• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Mysterious GPU Benchmarked in AOtS - Possible Upcoming AMD Graphics Card

Raevenlord

News Editor
Joined
Aug 12, 2016
Messages
3,755 (1.18/day)
Location
Portugal
System Name The Ryzening
Processor AMD Ryzen 9 5900X
Motherboard MSI X570 MAG TOMAHAWK
Cooling Lian Li Galahad 360mm AIO
Memory 32 GB G.Skill Trident Z F4-3733 (4x 8 GB)
Video Card(s) Gigabyte RTX 3070 Ti
Storage Boot: Transcend MTE220S 2TB, Kintson A2000 1TB, Seagate Firewolf Pro 14 TB
Display(s) Acer Nitro VG270UP (1440p 144 Hz IPS)
Case Lian Li O11DX Dynamic White
Audio Device(s) iFi Audio Zen DAC
Power Supply Seasonic Focus+ 750 W
Mouse Cooler Master Masterkeys Lite L
Keyboard Cooler Master Masterkeys Lite L
Software Windows 10 x64
Benchmarks of an as-of-yet-unknown GPU have surfaced in the Ashes of the Singularity benchmark database - one that could point towards performance levels on an upcoming AMD GPU (the RX 490?). With new cards based on Vega expected to arrive in early 2017 sporting a brand new architecture, hope is they will fill in AMD's utterly lacking (as in, equal to zero) enthusiast-class performance graphics solutions. The new GPU appears to be within spitting distance of the already established GTX 1080, with only a few percentage points separating the two (with the unknown graphics card having an 85.1 average frame-rate in all batches at 1080p standard settings, against 93.4 on a GTX 1080 - with both values potentially varying with particular overclocking characteristics and so on, so please don't take them at face value.





The data was attributed to a user Ceaj, and the GPU in question carries the Device ID 687F:C18. This is in-line with AMD's already-released Polaris generation of graphics cards, which also use the 0x6*** code - while still potentially denoting a different architecture, not being related to Polaris 10 or 11 - whereas rival NVIDIA makes use of 0x1***-type codes. The graphics chip appears to be much faster than Radeon RX 480, while offering comparable performance to GTX 1080. Judging from how AOTS benchmark recognizes dual-GPU graphics cards, the new card is most likely equipped with one processor. This device ID was not shown anywhere yet, so it's definitely something unreleased.

The Radeon RX 490 "Vega10" is rumored to sport the new GCN uARCH (GFX9). The chip is expected to deliver as much as 24 TFLOPs half-precision and 12 TFLOPs single-precision compute performance. It is also expected to feature up to 16 GB of HBM2 (now that it's a cheaper, more refined process than it was at Gen 1) that is clocked to operate at 512 GB/s. The cards based on this chip are expected to carry up to 225W rated TDPs.



View at TechPowerUp Main Site
 
Last edited:
exciting stuff - although AMD did say that Vega was going to be ~gtx 1080ish so this looks like further confirmation.
 
So a twin "480" GPU card that is as fast in the only benchmark AMD can really clobber Nvidia in..... who would think it!!!! https://www.techpowerup.com/reviews/AMD/RX_480_CrossFire/19.html


Also, water is wet, the wind blows, and snow is cold. Nothing against OP, but the AMD Hype train is either going to smash through some barriers and put a huge hole in their competition or derail in such horrifying brutal beautiful way no one will ever be able to take them seriously again. I hope for the first, but the second seems more of the cold reality at this point.
 
So a twin "480" GPU card that is as fast in the only benchmark AMD can really clobber Nvidia in..... who would think it!!!! https://www.techpowerup.com/reviews/AMD/RX_480_CrossFire/19.html


Also, water is wet, the wind blows, and snow is cold. Nothing against OP, but the AMD Hype train is either going to smash through some barriers and put a huge hole in their competition or derail in such horrifying brutal beautiful way no one will ever be able to take them seriously again. I hope for the first, but the second seems more of the cold reality at this point.

Except it's a single gpu chip.
 
After the rx480 hype train I am not holding my hopes high for 490. Not interested.
 
it doesn't look like anything to me. and yet able to match 1080 exactly. maybe if 400$ will try.
 
