Question Speculation: RDNA2 + CDNA Architectures thread

Page 174 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,698
6,393
146
All die sizes are within 5mm^2. The poster here has been right on some things in the past afaik, and to his credit was the first to saying 505mm^2 for Navi21, which other people have backed up. Even still though, take the following with a pich of salt.

Navi21 - 505mm^2

Navi22 - 340mm^2

Navi23 - 240mm^2

Source is the following post: https://www.ptt.cc/bbs/PC_Shopping/M.1588075782.A.C1E.html
 

moinmoin

Diamond Member
Jun 1, 2017
4,993
7,763
136
people say Nvidia hasn't been slacking like Intel regarding AMD's competition but Ampere seems like a clear example of exactly that. RTX and DLSS always seemed like heavily marketed by-products of Nvidia's data-center engineering.
I wouldn't say Nvidia has been slacking, but the hardware improvements mainly helped areas outside of gaming. ~2x RT performance? Yeah, but not much of that left in games. 2x FP32 performance? Nice, but essentially useless in games. Big improvements in FP16 DL and int8 DL! Nice, maybe somebody will write an autoplay AI for games to make better use of that? Aside of the software part (of which half was work of others with new Nvidia name tags on it) the RTX 3000 presentation was an audience mismatch.

It's also a bit questionable to have results with SAM enabled since most people will not be using these with 5000 series ryzen
As @Veradun mentioned it seems to be odd to assume people are willing to spend $500+ on a gaming card but then not go the long way and get a fitting CPU as well. We'll need to wait for the Ryzen 5000 benchmarks, but to me SAM enabled seems to be only a problem if Ryzen 5000 turns out not to be the best CPU for gaming.
 
Reactions: Tlh97 and Veradun

Hitman928

Diamond Member
Apr 15, 2012
5,562
8,696
136
As @Veradun mentioned it seems to be odd to assume people are willing to spend $500+ on a gaming card but then not go the long way and get a fitting CPU as well. We'll need to wait for the Ryzen 5000 benchmarks, but to me SAM enabled seems to be only a problem if Ryzen 5000 turns out not to be the best CPU for gaming.

There will be plenty of people who have a 9th or 10th gen Intel CPU that may not want to completely switch out their system for whatever benefits SAM gives. Same with Zen2 owners on 300 or 400 series boards. Having to spend hundreds of dollars for a 10% increase (maybe) and having to tear down and rebuild the whole system doesn't sound very appealing to me.
 

MrTeal

Diamond Member
Dec 7, 2003
3,584
1,743
136
There will be plenty of people who have a 9th or 10th gen Intel CPU that may not want to completely switch out their system for whatever benefits SAM gives. Same with Zen2 owners on 300 or 400 series boards. Having to spend hundreds of dollars for a 10% increase (maybe) and having to tear down and rebuild the whole system doesn't sound very appealing to me.
If they're already willing to pay an extra $800 for 10% going from a 3080 to a 3090, the platform shouldn't be a huge hinderance. Though I agree, someone on a 9900KS or 10900K might not pull the trigger or might go with a 3090 in that case.
 

moinmoin

Diamond Member
Jun 1, 2017
4,993
7,763
136
There will be plenty of people who have a 9th or 10th gen Intel CPU that may not want to completely switch out their system for whatever benefits SAM gives. Same with Zen2 owners on 300 or 400 series boards. Having to spend hundreds of dollars for a 10% increase (maybe) and having to tear down and rebuild the whole system doesn't sound very appealing to me.
If so they apparently plan to buy a PCIe 4 GPU (that's true for both GPU manufactures) for a PCIe 3 system. Either they knowingly and willingly decide to cut such corners, or they didn't plan well at all. ¯\_(ツ)_/¯
 

Hitman928

Diamond Member
Apr 15, 2012
5,562
8,696
136
If they're already willing to pay an extra $800 for 10% going from a 3080 to a 3090, the platform shouldn't be a huge hinderance. Though I agree, someone on a 9900KS or 10900K might not pull the trigger or might go with a 3090 in that case.

The people who spend on the 3090/6900XT, yeah they will probably build a new system to have the best because that is what the highest end tier is for. But that will be a very small amount of next gen GPU purchasers.

The SAM feature works in GPUs from $580 to $1000 and will extend below that mark soon. Even if we take the $650 card, you are talking about someone upgrading a very high performance CPU for an additional $600 or so minimum without downgrading on CPU cores. You are doubling the price of the card to get a minor bump in speed. You might as well just buy the 6900XT at that point, it will be cheaper and just as fast if not faster.

