General Discussions

[kingfish]

MD unveiled its new Radeon RX 5700 line of graphics cards with 7nm chips at E3 last month, and with just days to go before they launch on July 7th, the company has announced new pricing. In the "spirit" of competition that it says is "heating up" in the graphics market -- specifically NVIDIA's "Super" new RTX cards -- all three versions of the graphics card will be cheaper than we thought.

The standard Radeon RX 5700 with 36 compute units and speeds of up to 1.7GHz was originally announced at $379, but will instead hit shelves at $349 -- the same price as NVIDIA's RTX 2060. The 5700 XT card that brings 40 compute units and up to 1.9GHz speed will be $50 cheaper than expected, launching at $399. The same goes for the 50th Anniversary with a slightly higher boost speed and stylish gold trim that will cost $449 instead of $499.

That's enough to keep them both cheaper than the $499 RTX 2070 Super -- we'll have to wait for the performance reviews to find out if it's enough to make sure they're still relevant.

AMD fires back at 'Super' NVIDIA with Radeon RX 5700 price cuts 

[hardcoregames_]

ajlueke wrote:

I'm not having a lot issues with Radeon VII on the 20.2.2 drivers currently.  I will admit, most of the drivers after the launch drivers could not be overclocked and stay stable so I ignored most of those.  Also, I tried the initial release of the "upgraded" drivers 19.12.2 I  think it was?  And those hard locked up when just sitting on the desktop.  However, I have my overclock in place in 20.2.2 and raise the minimum clocks a bit, and so far no issues playing Mechwarrior 5 or running a couple Superposition benchmark runs.  

How well does the fuzzy donut work on you rig?

[black_zion]

WCCFTech showed it a while back, the RX 580 is faster than the Fury X, and it did absolutely blow that they locked the HBM to 500mhz. I can't remember the last time the Fury series had any performance improvements, probably right after Polaris released in 2016, afterwards RTG threw everything into Polaris, granted RTG was in absolute shambles with Raja having done what he did with Vega, and they had to do everything they could to make Polaris somewhat competitive with nVidia.

And the minimum requirement for Doom Eternal is at 1920x1080 at the lowest detail levels, so we will have to see in third party reviews just how true that is, especially since they recommend an RTX 2060 OR the weaker Vega 56 for a much higher level of detail.

WCCFTech's GTX 1660 Super review shows RX 580 outperforms Fury X, 1660S outperforms RX 580, equal to...‌

[pokester]

Must really have some big requirements. Doom 2016 actually ran great for me on my Phenom II x6 1090t  and HD 7950 at 1080p. Glad my 2 gaming machines are 2060 or better. I am really looking forward to this game.

[black_zion]

I'm guessing it's going to depend on the amount of customization you use, as well as the detail levels. Plus ray tracing is no doubt going to absolutely slaughter performance when they add it later down the line.

But still, preorder Doom Eternal now, and you can be a unicorn!

[pokester]

I saw on Steam they are giving Doom 64 as a bonus too. I had not seen the unicorn. Although I am pretty sure I can live without it! My kid will like it probably. I have never once paid for custom skins, hats, etc....

[black_zion]

It's a free think if you link accounts and stuff like that, beh.

[hardcoregames_]

black_zion wrote:

It's a free think if you link accounts and stuff like that, beh.

more of that trademark infringer epic again?

[black_zion]

Oh, updated specs were just posted. They're still saying an RX 480 will be sufficient for 1080p60 high (but not the highest details), but for 4K60 at the -highest- detail settings (Ultra-Nightmare), only a 2080Ti will do. Also note the Windows 10 requirement for anything over minimum, and the processor requirement for Ultra Nightmare.

Only time will tell if this is an actual game, or a Crysis and Bioshock eye candy showcase with little substance.

https://www.tweaktown.com/news/71141/doom-eternal-needs-geforce-rtx-2080-ti-in-new-ultra-nightmare-mode/index.html

[black_zion]

Tacking on their official post as well, because it specifically states Windows 8.1 WILL NOT BE SUPPORTED, because of AMD. Don't think this will affect many considering Windows 10 is required for anything over minimum.

https://slayersclub.bethesda.net/en/article/5Wx9QeorMSfMZCwLg6VpoS/doom-eternal-launch-details?utm_source=Community&utm_medium=Social

[hardcoregames_]

windows 7 is a security nightmare, not sure why anyone would want to risk it

[colesdav]

"Removed Windows 8.1 as supported operating system because it is no longer supported by AMD as of their most recent driver release"

Who is the comedian?

Most recent driver release being ...

