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

[JoeJ]

Process twice expensive, but smaller chips. Adding HBM2 is more expensive than GDDR6(?), the difference might be not that big.

It's just - looking at price, performance and features, Navi isn't the leap i hoped to happen for next gen, leaving me a bit disappointed.
But rumors about RDNA2 sound good. I'll wait for that.

 

[Kaotik]

Process twice expensive, but smaller chips. Adding HBM2 is more expensive than GDDR6(?), the difference might be not that big.

It's just - looking at price, performance and features, Navi isn't the leap i hoped to happen for next gen, leaving me a bit disappointed.
But rumors about RDNA2 sound good. I'll wait for that.

Smaller chips, yes, but the density hasn't doubled while the price (pretty much) has. Producing exact same chip is more expensive on 7nm.

 

[no-X]

No. The graph is related to 250mm² dies. It's not possible to say, that 500mm² 14nm die is less expensive than 250mm² 7nm die (based on the graph), because yields decrease exponentially with die size (the relation is not linear). We can say that one 250mm² 7nm die almost as expensive as two 250mm² 14nm dies. But we can't say that one 500mm² 14nm die is just as expensive as two 250mm² 14nm dies. Two smaller dies will have higher yield than one big, so the big one will be more expensive.

 

[Kaotik]

No. The graph is related to 250mm² dies. It's not possible to say, that 500mm² 14nm die is less expensive than 250mm² 7nm die (based on the graph), because yields decrease exponentially with die size (the relation is not linear). We can say that one 250mm² 7nm die almost as expensive as two 250mm² 14nm dies. But we can't say that one 500mm² 14nm die is just as expensive as two 250mm² 14nm dies. Two smaller dies will have higher yield than one big, so the big one will be more expensive.

True, didn't think of that part, but considering that on 250mm² dies the price almost doubled, yet you can fit only ~65% more transistors in 1 mm² (based on Vega10/Navi10 comparison), it's still likely the 7nm ends up more expensive, especially since mature 16/14/12nm process probably has better yields compared to relatively fresh 7nm

 

[Bondrewd]

especially since mature 16/14/12nm process probably has better yields compared to relatively fresh 7nm

No, N7 is in fact the best yielding node in TSMC history.
The best ramp ever, too.

 

[Kaotik]

No, N7 is in fact the best yielding node in TSMC history.
The best ramp ever, too.

Fastest ramp, yes for now (according to TSMC 5nm will be faster), but last time I heard they were "expecting to match 12/16nm yields in 2019" but haven't heard actual confirmation if they ever did. Also AMD didn't really use TSMC 12/16nm, they used Samsung/Glofo 14/12 which might have had better yields, too.
edit: and there would still be some road to go, since the price for mm² nearly doubled but density only grew around 65%

 

[Deleted member 13524]

Price for sq-mm doubled is applied to first wafers or lifetime orders?

Seems odd to me that with TSMC constantly retrofitting their fabs to support newer nodes, theoretically they would just keep doubling their revenue after each transition. This would mean a level of revenue growth that shouldn't really be sustainable for long.

 

[Entropy]

Price for sq-mm doubled is applied to first wafers or lifetime orders?

Price per wafer drops over time (as does defect rate). TSMC prices were reported to have dropped below $10000/wafer last autumn in spite of high order pressure. And Samsung is rumoured to have dropped their prices considerably below TSMCs in order to compete and improve their process utilization.
The effective cost/mm2 depends on a lot of factors, some of them changing over time.
Any argument based on a single approximate (and dated) figure is nigh on useless, and it doesn’t get better that some of the numbers referred to have some kind of baked in assumtions about design costs and production volume.
All other things being equal, a final mm2 on TSMC 7nm is obviously more expensive than a mm2 at 16nm. But all things aren’t going to be equal, and again, the ratios will change over time. Trying to draw conclusions without being close to having access to the necessary information is an exercise in futility.

