AMD Radeon RX 5950 XT - Finally a 2080 Ti Killer

viivo said:
A dead news forum would be much worse than occasional stories that are obviously untrue. Keep posting, erek.
Click to expand...

I think, it would be reasonable to speculate based on what we know

AMD's problem has been power consumption. Navi 10(5700) is same size as Polaris, so it should be possible to produce a die which is twice the size. Have they managed to solve their power problems yet?

In the past AMD had to use HBM to cut down on power consumption. Since gddr6 is cheaper than HBM they might go with 384 width memory bus this time. Also they might have to cut down on frequency if power consumption issues are still unresolved


https://hothardware.com/news/amd-radeon-rx-navi-21-gddr6-rdna-2
 
Marees said:
I think, it would be reasonable to speculate based on what we know

AMD's problem has been power consumption. Navi 10(5700) is same size as Polaris, so it should be possible to produce a die which is twice the size. Have they managed to solve their power problems yet?

In the past AMD had to use HBM to cut down on power consumption. Since gddr6 is cheaper than HBM they might go with 384 width memory bus this time. Also they might have to cut down on frequency if power consumption issues are still unresolved


https://hothardware.com/news/amd-radeon-rx-navi-21-gddr6-rdna-2
Click to expand...

That hothardware link literally contradicts itself several times. Starts by saying navi 21 will be twice as fast as 5700xt, then that it will compete with 2080 super/ti(that is not twice as fast.) Goes on to say it may be on par with 2080 super/ti and a significant upgrade to 5700xt, which could mean 10 %. It basically says nothing.
 
Last edited:
Marees said:
I think, it would be reasonable to speculate based on what we know

AMD's problem has been power consumption. Navi 10(5700) is same size as Polaris, so it should be possible to produce a die which is twice the size. Have they managed to solve their power problems yet?

In the past AMD had to use HBM to cut down on power consumption. Since gddr6 is cheaper than HBM they might go with 384 width memory bus this time. Also they might have to cut down on frequency if power consumption issues are still unresolved


https://hothardware.com/news/amd-radeon-rx-navi-21-gddr6-rdna-2
Click to expand...


While the RX 5700 XT stock burns way too much power even think about doubling performance, the underclocked/cut RX 5700 is a lot closer to sanity.

power-gaming-average.png


We could likely power an underclcoked RX 5700 *2 on a 384-bit bus . IF you got perfect scaling, it would match the RTX 2080 Ti:

relative-performance_3840-2160.png


144/74 = the 2080 Ti is 1.95 times faster than the RX 5700 at 4k. So a perfect doubling of performance would be a smidge faster.

By the time you will be able to buy Big Navi, Nvidia will probably have Ampere out. But I can see them announcing it at CES, then releasing it 3-6 months later (They did this with Vega 64 3 years ago).

For additional features/final die size, who the fuck knows?The only thing can say for certain is this will not support both HBM or GDDR6.


Even the RT support is still up in the air, as it could b something as simple as the RX 1660 Ti, or something more involved.
 
Last edited:
defaultluser said:
144/74 = the 2080 Ti is 1.95 times faster than the RX 5700 at 4k. So a perfect doubling of performance would be a smidge faster.
Click to expand...

I think you meant 144/"84"

But still keeping the Vega64 power consumption as a benchmark I think big Navi might match current RTX 2080ti in performance

AMD's track record doesn't inspire confidence of anything more than that. Let us see
 
"With that being said, Navi 21 should offer a significant performance uplift over existing Navi 10 cards, but it would still just be "on par" with NVIDIA's current high-end Turing graphics cards if this reporting is accurate. When NVIDIA's next-generation graphics architecture finally hits the market (likely some time in 2020), we could see repositioning in the GPU wars shifting yet again."

 
Marees said:
I think you meant 144/"84"
Click to expand...

