AMD Radeon RX Vega 64 Video Card Review @ [H]

DigitalGriffin said:
I noticed this discrepancy too. And looking at Anand, the Vega is faster on 4K. But Kyle said it's really not suitable at that res.

All and all, I think it's competent with after market cooling at $500. Above that is a no-go. The free-sync does make the picture more attractive. But not by that much.

I hope VR has improved.
Click to expand...
I don't consider 30fps great gaming performance so neither the 1080 or 64 IMO are 4k cards.
 
Stoly said:
Others who? It seems HBM2 memory doesn't overclock much.
Click to expand...

Buildzoid made note of whether the BIOS could be flashed to aid in RAM overclocking, as the HBM has a notable affect on performance for AMD.
 
Nihilus1 said:
I agree that there is a FreeSync advantage, but that is like saying a Lincoln is better than a Aston Martin because the roads are better in the U.S. than the U.K. (ok I struggled making an analogy, but hopefully you get the point).
With an AIO cooler, it will be too close to a 1080ti. With aftermarket air, it will not o/c as good as a 1080 11gbps. Also, you would be dumping a TON of heat in your case.
Click to expand...

I fully agree. I went for the AIO (for noise reasons), which was similarly priced to the 1080ti at the time of purchase (it's already £50 more since then). Luckily, I don't own but want to play both included games, which are still £39.99 each in the UK, so that brought the price in line.
As for the analogy, how about this: a Lincoln is better than an Aston Martin when the former allows you to ignore speed limits.
 
Thanks for the review and the many sampled, actually played games! Awesome!

I expect some improvements with drivers, hopefully with MSAA as well but I doubt it will really overtake the 1080 by that much if at all, power will always trail the 1080.

All cards are out, we will see how well they are restocked.

I would have like to see the FuryX with the review to give context to improvements from Fiji to Vega but then that would probably meant less games tested. It appears indeed to be around 20% over Fiji(that is terrible).

I maybe interested in the Vega Nano, just have to wait and see how it pans out.

The AIO version I look forward to seeing done here when the chance arrives plus AIB ones. Those should beat a 1080Fe but probably not by much.
 
CSI_PC said:
I would say Nvidia's advantage with their architecture is that it is designed to scale heavily or be cut heavily; in Essence the GP102 is Tesla/Quadro/Geforce, the GP104 is Tesla/Quadro/Geforce,etc.

The point being there is not much difference even between the various segmented GPUs even on Nvidia, the exception being the largest die P100 and V100 purely designed for HPC and these are true multi-precision GPUs; FP16/32/64 with ideal scaling.
Also Nvidia is better at segmenting this architecture for its various markets.
The GCN architecture works well up to a point.
Cheers
Click to expand...
That is essentially my point. HBM makes sense for compute, and NVIDIA offers that in the P100 and V100 using the GP100/GV100 dies.

NVIDIA has the GP100 for High-end compute, GP102 for low-end compute/high-end gaming, and GP104 for mid-range gaming. GP104 is a much smaller die and thus more profitable.

AMD is doing all of the high-end lifting with just one die (Polaris doesn't scale above the $250 market) and the inclusion of HBM is likely what led to numerous delays, as well as makes the board more expensive to produce.
 
Yea... once prices come back down to earth, I'm thinking it'll be a 1070 for me. Especially since I plan on being 1080p for a long time.
 
when they ditch the stupid cooler and let manufactures work with it, well it might make it to 1080 performance or a little more.
But the wattage is crazy.
I knew it would not be a ti killer but if it can't even beat the 1080. Maybe it will make N drop the prices some.
N pricing has been a BS session for too long.

edit

and after looking aftermarket 1080's are cheaper then the stock vega
WTF
 
Last edited:
sparks said:
when they ditch the stupid cooler and let manufactures work with it, well it might make it to 1080 performance or a little more.
But the wattage is crazy.
Click to expand...

Keep in mind most site are doing reviews against 1080 FE cards. The aftermarket cooled 1080s are notably better at holding higher boost temps. Vega 64 aftermarket VS 1080 aftermarket should be pretty close to even.
 
Well, that's a bit disappointing, but not all together unexpected based on what we had been hearing prior to launch, based on the professional Vega card.

Back to the good old drawing board, Raja.
 
Woe is me....my 1440p freesync monitor is weeping tears of sadness right now.

I'm thinking I'll get a Vega 56 and be slightly sad about it. Damnit.
 
If they lower the price by $100 and come out with a driver that fixes the AA issue and cuts power by at least 50W, I will buy one.

They must have gotten confused and thought they humiliated nVidia instead of Intel.
 
Ocellaris said:
Keep in mind most site are doing reviews against 1080 FE cards. The aftermarket cooled 1080s are notably better at holding higher boost temps. Vega 64 aftermarket VS 1080 aftermarket should be pretty close to even.
Click to expand...

but if you go look the aftermarket 1080's are cheaper than this stock vega.....SAD
 
Ultima99 said:
Don't worry guys. In 3 years the driver improvements will have this thing TRASHING the 1080 and getting close to the 1080Ti....
Click to expand...

oh good, just long enough for me to save up to buy one.....seeing as how I am so far behind the times to play current games as it is....3 years seems about right.

I just bought Wolfenstein 1 through one of the steam summer sales....LOL
 
DigitalGriffin said:
I noticed this discrepancy too. And looking at Anand, the Vega is faster on 4K. But Kyle said it's really not suitable at that res.

All and all, I think it's competent with after market cooling at $500. Above that is a no-go. The free-sync does make the picture more attractive. But not by that much.

I hope VR has improved.
Click to expand...

And if you look at numbers from Anandtech you can see Kyle's right. They range from Maybe, to Nope. And while you could probably get the FPS back by turning the quality down, 1440p upsampled generally looks better than native but low quality.
 
Rizen said:
That is essentially my point. HBM makes sense for compute, and NVIDIA offers that in the P100 and V100 using the GP100/GV100 dies.

NVIDIA has the GP100 for High-end compute, GP102 for low-end compute/high-end gaming, and GP104 for mid-range gaming. GP104 is a much smaller die and thus more profitable.

AMD is doing all of the high-end lifting with just one die (Polaris doesn't scale above the $250 market) and the inclusion of HBM is likely what led to numerous delays, as well as makes the board more expensive to produce.
Click to expand...
One needs to separate die and HBM in this context.
The GV102 is also a heavy lifter HPC GPU and is a primary Tesla card over £5k, with 24GB GDDR5X.
In fact GV102 is Nvidia's HPC FP32 card by a fair margin, and also DL card for massive performance inferencing (int8).
The HBM is more niche for Nvidia, but AMD has to use it unfortunately due to the overall TDP/TBP and BW/cache aspects they are still needing to improve to catch up with Nvidia.

I guess we are at cross points/perspective, because both make their die GPU and core architecture across all segments; point being Nvidia are not designing cards for each unique market apart from one model out of all of the range but that range is still heavily used in HPC down to Geforce.
Cheers
 
Last edited:
Over the last few years I've wanted to get a Freesync monitor because they have some large nice ones for not a lot of money on occasion. But the main thing stopping me was the whole AMD uncertainty thing when it came to performance and cost. I mean, It's my money and I know what I've spent my money on and when I did go AMD, I was always 2nd in performance to nVidia. I literally refused to spend the extra $150 or $200 to move over to nVidia even if it meant I had the slower video card with AMD. But now something crazy and insane has happened. AMD can no longer control pricing on their cards. be it demand, mining, etc.

So now, the ones of you out there that are still with AMD have an incredible incentive that I never had in the past. And that's cost toward performance. You can literally spend an extra $50 dollars to almost nothing at all to just move over to the 1080 ti. Had I been able to do that when I was deciding who to go with I sure as hell would have picked nVidia.

Also, the increase in power draw is going to add 1/3 in electricity cost. That's going to add up over the expected 24 months you guys will own these cards. If you look at the performance not being that great, the additional power draw, the inflated costs that make no sense whatsoever the choice is just super clear.

Who in their right mind would pass up a 1080 ti ? I mean you can find them used even for around $600 here in Kansas City and that's with no tax.

Also, nVidia is not going to stop advancing, they are going to release their new cards and the perfromance will be incredible and AMD will fall further behind. They are not going to raise prices and lower performance. In fact they might even lower the price and bit to kill off AMD for once and all. Imagine how many more customers would leave AMD? AMD only has a small part of the PC market as it is according to Steam Stats.

Mark my words. AMD will never ... ever .... again beat nVidia ....... just like AMD will never ever again beat Intel at single core performance. That ship has sailed.
 
DigitalGriffin said:
I noticed this discrepancy too. And looking at Anand, the Vega is faster on 4K. But Kyle said it's really not suitable at that res.
Click to expand...

My Pascal Titan is barely passable in most titles at 4k, and in some (Like Deus Ex Mankind Divided) I even have to resort to custom letterboxed 21:9 resolutions (3840x1646) and 50hz vsync (european TV mode) in order to get it to play right.

I'm tempted to jump on a Titan Xp for 4k, but I don't want to spend yet another $1,200 when the increase in performance may not be enough to make me truly happy.

I'll see what the Volta Titan brings, probably early next year.
 
might still be hopes for vega 56, all i want is freesync on wow with 3440x1440... the gtx 1060 cant quite give me smooth gameplay, adaptive sync helps but still slowdowns and tears...the vega 64 no one in Ontario Canada can afford to run wuth Liberal jacked Hydro rates lol
 
Last edited:
Good review as always, I just read the RX Vega 56 review on Guru3d, looks like the card to have, trashes the GTX 1070 in all but 2 games and uses a lot less power that the Vega 64. I think I will wait a while to see how things pan out before upgrading !!
Also it would be good to see how these cards do with VR games, as previous AMD cards were poor.
 
And in a few months you'll get new "eternal beta" drivers that make things better for game published a year ago.

I've been liking the renewed AMD stuff. Their CPUs are a fantastic value, and look awesome if you cherry pick your scenarios.

But man, after digging into the details I think Intel/Nvidia are still where's it at. If you do a mix of gaming and stuff than needs buckets of threads, Intel/Nvida has the most balanced solution. IMO.

I was hoping AMD would really knock it out of the park on Vega and Zen. Instead it's just "pretty good." And they have Intel spooked, which is fantastic. My first dual CPU rig was a dual Athlon 64 setup and it spanked Intel super hard. Hopefully AMD iterates on this new architecture and really amps things up.
 
Thanks for the review. As soon as I saw there was no GTA 1080 Ti card in the review I knew it wasn't the card we were all hoping for. It's definitely a step in the right direction but sub-1080 levels more than a year late just isnt good enough.
 
CSI_PC said:
One needs to separate die and HBM in this context.
The GV102 is also a heavy lifter HPC GPU and is a primary Tesla card over £5k, with 24GB GDDR5X.
In fact GV102 is Nvidia's HPC FP32 card by a fair margin, and also DL card for massive performance inferencing (int8).
The HBM is more niche for Nvidia, but AMD has to use it unfortunately due to the overall TDP/TBP and BW/cache aspects they are still needing to improve to catch up with Nvidia.

I guess we are at cross points/perspective, because both make their die GPU and core architecture across all segments; point being Nvidia are not designing cards for each unique market apart from one model out of all of the range but that range is still heavily used in HPC down to Geforce.
Cheers
Click to expand...
I don't think you can fully separate die and HBM, because you have to build the memory controller to support HBM, and it imposes manufacturing limitations for how large of a die you can build (it has to fit with the memory on the interposer). You're also beholden to HBM availability which is far worse than GDDR5X.

To your point though, perhaps the issue here is simply how they are designing the chips. I believe NVIDIA stated that starting with Maxwell they designed for low power and then scaled up to desktop GPU usage, which drove their efficiency increases as they built with mobile in mind. One would think that AMD has been doing this with their position on APUs, but maybe not? I don't see any other reason why they could be coming up so short on power efficiency unless the GF process is that bad.
 
and......... all sold out already.