Well, 10fps difference is quite significant, especially in a game where AMD has a huge edge due to their GPU design. I'd say this is more a competitor for GTX 1070 than GTX 1080. Still, I'm excited.
 
Except it's a single gpu chip.


Not saying you don't know that, but how do you know that? Historically we can safely assume its a two chip card. The performance is inline with 2 480 chips on one card. Considering the volume AMD is pulling for Sony and MS of good wafers and an improved process, plus Zen, it seems highly unlikely they would re-engineer a whole chip for the 490, and not have the fingerprints we usually get from shipping/tracking and other leaks.

My money is on carefully binned chips, two of them on a board, called the 490, as has been their naming scheme for years (7970 +1 = 7990 etc...) and AMD will tout how great multi-GPU setups work in DX12 with Windows 10 and how with some spin improvements they have cut power consumption by 25% (10% in reality) by turning off the other die when not in use, which they will have a few bugs for restarting and enabling it, but will gladly ignore the issues and tell you about how great a game http://steamcharts.com/app/228880 AOTS is despite half of the "active" users are probably just AMD reps running benchmarks at any given time and giving each other the old dutch rudder, a quarter are actual players, and the other quarter is Nvidia trying to figure out how to get Gameworks into it to fuck it up for AMD.

Now, I like AMD and hope I am dead wrong, but we have a horse here with two broke legs and a rubber crutch. If AMD were half as smart as we need them to be, they would have forced Sony and MS to brand the shit out of "Powered by AMD" or the like on their consoles and created a whole marketing strategy off it instead of getting a participation prize.
 
called the 490, as has been their naming scheme for years
Except for that whole R9 290, and R9 390... and then logically, RX 490. Both previous cards were single chip.
 
Not saying you don't know that, but how do you know that? Historically we can safely assume its a two chip card.

Right in the article:
"Judging from how AOTS benchmark recognizes dual-GPU graphics cards, the new card is most likely equipped with one processor."
 
"hope is they will fill in AMD's utterly lacking (as in, equal to zero) enthusiast-class performance graphics solutions"

idk about that, the Fury(X), R9 295 and to an extend I guess the Produo still do a pretty good job high up there.
Hell if AMD was smart they would drop the price a bit and bam, solid competition.
 
Right in the article:
"Judging from how AOTS benchmark recognizes dual-GPU graphics cards, the new card is most likely equipped with one processor."


Explicit Multi-Adapter

ExplicitLinked.PNG
 
Looks like early benches from Vega more than a dual-gpu card. But Aots hasn't been a very good representation of overall performance, and I'm a bit leery of a game so uber-tuned specifically for AMD. So few people played it that I'd almost call it more of a dx12 graphics demo. Let's see some more varied benches and actual pricing, but it's certainly looking like it might be getting AMD competitive finally with the 1070/1080's.
 
"hope is they will fill in AMD's utterly lacking (as in, equal to zero) enthusiast-class performance graphics solutions"

idk about that, the Fury(X), R9 295 and to an extend I guess the Produo still do a pretty good job high up there.
Hell if AMD was smart they would drop the price a bit and bam, solid competition.
none of those are good modern options for a new rig. The fury x only has 4GB of RAM, which is not going to be enough for 1440p ultra, not to mention 4k, which is where the fury x does best. The pro duo and 295x2 are dual GPU solutions, not single cards, and suffer from all the disadvantages of dual cards. Given that modern games rarely support proper dual gpus properly, I'd never recommend those cards. And they are also VRAM limited, specifically the 295x2. Not to mention how much more power they need and heat they put out compared to pascal.

Also, the pro duo cost more then a titan pascal, which will walk all over it most of the time at 4k, and has three times the VRAM. the few times you can find a 295x2, it cost $100 more then a 1070 that will be far more consistent in performance then a 295x2. without crossfire, the 295x2 cant touch the 1070, nevertheless a 1080.

DualGPU should NEVER be a stand in for a proper high end solution. AMD has nothing to directly compete against the 1070 and 1080, and nvidia's revenue shows that. AMD needs to get a full proper 1070 competitor, and a 1080 competitor, out soon.
 