Which makes the 2080ti like 72% faster. Nvidia has a better memory controller and bandwidth might be a bottleneck for AMD even with the 384 bit bus. Still the 2080ti is only like 50% faster than a 5700xt, so it would be enough to at least match the 2080ti.

The best idea of performance would be a doubling of the 5600xt when it comes out, which is 192 bit.
 
erek said:
"With that being said, Navi 21 should offer a significant performance uplift over existing Navi 10 cards, but it would still just be "on par" with NVIDIA's current high-end Turing graphics cards if this reporting is accurate. When NVIDIA's next-generation graphics architecture finally hits the market (likely some time in 2020), we could see repositioning in the GPU wars shifting yet again."

Click to expand...


Just someone else regurgitation the same rumor for clicks...
 
Snowdog said:
Erek, could you avoid posting low quality sources like WCCFT?

If you want a good quality source of leaks, and industry info, it's Videocardz.com.

There will be a lot less stories, but they will be of higher quality and veracity.
Click to expand...
Or better yet you could step in and post some threads and show how it's done, so erek isn't doing ALL the work. You do know he's only a volunteer here trying to liven the place up, yes? :)

WCCF garbage makes me roll my eyes too but I'm not going to criticize what I'm too lazy to do better myself.

(I'm not attacking you, btw)
 
Even rumors coming down the line at least enlighten us to possible advances. One can hope, yes?
 
erek is doing a damn good job.


Is it bad that I'm still kinda waiting for infinity fabric and chiplets to make it's way to AMD GPUs? I mean I really need an upgrade in 2 of my machines, been waiting for the price of the 5700 xt to drop a bit.... but damn if I don't want some tasty tasty chiplets.
 
Snowdog said:
If you want unfiltered rumors, just read WCCFT front page directly. I would like to see actual tech news, or leaks/rumors from a site with a good track record like Videocardz.
Click to expand...
Even rumors have an inkling of truth to them. Whether it eventually pans out, well, we wait and see
 
erek said:
i wasn't the only one to fall for this @




pendragon1 ^^^
Click to expand...

God... I miss the old [H]... I know this is now just a tech forum with some hardcore people on it but if that were true we wouldn't be posting bullshit from unverified sources on here. I think what others have been saying is that not all news is good news. I doesn't matter who falls for what bullshit leak. It matters that we are actually discussing and speculating on real shit here.

That's what keeps me coming back. I still dig through the forums for people who have posted their commentary on issues with hardware and software. I think it shouldn't be all that damn hard for us to find stories that are a bit more solid than "5950 XT". Honestly, why the hell would AMD jump from 5700 XT on 7nm to 5950 XT on 7+ nm and not flesh out the product stack? They worked backwards from the 5700, to be certain however, I find it more likely that they release a 5800 XT before going for something that sounds like it's a pair of 5700/5800 XT chips in Crossfire on a single board (at least that's what the 5950 XT seemingly invokes in my brain as any higher and it's a 6000 series board).

Kyle & crew may not be leading the charge for us anymore. Perhaps, it's time we did our own investigative journalism before shit posting / posting shit here. ;)
 
Azrak said:
Feel free to post them.
Click to expand...

Erek will shortly anyways. :p

Anyways, I'm sure AMD will eventually launch a 2080ti killer next year.. I mean they finally gave us the a 1080ti killer (barely) last year with the Radeon VII, hah.
 
  • Like
Reactions: Auer
like this
Azrak said:
Feel free to post them.
Click to expand...

I used to regularly post what I thought where good quality info, from Videocardz, Anandtech and ArsTechnica, and minor stuff from other sources when the forum wasn't being flooded with the rumors of the day.

Now I have given up searching out quality things to post when it is lost in a flood of low quality rumors.
 
Snowdog said:
I used to regularly post what I thought where good quality info, from Videocardz, Anandtech and ArsTechnica, and minor stuff from other sources when the forum wasn't being flooded with the rumors of the day.