All your base are belong to cryptominers.


So it all evens out. Excessive electricity usage during summer, lower heating bills during winter.
Actually, scratch that. Electric heating is never cheap compared to natural gas.
 
I was right there with everyone else when the original announcements about HBM were made a ways back. Cue the actual products hitting retail and it has only been disappointment after disappointment (HBM and now HBM2). It seems the advantage (s) has turned more theoretical than actual ... /sadpanda

I cheered for Ryzen and spent my own $$$ and I was hoping AMD could pull something out as well, just to create competition in the videocard marketplace. Just look at how Intel has course changed due to Ryzen and Threadripper. Having Nvidia being the sole graphics player is definitely slowing innovation.
 
erek said:
Maybe once we get WHQL drivers we'll finally see the performance we deserve after all this wait
Click to expand...

You won't, since this card is pretty much a 14nm Fiji card. AMD has been making Fiji drivers for 2 years now, they would have found a silver bullet if there was one.
 
This is hardly a surprise to me. Vega still at its heart was an Arctic Islands architecture and while they may have improved upon it, we saw its limitations with Polaris.

I have been saying for a year that this was going to happen. Navi is going to be the one to compete if the RTG ends up with a competitive card.

This isn't some wait till the next card every card Fanboy delusion. This is rationally taking a step back and realizing that Vega was already in development when Raja koduri took over, and while he was able to push the Arctic Islands architecture to this level, it was too late to go back and design something completely new given amd's lack of resources at the time and the millions of dollars already invested in research and development. They simply were not in a spot to call it a wash and start over. They needed badly to recoup their costs and thus with Polaris and Vega are doing their best to make lemonade from lemons.

Does this mean Navi is going to be the Saving Grace? Not necessarily, but it was always the card where we would finally find out what Raja really is capable of.

Navi is the first card of a brand new architecture that was overseen by Raja himself. There have been those who argued what difference a single engineer can make, I would Point them to rysen, as other than Jim Keller, the engineers were mostly the same for AMD and yet look at the results. Sometimes you just need someone with genius and vision leading a project and that was the key ingredient AMD was lacking prior to Ryzen, and could be what they were lacking prior to Raja although we do not know..

Therefore this isn't some AMD fanboyism of wait until the next card every card. Rather it always was Navi or bust. Navi it will be a brand new architecture and not restricted by the limitations of the Arctic Islands line.

If Amd is going to compete in graphics it will show with Navi. And if Navi is a flop then AMD is better off selling off the RTG to Intel like Kyle alluded that Raja wanted so as not to further drag down amd's profit margin. Because if Navi flops there is no next card. At least not under Raja's leadership, and as we can see with Polaris and Vega, it can take a decade or near that to get out from under a bad architecture design when you don't have the extra cash to start over like the big players do with Nvidia and Intel.
 
Lame, I guess I'm gonna keep my 290 for another year or two.
Too bad the new drivers went full retard with fan profiles, had to install MSI Afterburner for more reasonable and stable RPMs instead of 15 sec on high, 30 sec on low the drivers have.
 
The best thing about these Vega reviews is that mining performance/watt is very close to 1070 hashrates. So Vega prices will not go much higher than 1070's prices.
AMD will need pretty generous bundles to sell Vega to gamers. There are no bad products, only products with bad prices.
 
Singularity_Survivor said:
Wait, what does a Youtube video have to do with this review and the failure to mention how important these GPU's are/will be to Freesync owners? Please don't get defensive, I'm just pointing out this is a HUGE deal for a great many of us.
Click to expand...

Because a vega 64 with freesync was preferred over a 1080 ti with Gsync in that video while costing less overall. Completely subjective test and used Doom which is best situation for Vega but it went to show how much of a difference Freesync can make with lower FPS vs a faster card even if it's using Gsync.
 
You must log in or register to reply here.
Back
Top