If so they apparently plan to buy a PCIe 4 GPU (that's true for both GPU manufactures) for a PCIe 3 system. Either they knowingly and willingly decide to cut such corners, or they didn't plan well at all. ¯\_(ツ)_/¯

PCIe4 versus PCIe3 makes minimal if any difference in almost all games, even with the highest end cards. Again, plenty of people with 9th and 10th gen Intel systems will buy PCIe4 GPUs because the cost of replacing the full system is insane compared to the insignificant speed bump to get PCIe4 support.
 

Hail The Brain Slug

Diamond Member
Oct 10, 2005
3,216
1,615
136
If so they apparently plan to buy a PCIe 4 GPU (that's true for both GPU manufactures) for a PCIe 3 system. Either they knowingly and willingly decide to cut such corners, or they didn't plan well at all. ¯\_(ツ)_/¯

There's third party testing showing PCI-e 3.0 vs. 4.0 on Ampere cards is a worthless distinction.

Always has been people will upgrade the GPU in their existing system because it's always been the most important upgrade for improved gaming performance. Why do you seem to think this is poor planning or cutting corners?

Plenty of people with Zen/Zen2/Skylake/Coffeelake or even older systems that have plenty of CPU power for gaming and don't need to upgrade.

 
Reactions: psolord

pj-

Senior member
May 5, 2015
481
249
116
If they're already willing to pay an extra $800 for 10% going from a 3080 to a 3090, the platform shouldn't be a huge hinderance. Though I agree, someone on a 9900KS or 10900K might not pull the trigger or might go with a 3090 in that case.

Well I am the exact type of idiot who these products are targeted at and going with 5950x + 6900xt makes sense to me because I currently have a 3950x and a x570 board so nothing else needs to change. If I was on a 10900k I don't think I'd go through the hassle of switching to ryzen for SAM unless the CPU gaming performance gap was really significant, which I doubt it will be.
 
Reactions: lightmanek

moinmoin

Diamond Member
Jun 1, 2017
4,993
7,763
136
Let's wait for benchmarks, then people can make informed decisions about all the possible compromises they can afford to make with their partly cutting edge gaming systems. Does that mean that the additional performance possible with SAM (that likely makes good use of PCIe 4) shouldn't be tested? I don't think so.
 
Reactions: spursindonesia

Hitman928

Diamond Member
Apr 15, 2012
5,562
8,696
136
Let's wait for benchmarks, then people can make informed decisions about all the possible compromises they can afford to make with their partly cutting edge gaming systems. Does that mean that the additional performance possible with SAM (that likely makes good use of PCIe 4) shouldn't be tested? I don't think so.

I don't think anyone is saying don't test it, just that it shouldn't be presented as the default our stock configuration.
 

moinmoin

Diamond Member
Jun 1, 2017
4,993
7,763
136
I don't think anyone is saying don't test it, just that it shouldn't be presented as the default our stock configuration.
Why not? In the same way you could as well argue that all GPU tests should be done using only PCIe 3 since that's what most PC users still have. Do you want to know the best possible performance by default (as well as its system configuration) and go from there, or do you want some preset compromises that may well not match the compromises you would personally be willing to make?
 

kurosaki

Senior member
Feb 7, 2019
258
250
86
I don't see why this has to be the case. Compare N21 with Cypress (old 5000 series) which had XT, PRO, and LE that were the 5870, 5850, and 5830 respectively.

If you look at those cards compared to N21 the similarities are eerie. Cypress had 3 bins with 100/90/70 percent of hardware enabled. Navi21 is 100/90/75. The price spread for Cypress was 100/68/63 percent relative to full die cost. N21 is 100/65/58 which is similar.

AMD hasn't a die with 4 bins (I'm excluding refreshes where the same die on a much mature process was given a new designation like Tahiti Pro2) since RV770, which is a decade ago. It may be possible that no 4th bin exists at all. The TSMC 7nm is reported as being quite mature and having a low defect rate relative to other processes at this point in its lifecycle, so an even more cutdown die feels a little less likely.

Maybe we eventually do see something that's a lowest common denominator salvage die, but it would have to be something with reduced memory controllers or infinity cache or a disabled shader engine because it's hard to imagine too many of those existing and at some point the bin is bad enough that even a high clock N22 will come close to it. If the yields are good enough maybe that part is so rare that AMD just scraps them or saves them for some special OEM-only part.
But, do you remember what the cost for the 5000-series was? My memory fails me, i bet though, it was not 579usd for the 5850..
 

kurosaki

Senior member
Feb 7, 2019
258
250
86
PCIe4 versus PCIe3 makes minimal if any difference in almost all games, even with the highest end cards. Again, plenty of people with 9th and 10th gen Intel systems will buy PCIe4 GPUs because the cost of replacing the full system is insane compared to the insignificant speed bump to get PCIe4 support.
With SAM, we get 64GB/s fed to the CPU, instead of 25-ish for the DDR4-RAM. Thats quite a difference. Look ahead and PCIex gen 5 will deliver 128GB/s to the CPU. Might as well throw those sticks of RAM out the window, just buy a AMD 8900XT with 48 Gigs of GDDR7 feeding the CPU with the SAM-Ex-ALLCPU feature.. ;D
 

MrTeal

Diamond Member
Dec 7, 2003
3,584
1,743
136
With SAM, we get 64GB/s fed to the CPU, instead of 25-ish for the DDR4-RAM. Thats quite a difference. Look ahead and PCIex gen 5 will deliver 128GB/s to the CPU. Might as well throw those sticks of RAM out the window, just buy a AMD 8900XT with 48 Gigs of GDDR7 feeding the CPU with the SAM-Ex-ALLCPU feature.. ;D
You'd only get 63GB/s if you saturated both the up and downstream links with PCIe 4.0 x16, no? DDR4 3200 would already be 25.6GB/s for single channel one way and twice that in dual channel.
 
Reactions: kurosaki

Ajay

Lifer
Jan 8, 2001
16,049
8,089
136
There is one reason for it.

Layouts of the dies.

N22 may be almost 50% of N21. But N23 is not. Ergo. We have to wait. N23 is the biggest mystery of them all.
Shouldn't N22 be larger than 1/2 of N21 - given that it has 3/4 of IC and memory controllers?
 

Hitman928

Diamond Member
Apr 15, 2012
5,562
8,696
136
Why not? In the same way you could as well argue that all GPU tests should be done using only PCIe 3 since that's what most PC users still have. Do you want to know the best possible performance by default (as well as its system configuration) and go from there, or do you want some preset compromises that may well not match the compromises you would personally be willing to make?

Many reviewers still do test on PCIe3 platforms. If PCIe3 and PCIe4 had an appreciable difference in performance, there might be a point of discussion there, but there's not. SAM should not be on by default because a very large percentage, probably the vast majority, of gamers will not have access to it so if you turn it on and represent it as stock performance, it is not an accurate representation of what most buyers will get. This is a unique situation we haven't ever run into before that I can remember, but in general you should always give whatever the baseline experience is as the stock representation then turn on whatever features may be available to boost performance on top of the baseline as an extra data set. I actually think this is even more important when your system has to be vendor and generation specific.
 

Glo.

Diamond Member
Apr 25, 2015
5,759
4,666
136
Really looking forward to N22 GPUs, should hit my price and performance goals just fine. I play 1440p, 60Hz.
I would be considering playing on 1440p monitor if... there would be one with 144/165 Hz and 23.8 or 24.5 inch.

27 inch monitors are too large for me.
 

Mopetar

Diamond Member
Jan 31, 2011
7,993
6,410
136
The 5830 also didn't launch until 5 months after the 5870 and 4 months after the 5770 to plug a hole in the lineup, so by that point they would have had a good supply of bad dies on the relatively new and notoriously picky 40nm node.
I think the fact that AMD is leading with a fully enabled 80CU die is pretty indicative that yields are at the least reasonable.

That's a fair point and does show that depending on the process and the die it can take a while to get enough salvage parts to turn into a product category. It's just a weird situation no matter how you split it.First, having 4 bins isn't something AMD does very often and when we do see more than 3, it's usually a case of one or more just being the same cut of chip but with better clock speeds due to process improvements. The issue is that because the process is good we don't know if that means it will just take a long time to get a really low end salvage part, or that there's just so few dies that defective that there's no reason to make that kind of part.

The more I look at Evergreen, the more weird coincidences crop up. Juniper (the N22 analog) was half the size of Cypress, which is the same as N21 to N22. Also, all of the Cypress parts had the full 256-bus, so if history is any indicator (I don't really trust it since it's so far removed) then AMD might not bin the parts based on creating a chip with a 192-bit memory bus. If we go back to RV770, they did release a card with a 192-bit memory bus, but it took almost a full year after the initial launch for the RV770 CE (branded as the Radeon 4810) to come out, and I honestly don't even remember that card. It was probably a pretty low volume part.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |