Jump to content

AMD Radeon VII Benchmark/Launch Mega Thread

Taf the Ghost
12 minutes ago, cj09beira said:
19 minutes ago, JoostinOnline said:

 

radeon Vii has to have 4*4GB hbm stacks to reach 1Tb/s with 16GB

Does that theoretically mean that performance would drop with an 8GB version (assuming less than 8GB is used)?

Make sure to quote or tag me (@JoostinOnline) or I won't see your response!

PSU Tier List  |  The Real Reason Delidding Improves Temperatures"2K" does not mean 2560×1440 

Link to comment
Share on other sites

Link to post
Share on other sites

Just now, JoostinOnline said:

Does that theoretically mean that performance would drop with an 8GB version (assuming less than 8GB is used)?

there is 2 ways to go about it, there is the 2* 4GB (2 stacks 4Hi) and 4* 2Gb (4 stacks 2 Hi),

as long as there is no memory capacity problems the first would have lower performance the second would have the same performance, but the second might have some problems do to the hbm being shorter than it already is, thus more would have to be cut from the core for both to be at the same height 

Link to comment
Share on other sites

Link to post
Share on other sites

45 minutes ago, cj09beira said:

radeon Vii has to have 4*4GB hbm stacks to reach 1Tb/s with 16GB

That actually escaped me when I made my initial remarks about viability of same performance with less HBM, thanks for pointing that out, bonus points for calm and non confrontational tone ?

CPU: i7 6950X  |  Motherboard: Asus Rampage V ed. 10  |  RAM: 32 GB Corsair Dominator Platinum Special Edition 3200 MHz (CL14)  |  GPUs: 2x Asus GTX 1080ti SLI 

Storage: Samsung 960 EVO 1 TB M.2 NVME  |  PSU: In Win SIV 1065W 

Cooling: Custom LC 2 x 360mm EK Radiators | EK D5 Pump | EK 250 Reservoir | EK RVE10 Monoblock | EK GPU Blocks & Backplates | Alphacool Fittings & Connectors | Alphacool Glass Tubing

Case: In Win Tou 2.0  |  Display: Alienware AW3418DW  |  Sound: Woo Audio WA8 Eclipse + Focal Utopia Headphones

Link to comment
Share on other sites

Link to post
Share on other sites

2 hours ago, i_build_nanosuits said:

except the gimped the 16Tflops of FP64 from the actual compute card the VII is based on down to only 1.6Tflops on this card making it basically useless for double precision floating point compute.

Except that it is 3.52 TFLOPS not 1.6. It is not useless, it's actually half as fast as an nVidia Tesla V100 in FP64 at less than 1/10th the price.

CPU: i7 4790k @ 4.7 GHz

GPU: XFX GTS RX580 4GB

Cooling: Corsair h100i

Mobo: Asus z97-A 

RAM: 4x8 GB 1600 MHz Corsair Vengence

PSU: Corsair HX850

Case: NZXT S340 Elite Tempered glass edition

Display: LG 29UM68-P

Keyboard: Roccat Ryos MK FX RGB

Mouse: Logitech g900 Chaos Spectrum

Headphones: Sennheiser HD6XX

OS: Windows 10 Home

Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, JoostinOnline said:

 

I honestly think it would if they made an 8GB version for $100 less. It seems like the performance would be similar in gaming.

Still wouldnt be bought. Also it would cost a lot more for AMD to create. It only excists to stay relevant in the GPU market and to have something to show at CES

Link to comment
Share on other sites

Link to post
Share on other sites

51 minutes ago, GoldenLag said:

Still wouldnt be bought. Also it would cost a lot more for AMD to create. It only excists to stay relevant in the GPU market and to have something to show at CES

Why wouldn't it be bought? I get that you wouldn't buy it, but why do you think that nobody would want it?

Make sure to quote or tag me (@JoostinOnline) or I won't see your response!

PSU Tier List  |  The Real Reason Delidding Improves Temperatures"2K" does not mean 2560×1440 

Link to comment
Share on other sites

Link to post
Share on other sites

2 hours ago, JoostinOnline said:

Does that theoretically mean that performance would drop with an 8GB version (assuming less than 8GB is used)?

Yes, that it does mean.

 

Or they have to order completely different HBM stacks and ramp up production of another part with "2hi" Stacks, wich is pretty unlikely. So they will stick with 4hi and 8hi stacks (meaning 4 and 8 GiB HBM Stacks)...

 

So there is just no other way.