Revision Number

Crimson ReLive Edition 17.7.1 Optional

File Size

400 MB

Release Date

7/14/2017

[black_zion]

Almost sounds like nobody uses AMD cards there, and they just now checked to see if drivers exist, doesn't it?

[colesdav]

You can install Adrenalin 2019 19.12.1 Windows 7 driver on Windows 8.1 64 bit but it is not tested by AMD at all.
Since they are recommending 19.10.1 it must be the Windows 7 driver on Windows 8.1 they are using.

I could probably try to use Windows 8.1 device manager and try to install Adrenalin 20.2.2 INF files on Windows 8.1 64bit and see if that works.

As far as I am concerned though last useable GPU drivers from AMD are Adrenalin 19.12.1 and Radeon Pro for Enterprise 20.Q1 which now supports most GCN Radeon cards.

I am not going near that Adrenalin 2020 GUI/UI.

[hardcoregames_]

Refurbishing old machines I am often forced to use older Vista and 7 drivers for windows 10

[black_zion]

Interesting bit of information, nVidia could potentially be refreshing the RTX 2060 using 8GB GDDR6 as a way to position it higher than the RX 5600XT which has only 6GB. It's just an EEC submission right now, and may be a typo or even just speculative registering, but it is interesting nonetheless.

https://www.tomshardware.com/news/nvidia-geforce-rtx-2060-refresh

[hardcoregames_]

EVGA offered a RTX 2060 KO card as a gimmick more than anything.

All I see is that RTX 2060 in Canada is $120 cheaper than the RX 5700 XT (if you can even find one).

[ajlueke]

Seems like the game will heavily favor IPC over cores as well on the CPU side.

[ajlueke]

"But still, preorder Doom Eternal now, and you can be a unicorn!"

Yay?

[pokester]

It is likely very popular with the 100, 12 year old girls that will play the game. LOL

[ajlueke]

"It is likely very popular with the 100, 12 year old girls that will play the game. LOL"  

It is just such a weird skin to offer considering the overall tone of the game and the intended audience.  Oh well, I guess it is "something"?

[hardcoregames_]

Looking at the RX 580 vs my overclocked RX 480 and my card is a good bit more powerful. Sapphire overclocked my card quite a bit over 6TFLOPS so I believe this is why my card is able to play Halo etc at 4K without problems.

My RX 480 8GB beats the Fury with many games but the differences are close.

I play Doom 2016 with my RX 480 at 4K as well and it is playable.

[ajlueke]

I think the largest issue with leaving Fiji behind was that it didn't seem to be limited from a hardware perspective.  Fiji was ultimately very similar to Vega, but even the gap between those has widen over time.

[colesdav]

Four R9 Fury X cards is the best AMD can do in 3DMark Firestrike or TimeSpy using drivers from years ago, which unlock HBM frequency.

[hardcoregames_]

colesdav wrote:

Four R9 Fury X cards is the best AMD can do in 3DMark Firestrike or TimeSpy using drivers from years ago, which unlock HBM frequency.

All I know is that my RX 480 8GB OC tends to somewhat better with games. It varies from game to game.

No sure how to squeeze 4 cards into a rig as there is not enough room for more than 3 cards with ATX motherboards.

[ajlueke]

"Four R9 Fury X cards is the best AMD can do in 3DMark Firestrike or TimeSpy using drivers from years ago, which unlock HBM frequency."

Relatively meaningless though right?  Since crossfire is dying off with DX11 and most DX12/Vulkan engines don't have MultiGPU support past two GPUs if they have it at all.   How much higher would 4X Vega 64s be?  No one is going to bother to find out.

[colesdav]

It is not meaningless. 
3D Mark is a standard benchmark used in most reviews to compare GPU performance.

It shows how far behind AMD GPUs are in the sense that they have nothing that comes anywhere near the best performance that Nvidia offers from a single GPU.

There are still many people who are interested in 3DMark Benchmarking and Overclocking.

Three R9 FuryX's beat two RX Vega 64 liquids.

You cannot run four Vegas together.
You can only run two RX Vegas together in DX11 Crossfire or DX12 MultiGPU.
That is all that is supported in the Driver.

DX11 Crossfire is not dying off, it is dead.
The Adrenalin 2019 driver performance in DX11 Crossfire showed negative scaling for 3 R9 Fury X versus 2.
Adrenalin 2020 driver is a joke w.r.t. multiGPU. You do not even have the option of which cards you can pair up.

DX12 MultiGPU hardly even got off the ground at all.
I took a look at how that is performing on Adrenalin 2019 drivers at the start of 2020.
Very few of the few supported games show good results.

[ajlueke]

I didn't mean the benchmark is meaningless, only that 4X Fury is the best AMD can do, since as you say, other cards cannot run in 4X so there is no apples to apples comparator.