Now I have given up searching out quality things to post when it is lost in a flood of low quality rumors.
Click to expand...

I was about to post something to this effect earlier, but figured people would just blindly defend OP anyways. Some here seem to think that posting anything is better than posting nothing, but that's not the case when it deters people from visiting the news forum just to sort through the spam. Erek filtering his posts doesn't necessarily have to mean posting less either and the "then you post something better" argument doesn't apply just for the reason you stated.

Regardless of how you feel, when you see half the threads devolve into his posting habits, it's clearly worth addressing in some form.
 
Legendary Gamer said:
God... I miss the old [H]... I know this is now just a tech forum with some hardcore people on it but if that were true we wouldn't be posting bullshit from unverified sources on here. I think what others have been saying is that not all news is good news. I doesn't matter who falls for what bullshit leak. It matters that we are actually discussing and speculating on real shit here.

That's what keeps me coming back. I still dig through the forums for people who have posted their commentary on issues with hardware and software. I think it shouldn't be all that damn hard for us to find stories that are a bit more solid than "5950 XT". Honestly, why the hell would AMD jump from 5700 XT on 7nm to 5950 XT on 7+ nm and not flesh out the product stack? They worked backwards from the 5700, to be certain however, I find it more likely that they release a 5800 XT before going for something that sounds like it's a pair of 5700/5800 XT chips in Crossfire on a single board (at least that's what the 5950 XT seemingly invokes in my brain as any higher and it's a 6000 series board).

Kyle & crew may not be leading the charge for us anymore. Perhaps, it's time we did our own investigative journalism before shit posting / posting shit here. ;)
Click to expand...


Investigation here we go

Typical events "leaks" before amd releases anything:

1. Linux drivers adding architecture names, or hex codes identifying the GPU.
only GFX908 was spotted. (chip_arcurus) in last 3 months. (101x ones weren't updated in last 7 months - guessing navi)
D_qgN63UwAAuUOa.png

No other new hardware id was spotted in drivers.

ROCm (amd's machine learning) HIP (cuda converter to c++) added support for gfx908 around 2 month ago. (commit #685)

upload_2020-1-3_9-42-52.png

The committer (saadrahim) owns Tensile libs
https://github.com/saadrahim/Tensile/tree/gfx908

https://github.com/ROCmSoftwarePlatform/Tensile/wiki
Thus the presentation could be showing off some of its capabilities within this project. Noting there hasn't been any updates since.


We can see the fluidity of the GPU (the gpu being changed?) target needs feature-set suffixed
upload_2020-1-3_9-49-31.png

changes to call name, removing older name and adding SRAM-ECC?
Why amdgcn, and amdhsa doesn't tell us all that much? Navi cards were also described in same way, might be mistake in navi kernel descriptors or devs being lazy.

sram-ecc is interesting, but i personally have no clue on the actual use of sram - it might be something quite custom in fact or just similar to naming | or just another test board by amd.

From llvm.amdgcn AMDGPU drives we can see something beyond GFX908

; RUN: llc -march=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GCN,GFX906 -- Vega20
; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GCN,GFX908 -- Arcurus
; RUN: llc -march=amdgcn -mcpu=gfx1011 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GCN,GFX10,GFX1011 -- Navi 11
; RUN: llc -march=amdgcn -mcpu=gfx1012 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GCN,GFX10,GFX1011 -- Navi 12


some other interesting snips
https://fossies.org/linux/misc/cfe-9.0.0.src.tar.xz/cfe-9.0.0.src/lib/Basic/Cuda.cpp

upload_2020-1-3_10-16-0.png


We know that
gfx906 = Vega20

gfx907 = Instinct full vega20 (32G, 64CU's + HBCC)

gfx908 = Mi100 (apple's 2x vega20?)
GFX1032 = Van Gogh?

upload_2020-1-3_10-54-41.png


2. Leaked CompuBench results
No visible results at this moment, but there has been some within last 3 month from same users that uploaded other "leaked" cards developer model cards (due to changes in rocm for 908 i assume its the same one.).

Thus by all standards there is no big navi

AMD will likely by showing apple's dual vega or gfx908 (which i assume is a 32G HBM2e, 3rd gen vega card.)
 

Attachments

  • upload_2020-1-3_10-0-3.png
    upload_2020-1-3_10-0-3.png
    15.9 KB · Views: 0
  • upload_2020-1-3_10-14-7.png
    upload_2020-1-3_10-14-7.png
    16.7 KB · Views: 0
  • upload_2020-1-3_10-28-27.png
    upload_2020-1-3_10-28-27.png
    11.5 KB · Views: 0
I'll believe it when I see it. Somehow, I have this feeling that the best AMD will be able to achieve with Navi anytime soon is getting within 10-15% of the RTX 2080 Ti. Even if rough parity is achieved, or even if Navi performs slightly better, its late in the RTX 2080 Ti's life cycle. Meaning, AMD is still way behind.
 
EVEN if AMD releases a 2080ti killer, Don't you think Nvidia will respond? Think about it, Nvidia still needs to release their new 7nm products this year. I mean lets be honest they aren't about to release a new product line with the same performance as Turing.

Now dont get me wrong, having more options in the GPU market is really fucking badass. But, AMD will still be playing catch up EVEN if they release a 2080ti killer.

My 0.02c
 
Dan_D said:
I'll believe it when I see it. Somehow, I have this feeling that the best AMD will be able to achieve with Navi anytime soon is getting within 10-15% of the RTX 2080 Ti. Even if rough parity is achieved, or even if Navi performs slightly better, its late in the RTX 2080 Ti's life cycle. Meaning, AMD is still way behind.
Click to expand...

One advantage to that is we may be able to get a Radeon with close to 2080Ti performance for a few hundred dollars or so after the launch dust settles.
 
Brackle said:
EVEN if AMD releases a 2080ti killer, Don't you think Nvidia will respond? Think about it, Nvidia still needs to release their new 7nm products this year. I mean lets be honest they aren't about to release a new product line with the same performance as Turing.

Now dont get me wrong, having more options in the GPU market is really fucking badass. But, AMD will still be playing catch up EVEN if they release a 2080ti killer.

My 0.02c
Click to expand...
Yeah... We haven't seen the 2080Ti Super Duper Edition yet... :rolleyes:
 
Legendary Gamer said:
Yeah... We haven't seen the 2080Ti Super Duper Edition yet... :rolleyes:
Click to expand...

I don't think we will either. Nvidia plans to launch their own 7nm cards this year. I mean they could release a 2080ti super to rip off more people, but those people probably have money to burn. No need to buy it with new generation of cards coming this year.
 
Dan_D said:
I'll believe it when I see it. Somehow, I have this feeling that the best AMD will be able to achieve with Navi anytime soon is getting within 10-15% of the RTX 2080 Ti. Even if rough parity is achieved, or even if Navi performs slightly better, its late in the RTX 2080 Ti's life cycle. Meaning, AMD is still way behind.
Click to expand...
I think you are probably correct in terms of performance. However, I also feel a lot of people don't care, because a lot of people don't care about the RTX 2080 TI price range. If they can get a competitive product below $1000 that would be something I might consider.
 
erek said:
Click to expand...


well this assumes its just 5700xt with everything 2x. I highly doubt that is the case. RDNA 2 is actually a major architectural change given it will have ray tracing and other improvements.
 
elite.mafia said:
Can we please ban WCCFTECH and tweaktown from being posted here...
Click to expand...

Freedom of speech also means freedom to decide what news is fake or valid for ones self.

Clearly you fail in fundamental human liberty or you just dont care.

The 1st Amendmwnt guarantees the right of free press. To ban a website because you dont like thier stories means you are advocating that [H] and its members support tyranny and the usurpation of the constitutional rights of our members.
 
defaultluser said:
While the RX 5700 XT stock burns way too much power even think about doubling performance, the underclocked/cut RX 5700 is a lot closer to sanity.

View attachment 212467

We could likely power an underclcoked RX 5700 *2 on a 384-bit bus . IF you got perfect scaling, it would match the RTX 2080 Ti:

View attachment 212468

144/74 = the 2080 Ti is 1.95 times faster than the RX 5700 at 4k. So a perfect doubling of performance would be a smidge faster.

By the time you will be able to buy Big Navi, Nvidia will probably have Ampere out. But I can see them announcing it at CES, then releasing it 3-6 months later (They did this with Vega 64 3 years ago).

For additional features/final die size, who the fuck knows?The only thing can say for certain is this will not support both HBM or GDDR6.


Even the RT support is still up in the air, as it could b something as simple as the RX 1660 Ti, or something more involved.
Click to expand...

Again why the hell everyone is assuming RDNA2 is basically same as RDNA 1? Heck its well reported its not it will have more improvements. If they were going to do just double the 5700xt they wouldn't be waiting a whole year to do it.
 
  • Like
Reactions: Mega6
like this
tangoseal said:
Freedom of speech also means freedom to decide what news is fake or valid for ones self.

Clearly you fail in fundamental human liberty or you just dont care.

The 1st Amendmwnt guarantees the right of free press. To ban a website because you dont like thier stories means you are advocating that [H] and its members support tyranny and the usurpation of the constitutional rights of our members.
Click to expand...
Freedom of speech has nothing to do with this situation in the slightest but okay....
 
elite.mafia said:
Freedom of speech has nothing to do with this situation in the slightest but okay....
Click to expand...
What do you call your attempt to censor a Message Board by banning certain sites/postings that you do not like or prefer then?

censor
[ sen-ser ]

noun
an official who examines books, plays, news reports, motion pictures, radio and television programs, letters, cablegrams, etc., for the purpose of suppressing parts deemed objectionable on moral, political, military, or other grounds.
 
serpretetsky said:
I think you are probably correct in terms of performance. However, I also feel a lot of people don't care, because a lot of people don't care about the RTX 2080 TI price range. If they can get a competitive product below $1000 that would be something I might consider.
Click to expand...

Well price is everything. If we get a card that's close to the RTX 2080 Ti at $700, then it could be a winner. Priced like an RTX 2080 Ti is now, the card would be a dismal failure. Too little, too late. Everyone willing to spend that much that hasn't already will wait until NVIDIA's next launch before taking the plunge.
 
Mega6 said:
What do you call your attempt to censor a Message Board by banning certain sites/postings that you do not like or prefer then?

censor
[ sen-ser ]

noun
an official who examines books, plays, news reports, motion pictures, radio and television programs, letters, cablegrams, etc., for the purpose of suppressing parts deemed objectionable on moral, political, military, or other grounds.
Click to expand...

It has nothing to do with the right to free speech... because it's a private forum...
 
Mega6 said:
And who elected you spokesman for the entire Forum?
Click to expand...
Nobody? I see these sort of BS posts extremely frequently and almost all the thread is people saying NOT THIS SITE AGAIN
 
  • Like
Reactions: T4rd
like this
Well, wccf may be the fake news of hardware but I've learned some things from this thread.
 
NKD said:
Again why the hell everyone is assuming RDNA2 is basically same as RDNA 1? Heck its well reported its not it will have more improvements. If they were going to do just double the 5700xt they wouldn't be waiting a whole year to do it.
Click to expand...


Any additional efficiency gains are probably going to pay for the Raytracing hardware. So I think it's safe to assume the power consumption will be the same as if we doubled RX 5700.

I'm not going to assume that it will have the same 30% vampire power as Turing, but something around 10-20% wouldn't be unfounded. WE MAY FIND OUT NEXT WEEK.
 
You must log in or register to reply here.
Back
Top