 

[Kaotik]

https://www.io-tech.fi/artikkelit/testissa-amd-radeon-rx-5600-xt/

Our 5600XT Review is out
Also
https://www.techpowerup.com/review/sapphire-radeon-rx-5600-xt-pulse/
https://www.anandtech.com/show/15422/the-amd-radeon-rx-5600-xt-review
https://www.computerbase.de/2020-01/radeon-rx-5600-xt-test/
https://www.guru3d.com/articles-pages/sapphire-pulse-radeon-rx-5600-xt-6g-review,1.html
https://www.guru3d.com/articles-pages/asus-strix-radeon-rx-5600-xt-review,1.html
https://www.guru3d.com/articles-pages/gigabyte-radeon-rx-5600-xt-gaming-oc-review,1.html

With the BIOS update they announced after RTX 2060 price cut, the card matches RTX 2060 performance (or even slightly passes it) while being cheaper.

 

[Deleted member 2197]

RX 5600XT reviews (VideocardZ)

AnandTech SAPPHIRE Pulse
Benchlife SAPPHIRE Pulse
Benchmark SAPPHIRE Pulse
Bitwit POWERCOLOR Red Dragon
BPS Customs SAPPHIRE Pulse
ComputerBase SAPPHIRE Pulse
ComptoirHardware SAPPHIRE Pulse
Eteknix (video) ASUS TUF X3 Gaming
Expreview XFX Thicc II Pro
GamersNexus SAPPHIRE Pulse
GamerMeld XFX Thicc II Pro
GearSeekers SAPPHIRE Pulse
Golem SAPPHIRE Pulse
Greg Salzar SAPPHIRE Pulse
Guru3D SAPPHIRE Pulse
ASUS ROG STRIX TOP
GIGABYTE Gaming OC
HardwareBattle ASUS TUF X3 Gaming
HardwareCanucks SAPPHIRE Pulse
Hardware.info SAPPHIRE Pulse
HardwareUnboxed SAPPHIRE Pulse
MSI Gaming X
HardwareUpgrade (video) SAPPHIRE Pulse
HardwareLuxx ASUS TUF Gaming
SAPPHIRE Pulse
HardwareZone SAPPHIRE Pulse
Hexus SAPPHIRE Pulse
HKEPC SAPPHIRE Pulse
HotHardware SAPPHIRE Pulse
HwBox SAPPHIRE Pulse
igorl’sLAB POWERCOLOR Red Dragon
JayzTwoCents POWERCOLOR Red Dragon
KitGuru (video) SAPPHIRE Pulse

RX 5600 XT reviews (VideocardZ)

Lab501 SAPPHIRE Pulse
Optimum Tech XFX Thicc II Pro
Overclock3D SAPPHIRE Pulse
Overclockers XFX Thicc II Pro
PCLab SAPPHIRE Pulse
PC Games Hardware (video) SAPPHIRE Pulse
PCGamesN SAPPHIRE Pulse
PCWorld SAPPHIRE Pulse
PCPerspective SAPPHIRE Pulse
Phoronix SAPPHIRE Pulse
Pro Hi-Tech SAPPHIRE Pulse
PurePC SAPPHIRE Pulse
Quasarzone XFX Thicc II Pro
SweClockers SAPPHIRE Pulse
TechPowerUP SAPPHIRE Pulse
ASUS ROG STRIX
Tech YES City ASUS TUF Gaming
The FPS Review XFX Thicc II Pro
ThinkComputers MSI Gaming X
Tom’s Hardware SAPPHIRE Pulse
UNIKO’s Hardware SAPPHIRE Pulse
WASD (video) SAPPHIRE Pulse
Vortez (video) SAPPHIRE Pulse
XanxoGaming GIGABYTE Gaming OC
Xfastest Hong Kong SAPPHIRE Pulse
Xfastest Taiwan ASUS TUF X3 Gaming

 

[DegustatoR]

With the BIOS update they announced after RTX 2060 price cut, the card matches RTX 2060 performance (or even slightly passes it) while being cheaper.

 

[Deleted member 2197]

However the card does trail the RTX 2060 by 6%, which although is a smaller gap, it’s also about as much as the price difference between the cards. So stock-to-stock, AMD and NVIDIA have managed to almost perfectly align their products on a price-performance basis. This means that there are no bargains here between the RX 5600 XT and the RTX 2060 (or even the RX 5700), but no one is clearly trailing the pack either.

https://www.anandtech.com/show/15422/the-amd-radeon-rx-5600-xt-review

 

[Kaotik]

Spoiler

Spoiler

https://www.anandtech.com/show/15422/the-amd-radeon-rx-5600-xt-review

AMD has said all along this is 1080p card rather than 1440p card. Regardless, in our tests RX 5600 XT was faster than RTX 2060 (NVIDIA Founders Edition so factory OCd by default) at both 1080p and 1440p in every test except for Shadow of the Tomb Raider.
TPU lists 5600 XT slightly faster at both resolutions too, as does ComputerBase
Haven't crawled through the rest yet

 

[DegustatoR]

RX 5600 XT was faster than RTX 2060 (NVIDIA Founders Edition so factory OCd by default)

2060 FE spec is reference, there's no OC there. Then again, AMD's new BIOS is for OCed cards, as mentioned by Anandtech:

Now, to be sure, AMD has not changed the reference specifications for the Radeon RX 5600 XT. So despite what’s going on, the baseline hasn’t changed. Case in point: the Pulse 5600 XT’s quiet mode BIOS has the same 135W TGP (~150W TBP) power limit both with the old and new BIOS, as well as the same 1460MHz rated game clock.

I mean, the cards are basically even in performance in current titles but 5600XT trades it's lack of new features for $20 lower price. Seems like a fair trade to me. Wouldn't buy one personally still, don't see $20 as a huge save.

 

[Ryan Smith]

Personal aside: while I get why AMD made the last-minute BIOS changes that they did, I am rather concerned that it's going to make a mess of things in the long run. Not all RX 5600 XT cards are getting the 14Gbps factory overclock on the memory. Those cards perform a lot better, but they are well above what AMD actually guarantees as far as performance and specifications are concerned.

(The TechSpot graph embedded above is a great example of why this is a potential problem)

 

[Deleted member 2197]

One review without the new bios due to time constraints.
https://www.thefpsreview.com/2020/01/21/xfx-radeon-rx-5600-xt-thicc-ii-pro-review/

Edit: Bios update on Linux made performance worse.

I alerted the AMD Linux folks but unfortunately as of writing have not received any firm resolution as of launch day. But long story short, the new vBIOS that Sapphire is recommending for consumers sadly seems to mess up the Linux performance for the worse...

https://www.phoronix.com/scan.php?page=article&item=linux-rx5600xt-amd&num=2

 

[Lightman]

Well, we finally have a proper price and performance fight in $250 - $300 price segment!
It's messy with last minute BIOS updates, but it brings decent value into this segment.
I can see 5600XT going down to under $250 spot in couple of months time. Compared to RX580 it offers very decent uplift (finally) in the same MSRP bracket.

Now lets wait for 2nd half of 2020 to see the same fight in higher end of market!

 

[OlegSH]

RTX 2060 (NVIDIA Founders Edition so factory OCd by default)

RTX 2060 FE has reference clocks and unlike heavily factory overclocked 5600 XT cards, perf gains from RTX 2060 overclocking are well higher than 2%

 

[Deleted member 2197]

RTX 2060 FE has reference clocks and unlike heavily factory overclocked 5600 XT cards, perf gains from RTX 2060 overclocking are well higher than 2%

True, as can be seen from the GPU Shootout graph depicting RTX 2060 FE and AIB custom 2060's (MSI, Palit and Asus).

Spoiler: Shootout Graph

 

[Kaotik]

RTX 2060 FE has reference clocks and unlike heavily factory overclocked 5600 XT cards, perf gains from RTX 2060 overclocking are well higher than 2%

Oh? I thought all RTX 20 FE's were factory OCd. Was it just the highend then?