"It shows how far behind AMD GPUs are in the sense that they have nothing that comes anywhere near the best performance that Nvidia offers from a single GPU."

NVidia's best single GPU doesn't beat 4X Fury X either...so how does it show that exactly?

[Anonymous]

The FuryX was such a fantastic GPU with brilliant driver software, you can disable the powerplay in the driver which results in a stutter/lag free experience.

The Vega and VII are complete trash with built-in powerplay.

[colesdav]

They are still very good compute units for Blender MultiGPU and still win on performance versus price versus other AMD cards for that.
I am running a number of PC with 6 AMD GPUs installed for Blender Rendering.
In one case I am using an old R9 280x for display driver to view output + 4 R9 Fury X connected to PCIe ports + single XFX R9 Fury in Thunderbolt 3 eGPU box. 
Thats ~  41.5 TFLOPS of compute for Blender.

[hardcoregames_]

colesdav wrote:

They are still very good compute units for Blender MultiGPU and still win on performance versus price versus other AMD cards for that.
I am running a number of PC with 6 AMD GPUs installed for Blender Rendering.
In one case I am using an old R9 280x for display driver to view output + 4 R9 Fury X connected to PCIe ports + single XFX R9 Fury in Thunderbolt 3 eGPU box. 
Thats ~  41.5 TFLOPS of compute for Blender.

Running cards from a USB 3.0 cable like coin miners?

[pokester]

He said Thunderbolt. That is faster than USB 3 I believe. 

[colesdav]

Two R9 Fury X GPUs are running from standard PCIe3.0x8 slots.
One R9 Fury X on PCIe 2.0x1 to USB3.0 mining adapter.
The R9 280x is running on PCIe 2.0x1 to USB3.0 mining adapter.
One R9 Fury X is running at PCIe 2.0x2  from M.2 to ADT-Link eGPU Adapter.
XFX Fury is in a Razer Core X enclosure running from an Asus Thunderbolt EX3 card from a PCIe2.0x4 slot

[hardcoregames_]

colesdav wrote:

Two R9 Fury X GPUs are running from standard PCIe3.0x8 slots.
One R9 Fury X on PCIe 2.0x1 to USB3.0 mining adapter.
The R9 280x is running on PCIe 2.0x1 to USB3.0 mining adapter.
One R9 Fury X is running at PCIe 2.0x2  from M.2 to ADT-Link eGPU Adapter.
XFX Fury is in a Razer Core X enclosure running from an Asus Thunderbolt EX3 card from a PCIe2.0x4 slot

I have an HAF 932 EATX box and there is not nearly enough room. 

A third card would bump into my PSU. I tried my RX 480 and it would not fit.

I saw one mining setup with a wooden rack with video cards. That is were my EVGA GTX 1060 came from.

[colesdav]

They are not in the case (Corsair Obsidian 650D) - they sit around the PC.
Here is a picture of them running a Blender job:

The XFX Fury is in the Thunderbolt Box here:

[colesdav]

Can everyone spot the very cheap PSU upgrade?

[pokester]

I have never done any blender works. So out of curiosity how do those GPU's have to run at the slowest common speed or is it able to use each at their maximum speed available. Is it like a crossfire for instance. 

[colesdav]

I run all 4 Fury X at default speed - 1050 MHz.
I run the XFX Fury at 5% overclock to match the speed of the Fury X cards and then another +5% in attempt to compensate for cut down die. So it runs at 1100 MHz.
The R9 280x is just used as a display GPU. Blender 2.80 does not support GCN 1.0 cards (AMD needs to fix a driver problem).
It is best not to use a GPU that is used for Blender compute to display output as well.
Blender performs the rendering in 5 separate tiles, one for each Fiji based GPU.
The render time is really fast. 5 GPPUs run in ~ 1/5th the time of 1 GPU.

[pokester]

So each one isn't or is dependent on the other card being the same?

[colesdav]

It make more sense to use the same performance GPU so the tiles all complete at the same time.
If one tile is slower than the rest then it slows down the entire job.

They do not have to be the same AMD GPU at all though.
I can run 2 RX Vega 64 Liquid, RX Vega 56, R9 390x, and Radeon Pro Duo in the same Blender run on another machine (Maximus).
On that one though I tend to run Multiple Blender runs so

One blender job will use the 3 Vega card.
Second Blender job will run on R9 390x + Radeon Pro Duo.
Third  Blender job runs on on 14 of the 16 availble Ryzen CPU threads.
Another job can run on on the Primary Nvidia RTX2080.

[pokester]

Sorry to make you write so much I was just curious how that works. So thanks for taking the time. I learn so much from you!