To back up @Steevo, there have been other commentaries on this (days old news) suggesting exactly either a dual config or a very refined Polaris chip. One site (can't recall but i check lots) thinks it would be stupid for AMD to abandon the Fury X tag (or equivalent) so a RX490 chip would better suit the Polaris arch. However, it's doubtful a Polaris chip (not mentioned in any roadmap) could perform close to a 1080 unless it was effectively a double size Polaris (think GP104 compared to GP100). I always did think it was weird Polaris capped out so small when it was quite a reasonable little chip.

Regardless, the 1080 levelling result is, as stated i AotS with DX12, so it says little about DX11 performance. If this was Vega and it only matches 1080 in AotS at DX12, AMD have some problems unless the price it lower than 1080 but some sites have suggested comparable pricing to 1080....

Right now I'm more excited about Zen as Kabylake is showing it's true colours as being Skylake in a new box. So god damned dissapointing.
 
To back up @Steevo, there have been other commentaries on this (days old news) suggesting exactly either a dual config or a very refined Polaris chip. One site (can't recall but i check lots) thinks it would be stupid for AMD to abandon the Fury X tag (or equivalent) so a RX490 chip would better suit the Polaris arch. However, it's doubtful a Polaris chip (not mentioned in any roadmap) could perform close to a 1080 unless it was effectively a double size Polaris (think GP104 compared to GP100). I always did think it was weird Polaris capped out so small when it was quite a reasonable little chip.

Regardless, the 1080 levelling result is, as stated i AotS with DX12, so it says little about DX11 performance. If this was Vega and it only matches 1080 in AotS at DX12, AMD have some problems unless the price it lower than 1080 but some sites have suggested comparable pricing to 1080....

Right now I'm more excited about Zen as Kabylake is showing it's true colours as being Skylake in a new box. So god damned dissapointing.
It wouldnt surprise me if this was a 3072 or 3584 core GPU based on the polaris arch with a 384 bit bus.
 
It wouldnt surprise me if this was a 3072 or 3584 core GPU based on the polaris arch with a 384 bit bus.


Why, when they can nail another GPU onto the same card, or just "tape" two cores together.
 
Why, when they can nail another GPU onto the same card, or just "tape" two cores together.
Because dual GPU brings with it a host of problems. Modern games are Russian roulette as to if they will support dual GPU, how well they will support it, ece. On top of that, even in well supported games, dual gpus still introduces more frame latency then an equivalent single GPU and more dropped frames. A single GPU will also be more efficient then dual GPUs, and easier to cool, cheaper to make, ece. And driver updates can accidentally tank dual GPU support.

A dual GPU card is not a suitable replacement for a single powerful product, ESPECIALLY when you currently have no high end products. If anything, a dual GPU AMD card would be in titan territory. they need a single GPU solution to fight the 1070 and 1080.
 
Explicit Multi-Adapter
This stuff has been tried for years. Compute is easy, but splitting rendering across multiple GPUs doesn't scale due to time constraints. You will be limited to only synchronize a few MB each frame without causing major delays. AFR already scales well, and is easy.
 
1080p standard is more like a processor test than anything else. How that crazy tests compares to the rest? That aots benchmark db is quite horrible and the test itself is useless without same version of the test and settings.
 
Will be interesting, though its a bit tardy to the party so it has to be at a lower cost to entice people in it even if it is equal in all respects (Which we know it won't). Look forward to it, but this is not enough to really judge it.
 
Well, how possible would it be to link two otherwise separate GPU's on a single die and present it to the system as single GPU? Remember how Intel and AMD used to create CPU's by basically merging two separate cores to create a single dual core die and those two shared rest of the processor components. Is that even possible with GPUs?

Or potentially stack two cores on a single interposer and have all the other processor units outside of that. It should be less problematic than two separate GPU's on two separate interposers linked by "long" links throughout the PCB. Because frankly, I think that will be the way to go in the future. Stacking more smaller cores and somehow presenting them as single GPU tot he system than creating super expensive massive GPU's that are hard to make because you have to throw out so many defective ones.
 
If a big boy vega, then it would have to be clocked pretty low to lose to a 1080.
 
crazy present 1080p it gets just 38fps? Seems low..

Well, I tested newest DLC AOtS with 980ti factory oc 1420mhz and got 52fps at crazy preset 1080p dx12.


noaa, same 52fps?


4xmsaa 52fps..


cpu 42fps
 
Last edited:
Back
Top