AMD Radeon Resizable BAR / Smart Access Memory

[Julhes]

Thanks to my rtx 2080 ti bought for € 429 (second-hand)

 

[bug]

It isn't proprietary, stop making things up: https://www.pcgamer.com/amd-smart-access-memory-not-proprietary-promise/

Well, tbh, they haven't even worked with themselves to bring this to Zen2 CPUs.
But I was suspecting that's a temporary situation and they just focused on one CPU gen to get to the market faster.

Plus, now that we see it doesn't actually do much difference irl, we can really stop worrying about it. Those who can get it will need to turn on their monitoring HUD to see the few fps they enjoy, the rest can rest assured they're not missing much.

And while typing this, it just hit me not everything is just about fps, I wonder is enabling SAM affects power draw somehow. I'd be more psyched if SAM would allow you to shave off a few W in memory intensive scenarios, rather than gaining a few fps.

 

[mechtech]

It's the default coloring of my charts. Let me fix this

"Red–green color blindness affects up to 8% of males and 0.5% of females of Northern European descent. "

Color blindness - Wikipedia

en.wikipedia.org

May I suggest orange and blue please/thanks

 

[defaultluser]

We have been using the same 256-MB PCIe memory window from the 32-bit days for the last 15 years without a problem.

Making this infinite-sized is not going to suddenly magically make the PCIe bus any faster, you just get slightly higher throughput (due to less protocol overhead). It's like all the hype of Ethernet Jumbo Frames, and yet it does very little (unless you are processor limited)

Once drivers and games get optimized for this, it will be a more reliable performance improvement...but it's still going to be 5% or less.

 

[Steevo]

We have been using the same 256-MB PCIe memory window from the 32-bit days for the last 15 years without a problem.

Making this infinite-sized is not going to suddenly magically make the PCIe bus any faster, you just get slightly higher throughput (due to less protocol overhead). It's like all the hype of Ethernet Jumbo Frames, and yet it does very little (unless you are processor limited)

Once drivers and games get optimized for this, it will be a more reliable performance improvement...but it's still going to be 5% or less.

Less than 5% is all the 3080 is faster than the 6800XT, and SAM is free, so......

It may improve things like F@H and unfortunately also mining efficiency, the former is welcome, the latter can light themselves on fire for all I care. But it would also be impressive to see how SQL lookup times would change if the CPU being able to use GPU memory and hardware.

 

[defaultluser]

Less than 5% is all the 3080 is faster than the 6800XT, and SAM is free, so......

This is not some proprietary solution. I would expect both implementations of this standard feature to give you similar increases in FPS. We were just waiting for the 64-bit conversion to finish before this increase in memory-window size made any sense.

You are not going to magically unlock more FPS from the 6800 than it has today.

 

[Steevo]

This is not some proprietary solution. I would expect both implementations of this standard feature to give you similar increases in FPS. We were just waiting for the 64-bit conversion to finish before this increase in memory-window size made any sense.

You are not going to magically unlock more FPS from the 6800 than it has today.

Did you read the review? It showed "unlocking" 2% performance improvement, and with further implementation I'm sure it will gain a little more. So other than one game engine where is your data to back up that it doesn't improve performance today?

It will be interesting to see how Nvidia implements this and what performance increase they too gain, none of these companies are adding features to slow down their hardware....

 

[zlobby]

Holly schnitzel! Eat $hit, nvidia!

It's the default coloring of my charts. Let me fix this

Where are the comparisons between frame times and input latencies?

 

[W1zzard]

Where are the comparisons between frame times and input latencies?

thanks for reminding, frametimes have been added. Nothing to see there other than just lower frametime with BAR resizing

 

[zlobby]

thanks for reminding, frametimes have been added. Nothing to see there other than just lower frametime with BAR resizing

Sweet! Yer a wizard, W1zzard!

Not ideal, but yes, better.

Ideally, I'd have AMD stick to red, Nvidia to green and Intel to blue. And then I'd use different shades for variations of the same product.

Different shading and fill patterns can also be used.

 

[Space Lynx]

@W1zzard

just an fyi, min frame rates increase a lot with SMA enabled, timestamped linus video for you where he talks about it

 

[HD64G]

So, when using the Intel setup 6800XT loses to 3080 by 4% and when using the 5900X setup it closes the gap to 1% without SAM and wins by 1% using SAM at 1440P on average. Good to know. PCIE4 could play its role there me thinks.

 

[Space Lynx]

So, when using the Intel setup 6800XT loses to 3080 by 4% and when using the 5900X setup it closes the gap to 1% without SAM and wins by 1% using SAM at 1440P on average. Good to know. PCIE4 could play its role there me thinks.

hardware unboxed review of the 6800 xt is the best of all of them I think, I watched them all. ray tracing really doesn't matter to me i just want high frames 1080p/1440p gaming. so yeah. rock on.

 

[arbiter]

So AMD claims big gains like near 10% but in independent review only gives at most 2%. Yea sounds like AMD marketing is still like it used to be.

 

[crazyeyesreaper]

So AMD claims big gains like near 10% but in independent review only gives at most 2%. Yea sounds like AMD marketing is still like it used to be.

Well to be honest there are games that do see such a boost.

Detroit Become Human
1080p 8% performance gain.
1440p 7%
4k 4%

Gears 5
1080p 9%
1440p 11%
4k 6%

Hitman 2
1080p 14%
1440p 11%
4k 7%

Granted 3 games out of many is pretty limited. However, its likely with time, drivers and game engine enhancements especially with consoles being AMD CPU / GPU we will see SAM offer a more sustained boost. For now its extremely limited but in some situations a 4-14% boost in performance is nothing to sneeze at especially in an age where GPU and CPU overclocking are becoming increasingly redundant due to power limits and other factors. I imagine with Xbox essentially just being a pre-spec PC we will see more titles use SAM and NVIDIA's implementation as well. A free performance bump is a free bump.

 

[Space Lynx]

Well to be honest there are games that do see such a boost.

Detroit Become Human
1080p 8% performance gain.
1440p 7%
4k 4%

Gears 5
1080p 9%
1440p 11%
4k 6%

Hitman 2
1080p 14%
1440p 11%
4k 7%

Granted 3 games out of many is pretty limited. However, its likely with time, drivers and game engine enhancements especially with consoles being AMD CPU / GPU we will see SAM offer a more sustained boost. For now its extremely limited but in some situations a 4-14% boost in performance is nothing to sneeze at especially in an age where GPU and CPU overclocking are becoming increasingly redundant due to power limits and other factors. I imagine with Xbox essentially just being a pre-spec PC we will see more titles use SAM and NVIDIA's implementation as well. A free performance bump is a free bump.

you haven't added in SAM or high tuned ram yet either... i will be getting 20-40% gains due to my ram and SAM... so yeah...I am aware not every will be able to activate SMA, but thats not my problem.

 

[crazyeyesreaper]

you haven't added in SAM or high tuned ram yet either... i will be getting 20-40% gains due to my ram and SAM... so yeah...I am aware not every will be able to activate SMA, but thats not my problem.

Point remains 22 games 4K res settings most of the time its 0 change but the change we do see is a good indicator of the future in those select titles. no one will turn down free performance after all. And considering the AMD cards can actually OC a bit... well i would say they are sitting in a good spot seeing an OC increase performance by 9% is a nice change of pace, add in rage mode for another 1-2% add in SAM for even if its just 2% on average thats still a nice up tick of 12% with the capacity to push up further. for instance in a title at 1440p already seeing a boost of 10-11% via SAM, add in 9% OC and rage mode and oh look you gained 20%. Thats a nice feather in AMD's cap but it needs to become a mainstream aspect not a limited to select scenarios bonus.

 

[zlobby]

@W1zzard , I know you are a skilled coder and you have good knowledge of hardware and how it works on a low level, but can you please elaborate why are you a fan of CSM in 2020? I know UEFI is far from perfect, but still?

Well, tbh, they haven't even worked with themselves to bring this to Zen2 CPUs.
But I was suspecting that's a temporary situation and they just focused on one CPU gen to get to the market faster.

Plus, now that we see it doesn't actually do much difference irl, we can really stop worrying about it. Those who can get it will need to turn on their monitoring HUD to see the few fps they enjoy, the rest can rest assured they're not missing much.

And while typing this, it just hit me not everything is just about fps, I wonder is enabling SAM affects power draw somehow. I'd be more psyched if SAM would allow you to shave off a few W in memory intensive scenarios, rather than gaining a few fps.

I'm more worried that SAM will bring tons of new vulnerabilities for the hax0rz to play with.

you haven't added in SAM or high tuned ram yet either... i will be getting 20-40% gains due to my ram and SAM... so yeah...I am aware not every will be able to activate SMA, but thats not my problem.

It'd be interesting to see how SAM performs with heavily-OC'd VRAM, as well as with speedy system RAM. W1zz, get cracking!

 

[Space Lynx]

It'd be interesting to see how SAM performs with heavily-OC'd VRAM, as well as with speedy system RAM. W1zz, get cracking!

@W1zzard it's none of my business, but I just don't want to see you waste your limited time, imo any ram oc'ing tests should wait for the next BIOS, AMD will be make 4000 1:1 much easier to obtain, etc. i have my ram stable at 3600 cas 14-14-14 with my 5600x, and it is dual ranked. so i think thats as good as it gets right now until next bios. /shrug

Point remains 22 games 4K res settings most of the time its 0 change but the change we do see is a good indicator of the future in those select titles. no one will turn down free performance after all. And considering the AMD cards can actually OC a bit... well i would say they are sitting in a good spot seeing an OC increase performance by 9% is a nice change of pace, add in rage mode for another 1-2% add in SAM for even if its just 2% on average thats still a nice up tick of 12% with the capacity to push up further. for instance in a title at 1440p already seeing a boost of 10-11% via SAM, add in 9% OC and rage mode and oh look you gained 20%. Thats a nice feather in AMD's cap but it needs to become a mainstream aspect not a limited to select scenarios bonus.

hardware unboxed and ars technica review (i think) had much different performance gains for SAM than w1zz did. lot of variables this time around, who knows what the real answer is. either way im enabling SAM when my card gets here, meh why not. also its not about fastest fps with SAM, linustechtips shows us min fps increases a lot with SAM on

 

[crazyeyesreaper]

@W1zzard it's none of my business, but I just don't want to see you waste your limited time, imo any ram oc'ing tests should wait for the next BIOS, AMD will be make 4000 1:1 much easier to obtain, etc. i have my ram stable at 3600 cas 14-14-14 with my 5600x, and it is dual ranked. so i think thats as good as it gets right now until next bios. /shrug



hardware unboxed and ars technica review (i think) had much different performance gains for SAM than w1zz did. lot of variables this time around, who knows what the real answer is. either way im enabling SAM when my card gets here, meh why not. also its not about fastest fps with SAM, linustechtips shows us min fps increases a lot with SAM on

True Gamer's Nexus on the other hand found that SAM had next to no real benefit other than 1 or 2 titles. So its likely highly dependent on the title / game engine and driver optimization.

 

[alres3]

I have a 2700+ B450 and pre-ordered 3080 (maybe I'll get it in a week). To really benefit from SAM I'd need to upgrade my MB to X570 and new CPU, correct? Probably better off with my 3080 coming and gaming at 4k 60fps+

 

[ratirt]

@W1zzard Is everything OK with the charts? For instance the F1 2020 shows some weird numbers.
Is this the lowest quality possible? 486.2 FPS for 3080 at 4k? In the 6800XT review it had 145.5 FPS.

 

[W1zzard]

@W1zzard Is everything OK with the charts? For instance the F1 2020 shows some weird numbers.
Is this the lowest quality possible? 486.2 FPS for 3080 at 4k? In the 6800XT review it had 145.5 FPS.

I copied over my settings from the Intel rig, but when F1 detects a change in CPU, it will reset to different settings. I noticed this only after testing.

Results within this article are perfectly comparable, just not comparable to my other reviews

 

[ratirt]

I copied over my settings from the Intel rig, but when F1 detects a change in CPU, it will reset to different settings. I noticed this only after testing.

Results within this article are perfectly comparable, just not comparable to my other reviews

over 480FPS at 4k That is crazy

 

[Spanners]

LMAO.

AMD sure loves a taste of proprietary just like NVIDIA.

Thanks for the reminder I wasn't logged in.