Radeon 5700 Reviews Thread

[AtenRa]

[videocardz.com] MSI to launch seven custom Radeon RX 5700 graphics cards

https://videocardz.com/newz/msi-to-launch-seven-custom-radeon-rx-5700-graphics-cards

 

[guachi]

I was thinking about the 5700, but I think you just convinced me to get the 5700 XT. You're absolutely right. An extra $50 won't kill me. My last bar/dinner tab was $60.

The 5700XT costs 1/7th more and it's 1/8th faster so the performance increase almost equals the price increase.

 

[Timorous]

When looking at the Anandtech review of the 2070s and the 5700XT what strikes me is that the clockspeeds for the XT are far more variable.

When looking at 1440p results (for some reason it tanks at 4k) it ties two games with a clockspeed deficit (TR and F1), wins two games with a clockspeed deficit (Metro and Forza), it nearly ties one game with a slight clockspeed advantage (AC) and loses in GTA V with a small clock advantage. With the other 3 game it loses with a small deficit in clockspeed.

If it could maintain 1875Mhz in most games like the 2070s can I think it would pretty much average out to be neck and neck just like the computerbase.de results for the 1.5Ghz 5700 and the 2070.

At 4k it gets crushed though, hopefully this is some sort of driver issue rather than an architectural one although even if it is i would hope they can fix it for rdna2.

 

[Stuka87]

When looking at the Anandtech review of the 2070s and the 5700XT what strikes me is that the clockspeeds for the XT are far more variable.

When looking at 1440p results (for some reason it tanks at 4k) it ties two games with a clockspeed deficit (TR and F1), wins two games with a clockspeed deficit (Metro and Forza), it nearly ties one game with a slight clockspeed advantage (AC) and loses in GTA V with a small clock advantage. With the other 3 game it loses with a small deficit in clockspeed.

If it could maintain 1875Mhz in most games like the 2070s can I think it would pretty much average out to be neck and neck just like the computerbase.de results for the 1.5Ghz 5700 and the 2070.

At 4k it gets crushed though, hopefully this is some sort of driver issue rather than an architectural one although even if it is i would hope they can fix it for rdna2.

I think this is where aftermarket cards will really shine. They will be able to keep temps down, which should allow it to boost more often.

 

[.vodka]

https://www.computerbase.de/2019-07/radeon-rx-5700-xt-test/

Equalizing core count and frequency (TFLOPS when not possible), memory bandwidth:

Credits to this guy over at B3D for adding the percentages.

RDNA is REALLY good, equal to Turing in efficiency. As you can see it wipes the floor with both Vega and Polaris GCN when it comes to games. This is AMD's Maxwell.

It already has this kind of performance on launch drivers... There's tuning to be done going forward. Things are looking nice.

RDNA2 next year adds hardware accelerated RT support to the mix. It remains to be seen what nV has in store for 7nm, of course, but this is one hell of an upgrade architecture wise.

Big Navi, when 7nm allows for bigger chips at decent yields, definitely has a chance of hitting 2080Ti levels of performance (if not better?)



I'm excited for the GPU market again. I'm even more excited that this architecture is powering next gen consoles. I'm even more excited that Zen2 is the CPU for these consoles (better AI, better everything non graphics wise over netbook-grade Jaguar CPUs)

Games this coming decade should also wipe the floor with what we've had so far.

 

[IEC]

Preliminary look at 5700/XT by Wendell and EposVox for streaming/content creation (it's buggy, but promising):

Some highlights:
1) OBS w/AMF plugin is kinda broken right now, need clean install of new version of plugin
2) FFMPEG basically broken for now (looks like trash)
3) Hardware AMF encoder is equivalent to x.264 Fast, which is an improvement over previous generations including Radeon VII
4) HEVC encoding is "mind-blowing" - doing 5 or 6 FFMPEG encodes at once
5) The new CAS feature (Content Adaptive Sharpening, 1440p upscaling to 4K) is very good without noticeable blurring or over-sharpening - may be a viable option for playing on 4K HDTVs while getting 1440p performance
6) Radeon Anti-Lag is different from Fast Sync - they used high speed camera setup to capture the difference. Best case scenario vs nV (for nV): AMD is 7ms ahead / Worst case: AMD is 16-17ms ahead (!)

So for competitive gaming/e-sports titles, Anti-Lag can actually offer a surprisingly big difference. Of note: there probably isn't anything that would stop nV from making a similar feature

 

[Leadbox]

Preliminary look at 5700/XT by Wendell and EposVox for streaming/content creation (it's buggy, but promising):

Some highlights:
1) OBS w/AMF plugin is kinda broken right now, need clean install of new version of plugin
2) FFMPEG basically broken for now (looks like trash)
3) Hardware AMF encoder is equivalent to x.264 Fast, which is an improvement over previous generations including Radeon VII
4) HEVC encoding is "mind-blowing" - doing 5 or 6 FFMPEG encodes at once
5) The new CAS feature (Content Adaptive Sharpening, 1440p upscaling to 4K) is very good without noticeable blurring or over-sharpening - may be a viable option for playing on 4K HDTVs while getting 1440p performance
6) Radeon Anti-Lag is different from Fast Sync - they used high speed camera setup to capture the difference. Best case scenario vs nV (for nV): AMD is 7ms ahead / Worst case: AMD is 16-17ms ahead (!)

So for competitive gaming/e-sports titles, Anti-Lag can actually offer a surprisingly big difference. Of note: there probably isn't anything that would stop nV from making a similar feature

Doesn't the fact that nvidia have software scheduling vs Amd's hardware sheduling mean that they cannot go any faster than they already do in the best case scenerio? They claim to have had this feature or something similar for yonks.

 

[PontiacGTX]

I haven't seen one yet. The most relevant search result is me asking the same question

https://news.mydrivers.com/1/635/635495_2.htm

amd finally has a comparable 32 bit int performance to nvidia

 

[Bouowmx]

https://news.mydrivers.com/1/635/635495_2.htm

amd finally has a comparable 32 bit int performance to nvidia

Thanks for the link. But, I don't think you read the chart with AIDA64 right. Radeon RX 5700 XT 32-bit int ops much less than 32-bit FLOPS. 24-bit int ops is fine, but AMD has had full-rate INT24 for a while now.

 

[PontiacGTX]

Thanks for the link. But, I don't think you read the chart with AIDA64 right. Radeon RX 5700 XT 32-bit int ops much less than 32-bit FLOPS. 24-bit int ops is fine, but AMD has had full-rate INT24 for a while now.

maybe I was recalling Pascal performance , now I check Turing is much better, sorry I had in mind this chart
https://www.servethehome.com/wp-content/uploads/2019/03/Nvidia-Titan-RTX-AIDA64-GPGPU-Part-2.jpg but well I mis remember the digits I thought the GTX 1080 was closer

 

[Bouowmx]

The reference model uses graphite pad, not TIM. Another baffling decision

Spoiler: Gamers Nexus

 

[railven]

The reference model uses graphite pad, not TIM. Another baffling decision

Spoiler: Gamers Nexus

In another thread someone told me costs are sky rocketing. Explains why AMD couldn't include the washers. Haha.

Joking aside, this cooler is a joke, and AMD deserves all the flak for trotting it out. If you watched his tear down video, he tried to remove the dent from the shroud (which is 100% cosmetic).

 

[Stuka87]

The reference model uses graphite pad, not TIM. Another baffling decision

Spoiler: Gamers Nexus

The vast majority of GPU's use thermal pads. But not all thermal pads are created equally. TIM doesn't always age well. Thermal pads can last indefinately in most cases.

The issue here is that the pad is thicker than it should be, and the pressure that is applied to the cooler is not adequate. Often these two items would not be an issue if the cooler could pull heat away fast enough to overcome this. But most blowers cannot.

 

[tviceman]

RDNA is REALLY good, equal to Turing in efficiency. As you can see it wipes the floor with both Vega and Polaris GCN when it comes to games. This is AMD's Maxwell.

It already has this kind of performance on launch drivers... There's tuning to be done going forward. Things are looking nice.

RDNA2 next year adds hardware accelerated RT support to the mix. It remains to be seen what nV has in store for 7nm, of course, but this is one hell of an upgrade architecture wise.

Big Navi, when 7nm allows for bigger chips at decent yields, definitely has a chance of hitting 2080Ti levels of performance (if not better?)



I'm excited for the GPU market again. I'm even more excited that this architecture is powering next gen consoles. I'm even more excited that Zen2 is the CPU for these consoles (better AI, better everything non graphics wise over netbook-grade Jaguar CPUs)

Games this coming decade should also wipe the floor with what we've had so far.

Navi introduced a huge boost in efficiency - much larger than what past AMD architectures presented. In my opinion, Navi SHOULD win on efficiency because it's on a full generational node ahead of Turing. But it isn't winning; it's just tying. To me, it's amazing that Turing is able to keep up with Navi despite being on a 5 year old fabrication process, vs. Navi being on a 1 year old process.

Big Navi performance will entirely depend on how much TDP headroom AMD wants to push. AMD made the right move getting to 7nm 9+ months before Nvidia, so they need to do everything they can to take advantage of their competitive position before Nvidia releases Ampere. Hopefully big Navi can get here with a decent amount of time before Ampere hits, because Ampere's x04 GPU will be around 50% faster than the RTX 2080, if going by historical gains (making it about 80-85% faster than the 5700 XT).

 

[IntelUser2000]

To me, it's amazing that Turing is able to keep up with Navi despite being on a 5 year old fabrication process, vs. Navi being on a 1 year old process.

Are we in 2021?

16nm isn't 5 years old.

 

[ozzy702]

Are we in 2021?

16nm isn't 5 years old.

Didn't TSMC technically start 16nm in late 2013?

 

[IntelUser2000]

Didn't TSMC technically start 16nm in late 2013?

Yes, but products came much later. 20nm Apple chips were late 2014, and 20nm is really a half node process. Snapdragon 820 was 2015.

Not to mention GPUs are later and it has different requirements so you can't put them in the same categories. 16nm Pascal was in 2016.

 

[DarthKyrie]

1. AMD and EPIC already working together
2. If you combine AMD PC & Consoles they have the majority of the market vs NV

It's going to get even worse because now AMD tech will be going into the mobile side because of the deal with Samsung.

 

[GodisanAtheist]

It's going to get even worse because now AMD tech will be going into the mobile side because of the deal with Samsung.

True, but the wintel hegemony on PCs ensures that AMD isn't going to see the benefits directly translate to PC (unless Vulcan somehow becomes the defacto API for everything including Windows).

 

[DarthKyrie]

True, but the wintel hegemony on PCs ensures that AMD isn't going to see the benefits directly translate to PC (unless Vulcan somehow becomes the defacto API for everything including Windows).

Don't forget that MS has every incentive to maximize the Windows 10 kernel to harvest the most performance out of Zen2 because of Project Scarlett. This is the first time that MS has used the same kernel for both the PC and XBox while the XBox is utilizing a powerful AMD CPU.

 

[AtenRa]

because Ampere's x04 GPU will be around 50% faster than the RTX 2080, if going by historical gains (making it about 80-85% faster than the 5700 XT).

Well first, NAVI 10 is not a direct competitor to current x04 GPU. RTX2080 is using the much bigger TU104 that has 13.6B transistors with a die size of 545mm2. In comparison NAVI 10 only has 10.3B transistors with a die size of 251mm2. NAVI 10 directly competes with TU106 that has 10.8B transistors and comes with a die size of 445mm2.
Now when NVIDIA will go to 7nm/7nm+ the NAVI 10 competitor will be the x06 again and not the x04.

Secondly, now if you want to get 50% more performance out of the TU106 found in RTX2060 Super then they will have to increase the transistor count, making the die bigger than NAVIs 10 251mm2.

For example lets take the GTX980 vs the GTX1080,
GTX980 has 5.2B transistors with a die size of 398mm2 at 28nm and the GTX1080 came with 7.2B transistors with a die size of 314mm2 at 16nm.
So if they will increase the transistor count of the x06 Ampere they could be come close to or bigger than 300mm2 and that will compete with another NAVI and not NAVI 10.

 

[Tarkin77]

https://www.tomshw.de/2019/07/11/un...erplaytables-fuer-die-rx-5700-und-rx-5700-xt/

Igors Lab has posted SoftPowerPlayTables for Radeon RX 5700 und 5700 XT

He achieved really nice results! Core to 2.2GHz with a power draw of 250w and 2080 FE like performance! ... and thats without overclocking the Memory!

 

[PhonakV30]

damn , at clock 2200mhz , Temp is quite low.

 

[n0x1ous]

neat to see whats its capable of but he notes that its at dangerous voltage levels that are unsustainable.

 

[EXCellR8]

but at least his Kuhlung system is on point