And as said earlier, High End Cards SHOULD have more RAM than the mid range. At least 50%, more (like Radeon HD7970 vs. Radeon 7870).

"Hell is full of good meanings, but Heaven is full of good works"

Link to comment
Share on other sites

Link to post
Share on other sites

30 minutes ago, JoostinOnline said:

Why wouldn't it be bought? I get that you wouldn't buy it, but why do you think that nobody would want it?

i wouldnt buy any card at that price. this isnt about me buying it. its about the general public buying it. 

 

in general it wouldnt sell well at all. its allways been like this with AMD and Nvidia. its not that nobody want it, but it would be the exact same people who want it now who would buy it then. 

 

its simply not worth it in any financial way for AMD to do so. they could spend the hours of making a 512-bit GDDR6 variant of a vega 7nm. over engineering the memmory layout to make the impossible happen. just to lower the price of the card and make it 100-150$ cheaper. what would they get in return? nothing, that is what they would get. 

 

it wouldnt sell well because Nvidia, everytime AMD has had a compelling or straight up better product, people bought Nvidia. why?, because mindshare. 

Link to comment
Share on other sites

Link to post
Share on other sites

In case anyone wants to know a bit more about the Radeon VII driver situation I have been testing my card for a couple of hours and have found a few things out.

I have found that the card can and will overclock just fine but overclocks don't get applied at all in fullscreen modes or with some game engines.

If I run anything in fullscreen mode it will always use the stock clocks, but in windowed mode overclocks apply ok. Sometimes you have to reset wattman to defaults and reapply the overclock to get it to work again.

For actual overclocking gains I have been able to get an extra 10% on FPS so far in Heaven/Valley but in superposition the clocks do not boost up at, although I might be CPU bottlenecked here.

I've screenshot of wattman with Radeon VII at 2000Mhz:J5ChKTm.png

Stock Clocks:

SAanuEc.png

2000Mhz Core:

QsQvd1Q.png

 

Gaming Rig:CPU: Xeon E3-1230 v2¦RAM: 16GB DDR3 Balistix 1600Mhz¦MB: MSI Z77A-G43¦HDD: 480GB SSD, 3.5TB HDDs¦GPU: AMD Radeon VII¦PSU: FSP 700W¦Case: Carbide 300R

 

Link to comment
Share on other sites

Link to post
Share on other sites

1 minute ago, GoldenLag said:

i wouldnt buy any card at that price. this isnt about me buying it. its about the general public buying it. 

 

in general it wouldnt sell well at all. its allways been like this with AMD and Nvidia. its not that nobody want it, but it would be the exact same people who want it now who would buy it then. 

 

its simply not worth it in any financial way for AMD to do so. they could spend the hours of making a 512-bit GDDR6 variant of a vega 7nm. over engineering the memmory layout to make the impossible happen. just to lower the price of the card and make it 100-150$ cheaper. what would they get in return? nothing, that is what they would get. 

 

it wouldnt sell well because Nvidia, everytime AMD has had a compelling or straight up better product, people bought Nvidia. why?, because mindshare. 

It wasn't really ever "mindshare" that moved the market. It was some extremely shrewd moves to controlling the retail channel at the right time. Nvidia went all-in with their Software Suite rather than their hardware. We forget that Maxwell saved Nvidia's bacon before their CUDA Ecosystem finally took off. The market was much more 50/50 until fairly recently.

Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, Stefan Payne said:

Yes, that it does mean.

 

Or they have to order completely different HBM stacks and ramp up production of another part with "2hi" Stacks, wich is pretty unlikely. So they will stick with 4hi and 8hi stacks (meaning 4 and 8 GiB HBM Stacks)...

 

So there is just no other way.

And as said earlier, High End Cards SHOULD have more RAM than the mid range. At least 50%, more (like Radeon HD7970 vs. Radeon 7870).

Fair enough. I thought halving/doubling the memory on a GPU didn't require a redesign. But my reasoning was it would then compete with the RTX 2070, rather than the RTX 2080. While still pretty expensive and outside the mainstream market, the better performance would give you a compelling reason to choose it over Nvidia for gaming.

 

I hope Navi really hits it out of the park, because the GPU market has been depressing af.

1 hour ago, GoldenLag said:

it wouldnt sell well because Nvidia, everytime AMD has had a compelling or straight up better product, people bought Nvidia. why?, because mindshare. 

That's not really true. If it was, AMD wouldn't exist. You're making huge generalizations.

Make sure to quote or tag me (@JoostinOnline) or I won't see your response!

PSU Tier List  |  The Real Reason Delidding Improves Temperatures"2K" does not mean 2560×1440 

Link to comment
Share on other sites

Link to post
Share on other sites

34 minutes ago, JoostinOnline said:

I thought halving/doubling the memory on a GPU didn't require a redesign.

That depends.

On how it is archieved

What Memory is available.

And where you put the Memory.

 

For example look up the 256MiB (earlier) Radeon 9700 and compare that with the Radeon 9700 standard.

You have double the memory chips on the PCB in this case because that's what was available.

In this case we're talking about special Memory on the Package of the Chip...

 

And you can only use this Memory for this chip. And AMD seems to have already two different stacks (8GiB/Stack for the bigger card and 4GiB/Stack for the 56 CU version)...

 

34 minutes ago, JoostinOnline said:

But my reasoning was it would then compete with the RTX 2070, rather than the RTX 2080. While still pretty expensive and outside the mainstream market, the better performance would give you a compelling reason to choose it over Nvidia for gaming.

As said many times, you're wrong.

Because too many people would not buy AMD at all.

Just look at the GPU Forum, how many people consider the 1050ti over the RX570 although the RX570 isn't much more but way faster.
Or any other Graphics card. Doesn't matter what. 

People always try to justify the more expensive, worse card from nVidia over the better AMD Card. Always.


In the lower price regions, people are not that "bound by label" as in the higher end. And thus it makes sense to market your higher end products towards non gaming audience - like Miner, Compute Stuff and other things. If your card does a better job at lower cost than the competition, people will buy it. In some cases its just a business.

 

Meaning: All people care about is what does it cost, how much money can it make for me...

34 minutes ago, JoostinOnline said:

I hope Navi really hits it out of the park, because the GPU market has been depressing af.

You can start by telling People to buy AMD Products, especially when they are superior to the nVidia Products.

Everyone here can do something.

But its a hard, stony road...

 

But demanding AMD to make products you like, so that nVidia lowers the prices, just to buy the nVidia Product, wich is what many people are doing, is just not sustainable. 

 

34 minutes ago, JoostinOnline said:

That's not really true. If it was, AMD wouldn't exist. You're making huge generalizations.

https://store.steampowered.com/hwsurvey/videocard/

 

 

Look at that!

VEGA has about the same amount there as a 2080ti has. (0,16% for the VEGA, 0,15% for the 2080ti and that's BOTH VEGA)

A 2080 is double that at 0,31%

 

RX580 is a meager 0,82%, 470 is 0,25%, 570 0,23%, RX480 is at 0,6%...

 

The 1060 is at almost 15%, the 1050ti at 9,34%, 1070 at 4,3%.  even the expensive 1080ti is at 1,64%.

 

Do you really think I'm wrong after all these Numbers??

"Hell is full of good meanings, but Heaven is full of good works"

Link to comment
Share on other sites

Link to post
Share on other sites

2 hours ago, JoostinOnline said:

That's not really true. If it was, AMD wouldn't exist. You're making huge generalizations.

Tell that to the stores who had R9 fury and nano parts who they pretty much sold at a loss just to get rid of them.......

 

And AMD/Ati everytime they had a better product, better priced product, no hardware failures (not literally none, but not a product that consistantly failed), less powerhungry and colder cards.

 

Not saying Nvidia was bad, but AMD never got the market cap they should have gotten. (They cucked some things themselves aswell). 

Link to comment
Share on other sites

Link to post
Share on other sites

On 2/8/2019 at 10:51 PM, Madgemade said:

Stock Clocks:

SAanuEc.png

2000Mhz Core:

QsQvd1Q.png

 

Well that's good to see. So there is 10% more performance left in the tank for overclockers to unlock...

 

I think AMD has clocked it pretty high to begin with because they wanted to make sure they at least beat the RTX 2080 in some games. But 7nm is still pretty new for big GPUs so we may see better overclocking Radeon 7s as the process matures.

Link to comment
Share on other sites

Link to post
Share on other sites

2 hours ago, Humbug said:

Well that's good to see. So there is 10% more performance left in the tank for overclockers to unlock...

I think AMD has clocked it pretty high to begin with because they wanted to make sure they at least beat the RTX 2080 in some games. But 7nm is still pretty new for big GPUs so we may see better overclocking Radeon 7s as the process matures.

I don't know if I got a good chip or if AMD screwed up but the voltages are too high stock. Default voltage is 1025mv but I can run at 975mv no problem and the fan stays much quieter than stock. The highest I can get it is 2080Mhz at about 1125mv. If I try to go higher than that the junction temp hits 110+ and freezes the system. Once they have fixed the drivers I hope the professional overclockers will give it another go because it looks like the press drivers were really broken.

I was able to get another couple of FPS in heaven by running at 2050 core and 1200 memory:

1087054366_ocop.jpg.e8daf53fddc18197dd22ec6e2d79386a.jpg

My scores for superposition are in the benchmark thread. It is a lot better when overclocked. Given that the max allowed memory is 1200 and it run at that without issue and that on Air I can get 2080mhz sustained, I think with LN2 it would be possible to easily put every slider all the way in Wattman. So we need to get some support in custom tools to increase those limits and actually see some world record level stuff.

 

Gaming Rig:CPU: Xeon E3-1230 v2¦RAM: 16GB DDR3 Balistix 1600Mhz¦MB: MSI Z77A-G43¦HDD: 480GB SSD, 3.5TB HDDs¦GPU: AMD Radeon VII¦PSU: FSP 700W¦Case: Carbide 300R

 

Link to comment
Share on other sites

Link to post
Share on other sites

44 minutes ago, Madgemade said:

I don't know if I got a good chip or if AMD screwed up but the voltages are too high stock.

Typical AMD imo lol. Their cards need mandatory manual undervolting. 

 

Performance/Watt is similar to 2080 after undervolting but ymmv.

1960478650_OperaSnapshot_2019-02-10_175546_www.computerbase_de.png.ed2bdcdb17ca4d9c27b1104d2f5a929b.png

| Intel i7-3770@4.2Ghz | Asus Z77-V | Zotac 980 Ti Amp! Omega | DDR3 1800mhz 4GB x4 | 300GB Intel DC S3500 SSD | 512GB Plextor M5 Pro | 2x 1TB WD Blue HDD |
 | Enermax NAXN82+ 650W 80Plus Bronze | Fiio E07K | Grado SR80i | Cooler Master XB HAF EVO | Logitech G27 | Logitech G600 | CM Storm Quickfire TK | DualShock 4 |

Link to comment
Share on other sites

Link to post
Share on other sites

9 minutes ago, xAcid9 said:

Performance/Watt is similar to 2080 after undervolting but ymmv.

0v8tRg.md.png

Got another link/source for that? It's not working for me at all.

Link to comment
Share on other sites

Link to post
Share on other sites

26 minutes ago, leadeater said:

Got another link/source for that? It's not working for me at all.

Yeah weird.. it broke when i click save. I can see the image while editing.

 

https://www.computerbase.de/2019-02/amd-radeon-vii-test/4/#abschnitt_performance_pro_watt

 

But that only compare to 2080 FE, if you have this aftermarket bad boy.. 

power_average.png

 

You may want to start learning how to undervolt too. ?

| Intel i7-3770@4.2Ghz | Asus Z77-V | Zotac 980 Ti Amp! Omega | DDR3 1800mhz 4GB x4 | 300GB Intel DC S3500 SSD | 512GB Plextor M5 Pro | 2x 1TB WD Blue HDD |
 | Enermax NAXN82+ 650W 80Plus Bronze | Fiio E07K | Grado SR80i | Cooler Master XB HAF EVO | Logitech G27 | Logitech G600 | CM Storm Quickfire TK | DualShock 4 |

Link to comment
Share on other sites

Link to post
Share on other sites

51 minutes ago, Madgemade said:

I don't know if I got a good chip or if AMD screwed up but the voltages are too high stock. Default voltage is 1025mv but I can run at 975mv no problem and the fan stays much quieter than stock.

Haha, that sounds like Vega. Got a bit buff and bought a new suit, but still Vega ?

Link to comment
Share on other sites

Link to post
Share on other sites

26 minutes ago, xAcid9 said:

But that only compare to 2080 FE, if you have this aftermarket bad boy.. 

lol wow 60W jump just from an AIB card

Link to comment
Share on other sites

Link to post
Share on other sites

That intro.. lmao. ?

 

TL;DW
The software is buggy as hell, the card has good OC potential.
Liquid metal didn't improve the temperature as much.

 

 

44 minutes ago, leadeater said:

lol wow 60W jump just from an AIB card

Somehow Zotac 2080 AMP Extreme default board power limit is 280w, way higher compare to others aftermarket 2080.

power-limit.png

 

Spoiler

tdp-adjustment-limit.png

 

| Intel i7-3770@4.2Ghz | Asus Z77-V | Zotac 980 Ti Amp! Omega | DDR3 1800mhz 4GB x4 | 300GB Intel DC S3500 SSD | 512GB Plextor M5 Pro | 2x 1TB WD Blue HDD |
 | Enermax NAXN82+ 650W 80Plus Bronze | Fiio E07K | Grado SR80i | Cooler Master XB HAF EVO | Logitech G27 | Logitech G600 | CM Storm Quickfire TK | DualShock 4 |

Link to comment
Share on other sites

Link to post
Share on other sites

Steve needs to get to the overclocking/undervolting, that's where this card (will) shine. But I think it will take him a while because he won't have been able to do it with the broken press drivers. Only release day drivers allowed it and overclocking is still very glitchy.

I ran some Time Spy this morning and here's the results. I didn't do the CPU test as my CPU is 7 years old and tanks in that test.

At stock speeds and undervolted to 975mv:

1407793262_timespyundervolt975mv.thumb.jpg.5a5995726f0606c005901c5b4ceb1041.jpg

Overclocked to around 2030-2050 and 1200 mem: (Wattman shows a higher speed but it only ever peaks at 50 less than the peak number)

602019003_timespyoverclock2ghz.thumb.jpg.e42558acdd0707208c2d7890dab37575.jpg

Yes that's a 21% improvement on Air. Now we need the drivers to be fixed so that benchmarks can be run in fullscreen mode without the clocks resetting.

Gaming Rig:CPU: Xeon E3-1230 v2¦RAM: 16GB DDR3 Balistix 1600Mhz¦MB: MSI Z77A-G43¦HDD: 480GB SSD, 3.5TB HDDs¦GPU: AMD Radeon VII¦PSU: FSP 700W¦Case: Carbide 300R

 

Link to comment
Share on other sites

Link to post
Share on other sites

3 minutes ago, Madgemade said:

Steve needs to get to the overclocking/undervolting, that's where this card (will) shine. But I think it will take him a while because he won't have been able to do it with the broken press drivers. Only release day drivers allowed it and overclocking is still very glitchy.

I ran some Time Spy this morning and here's the results. I didn't do the CPU test as my CPU is 7 years old and tanks in that test.

At stock speeds and undervolted to 975mv:

1407793262_timespyundervolt975mv.thumb.jpg.5a5995726f0606c005901c5b4ceb1041.jpg

Overclocked to around 2030-2050 and 1200 mem: (Wattman shows a higher speed but it only ever peaks at 50 less than the peak number)

602019003_timespyoverclock2ghz.thumb.jpg.e42558acdd0707208c2d7890dab37575.jpg

Yes that's a 21% improvement on Air. Now we need the drivers to be fixed so that benchmarks can be run in fullscreen mode without the clocks resetting.

welcome amd to the 2ghz clan

Link to comment
Share on other sites

Link to post
Share on other sites

 

Seems the Launch Driver was a lot more stable than the Review Driver. AMD really needs to fix that problem for Navi.

Link to comment
Share on other sites

Link to post
Share on other sites

44 minutes ago, Madgemade said:

Steve needs to get to the overclocking/undervolting, that's where this card (will) shine. But I think it will take him a while because he won't have been able to do it with the broken press drivers. Only release day drivers allowed it and overclocking is still very glitchy.

I ran some Time Spy this morning and here's the results. I didn't do the CPU test as my CPU is 7 years old and tanks in that test.

At stock speeds and undervolted to 975mv:

1407793262_timespyundervolt975mv.thumb.jpg.5a5995726f0606c005901c5b4ceb1041.jpg

Overclocked to around 2030-2050 and 1200 mem: (Wattman shows a higher speed but it only ever peaks at 50 less than the peak number)

602019003_timespyoverclock2ghz.thumb.jpg.e42558acdd0707208c2d7890dab37575.jpg

Yes that's a 21% improvement on Air. Now we need the drivers to be fixed so that benchmarks can be run in fullscreen mode without the clocks resetting.

RX Vega 56 left a lot of room for the Tinkerers, and Radeon VII follows right along. Most of the Cards will undervolt like champs, but AMD's much smaller production run means they have to clock everything to the point that they can use all of the dies. (I think this is something we forget when it comes to AMD GPUs.) Most users will get a solid Underclock, as a result.

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now


×