware.com/reviews/nvidia-quadro-p6000-and-p5000-workstation-gpu-reviews?page=6

but for Vega right now i'm willing to believe there is something wrong with the result.

and for the HBM case i don't think there is any shady deals happen to prevent AMD from buying samsung HBM2. if anything it is proven that SK Hynix can produce much faster HBM2 than Samsung when we look at the HBM2 speed used on Vega and Tesla V100. only that nvidia end up with more bandwidth because they were using 4 stack instead of 2 like AMD Vega resulting memory bus width of 4096bit on tesla versus 2048bit on Vega. AMD using 2 stack only as opposed to 4 (Fury X is using 4 stack) most likely to increase yield and reduce cost.
 
So, given the actual numbers, it's way less than 40% over the Fury X "consumer" for the Vega "pro" edition... Around 25% for the games tested there. Drivers won't make a massive difference (5%-10% range) and other additional tweaks they might bring to the table (better cooling, higher base clock, different memory configuration), *might* put it in the 40% range over Fury X. That is a bit above the 1080, but hardy a gap that might justify over USD$500 for the consumer version? I hope it gets closer to the 1080ti and it doesn't seem that far-fetched for the "consumer" version of it, but no way it will be faster consistently.

Let's see how AMD plays this card. Looks like nVidia has free reign for a while, again.

Cheers!
 


will tiled rasterzation going to increase game performance in term of FPS? AFAIK the TBR in nvidia cards is not about increasing performance but more on making the architecture more efficient? that's why maxwell is power efficient despite being built on the same node as kepler. also i see some people thinking HBCC will make the game running faster as well in terms of minimum frame rate. though from what i can understand HBCC is not about making the game faster but to prevent performance drop when the GPU itself cannot provide enough VRAM to the game. but with Vega Frontier having 16GB HBM2 i don't think HBCC is needed at all. there is no game that can fill that much VRAM even at 4k.
 
All this and with AMD maybe skipping the release could mean that AMD is having trouble with the cards, eg. due to cutting staff, low capability of the GPU organization or unexpected issues with the new architecture.

Is it anything usual that GPU launches get delayed over and over? Besides shouldn't we start waiting for Volta+AMD's next big thing?
 
If they are still sticking to this roadmap:
https://www.bing.com/images/search?view=detailV2&ccid=%2fet6Lpsi&id=3B0EA4556F7C8BFA36573EE3D1E8B55946EBB67D&thid=OIP._et6LpsinrdB_3YLB70DTgEsCp&q=amd+gpu+roadmap&simid=608010733229574646&selectedIndex=0&ajaxhist=0

Then Navi is AMD's next lineup and that's not until 2018. Actually, Vega hasn't been delayed. AMD said 1H 2017 and the Pro cards are out...just in time. Officially, AMD never said otherwise, but rumors and wishful thinking had it slated for October 2016, then December 2016, then Q1 2017, and then May 2017. but AMD never changed from 1H 2017.

I'm looking forward to Vega consumer cards and by the time they come out, I hope to have all the money for my new build. If AMD delivers with Vega, it's very possible my new build could be my first ever all AMD computer. I haven't had a AMD CPU since my K6-2-300.

At this point in the grand scheme of things, all the kinks and bugs should have been found and corrected. They have released products based on Vega and issues that would effect the consumer cards will likely effect the Pro cards.

If they are delaying the consumer cards, it's likely yield issues and it's taking them longer than expected to build up stock. That or it could be driver issues, but I doubt its an architectural problem.
 


is Vega delayed? well if we looking at the history i don't think so. AMD stop releasing flagship GPU on every year basis after 7970. current trend with AMD is replacing flagship after two years.

2011 - 7970
2013 - 290X
2015 - Fury X
2017 - Vega

last year some people speculate that Vega will be coming out in 2016 but after looking at AMD past releases i think 2017 release is more likely (and i was right). AMD might release new GPU after releasing their flagship but the said GPU will not going to be faster than their flagship (like how polaris is still slower than Fury X). that's why i think Navi will be 2019 product. meaning until Navi arrives Vega will be AMD fastest GPU.
 
375W power consumption for 1000 € GPU? I don't think so. I don't need another heater in the room. Even 160W GPU heats my room to intolerable heights during summer months. When will AMD learn how to make power efficient GPUs?
 


when they have more money? but even if money is not really an issue the problem probably have to do with the very dated GCN architecture. AMD need major overhaul on the architecture not just add new thing to the base GCN architecture. but if they make drastic change to the point it is not GCN anymore it might not line up well with the so called "AMD Master Plan".
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom