AMD Navi Product Reviews and Previews: (5500, 5600 XT, 5700, 5700 XT)

[DavidGraham]

Fury X is hitting 4GB memory limit hard in many cases

Yeah I agree, but that doesn't fully explain why it will fall behind the regular 980 with also 4GB memory. It seems to fit a geometry bottleneck too.

The situation is the same with Kepler, it hit memory limit quickly coupled with a large defecit in compute throughput, especially in DX12 and Vulkan titles.

 

[Davros]

On the navi cards whats with the dent on the top like it's been hit with a axe /rant mode off

 

[bridgman]

Do you, or anyone, think it would make sense to keep iterating on Vega for this space/compute efficiency?

For a while yes, but not forever...

In that sense having a compute centric card would be to their benefit I'd think, less leakage from parts that go basically unused. But if AMD puts a majority of its investment into rDNA, how long before a Navi design is more power efficient regardless?

... for exactly this reason.

On the navi cards whats with the dent on the top like it's been hit with a axe /rant mode off

My original assumption was that we added it to make it fit better in Dr. Su's hand, but supposedly the corresponding internal guide helps with airflow as well.

 

[digitalwanderer]

My original assumption was that we added it to make it fit better in Dr. Su's hand, but supposedly the corresponding internal guide helps with airflow as well.

 

[w0lfram]

I suspect, all the current 5000 series to get a discount at CES. With big navi coming in at VII price point.

I would think the 5700xt @ $349.

 

[techuse]

Covers the various pricing tiers. Hard to disagree with.

 

[pharma]

 

[DavidGraham]

There are very few high end titles that AMD doesnt outperform pascal in these days. You will mostly be limited to Ubisoft games.

The recently released Detroit game shows a huge advantage for Pascal over Vega.
https://gamegpu.com/rpg/ролевые/detroit-become-human-test-gpu-cpu

Looking at the latest TPU performance summary, a GTX 1080 has regressed all the way to Vega 56 levels at 4K.

4K was never the GTX 1080 strong, nor any card of that level anyway. So it really doesn't matter that much at 4K.

At any rate, Computerbase did a fresh 2020 GPU comparison with all new games from 2019, retesting all cards with latest game versions and drivers.

Results:
GTX 1080 is neck and neck with Vega 64, up to 1440p. At 4K both were omitted from the charts.
https://www.computerbase.de/2019-12...rs-2020/2/#abschnitt_benchmarks_in_2560__1440

 

[Kaotik]

The recently released Detroit game shows a huge advantage for Pascal over Vega.
https://gamegpu.com/rpg/ролевые/detroit-become-human-test-gpu-cpu


4K was never the GTX 1080 strong, nor any card of that level anyway. So it really doesn't matter that much at 4K.

At any rate, Computerbase did a fresh 2020 GPU comparison with all new games from 2019, retesting all cards with latest game versions and drivers.

Results:
GTX 1080 is neck and neck with Vega 64, up to 1440p. At 4K both were omitted from the charts.
https://www.computerbase.de/2019-12...rs-2020/2/#abschnitt_benchmarks_in_2560__1440

In other words with new drivers and games Vega 64 has caught up with GTX 1080, which at Vega 64 launch was solid 4-5 % faster at 1080p & 1440p at the same site

 

[DavidGraham]

In other words with new drivers and games Vega 64 has caught up with GTX 1080, which at Vega 64 launch was solid 4-5 % faster at 1080p & 1440p at the same site

Yes, but 5% up and down are to be expected anyway in any GPU segment, this is a far cry from the Kepler or Fiji era, where an entire class of GPU was demoted to a lower tier than before.

 

[techuse]

The recently released Detroit game shows a huge advantage for Pascal over Vega.
https://gamegpu.com/rpg/ролевые/detroit-become-human-test-gpu-cpu


4K was never the GTX 1080 strong, nor any card of that level anyway. So it really doesn't matter that much at 4K.

At any rate, Computerbase did a fresh 2020 GPU comparison with all new games from 2019, retesting all cards with latest game versions and drivers.

Results:
GTX 1080 is neck and neck with Vega 64, up to 1440p. At 4K both were omitted from the charts.
https://www.computerbase.de/2019-12...rs-2020/2/#abschnitt_benchmarks_in_2560__1440

I disagree that 1080 was never meant to be used at 4k. Nvidia marketed it specifically for 4k and/or 1440p high refresh rare gaming. Besides, 4K performance tends to represent performance at lower resolutions as games become more demanding. Im not seeing the huge Detroit advantage. The game has severe CPU bottlenecking issues so at 1080p AMD gpus are artificially hamstrung.

The Computerbase article is absolutely a valid data point. Those guys are consistently among the most reliable and reputable websites out there. looking at the individual results that comprise the average i still stand by my opinion that Vega typically outperforms in newer, more advanced titles. I also went back and checked the launch results for Vega 64 and it was slower than a 1080 at all 3 resolutions in the performance summary, even 4K.

 

[w0lfram]

There is no question that Vega has gotten better over time. Who is arguing against that..?

Question I have is, who cares? Vega is no longer AMD's gaming uArch, rDNA is. And little ole Navi10 sometimes beats up on Vega VII in games. And only getting better with each driver release. Yields seem to be good and "bigger" navi is coming in a months time.

I would expect to see rdna(1) cards to get a price cut, when rdna(2) hits the shelves. Vega is obsolete.

 

[techuse]

There is no question that Vega has gotten better over time. Who is arguing against that..?

Question I have is, who cares? Vega is no longer AMD's gaming uArch, rDNA is. And little ole Navi10 sometimes beats up on Vega VII in games. And only getting better with each driver release. Yields seem to be good and "bigger" navi is coming in a months time.

I would expect to see rdna(1) cards to get a price cut, when rdna(2) hits the shelves. Vega is obsolete.

It wasnt about Vega getting better, my stance is that Nvidia GPUs have historically aged poorly compared to AMD and that should be a purchasing factor for those who dont upgrade every generation.

 

[gamervivek]

Looking at 5500XT AT review, its tessellation performance is not half of 5700XT, 5700XT's is not even 50% higher,

https://www.anandtech.com/show/15206/the-amd-radeon-rx-5500-xt-review/8

https://www.anandtech.com/show/14618/the-amd-radeon-rx-5700-xt-rx-5700-review/14

Is the single SE doing better in 5500XT or the drivers were not updated for 5700XT? Or 5500XT is a double SE design doing poorly?

It is not... Like at all. It's on par with 1650 Super (which actually has 15% lower mem BW), while the 1660 Super is nearly 30% faster.

https://www.techpowerup.com/review/sapphire-radeon-rx-5500-xt-pulse-4-gb/27.html

The 1660 Super is even 42% faster at 4K where the memory BW difference is likely showing its strenght.

The 4GB is most likely running out of VRAM and running PCIE 3.0 x8, pcgh have shown huge performance improvements for the 5500XT 4GB models in such scenarios with PCIE4.0.

https://www.pcgameshardware.de/Rade...pecials/PCI-Express-3-vs-PCI-E-4-GPU-1339415/

TPU's MSI 5500XT 8GB review has 1660 super only 18% faster at 4k though in that case the AMD card has the VRAM advantage.

 

[Kaotik]

Looking at 5500XT AT review, its tessellation performance is not half of 5700XT, 5700XT's is not even 50% higher,

Is the single SE doing better in 5500XT or the drivers were not updated for 5700XT? Or 5500XT is a double SE design doing poorly?

RDNA has one geometry engine regardless of how many shader engines the chip has.

 

[DavidGraham]

The 4GB is most likely running out of VRAM and running PCIE 3.0 x8, pcgh have shown huge performance improvements for the 5500XT 4GB models in such scenarios with PCIE4.0.

Why the hell would AMD limit the card to PCIe 8X in the first place?

 

[Frenetic Pony]

In other words with new drivers and games Vega 64 has caught up with GTX 1080, which at Vega 64 launch was solid 4-5 % faster at 1080p & 1440p at the same site

At least so far? Can imagine 2 would be a possibility for a bigger GPU for the 7nm+/Raytracing update, would seem kind of necessary really.

Anyway, supposed 5600 benchmarks: https://www.notebookcheck.net/Leake...5600-XT-beat-Nvidia-GTX-1660-Ti.448585.0.html

 

[gamervivek]

RDNA has one geometry engine regardless of how many shader engines the chip has.

So that is effectively reducing the tessellation performance of the bigger chip because the RDNA white paper mentions the primitive units being responsible for ff tessellation. And they're one per shader array.

Why the hell would AMD limit the card to PCIe 8X in the first place?

Same as with RX560, smaller chips not doing the full X16.

 

[DavidGraham]

Same as with RX560, smaller chips not doing the full X16.

This is a mess, NVIDIA doesn't do this, even the lowly regular 1650 enjoys a full 16X PCIe speed. The 5500 XT is limited by it's 8X speed, not by PCIe 3.

Hardware Unboxed did a further investigation into this, the uplift from PCIe 4 is game dependent really, if the game is pushing insane amount of textures, the 4GB version will deal with the main RAM, saturating the 8X PCIe speed more, in these situations it's often way behind the 1650 Super (with it's full 16X PCIe speed), so the uplift in performance is barely helping the 5500XT 4GB catch up to the 1650 Super.

Interesting also this doesn't happen with an Intel system, the 5500XT 4GB is often faster with an Intel CPU than any PCIe 4 AMD CPU.

Again naturally, the RX 580 and 1650 Super cards never suffer any of this messy problems, as such I deem them the much more preferable cards than the 5500 XT.

 

[Rootax]

It's weird. Are pci-e lanes disabled because some other part of the gpu are ? If not, what is the point of disabling pci-e lanes ? Power usage ? *mumble* Where is agp texturing when you need it *mumble*