Speculation: GPU Performance Comparisons of 2020 *Spawn*
- Thread starter eastmen
- Start date
- Status
Honestly I don't think AMD can ever dethrone Intel, at least in volume terms (AFAIK Intel's production is 800K wafers per month and AMD is sub 100k total). Their stated goal is to reach 20% market share in servers and I can see them reaching that by end 2021 with Milan ramping (They've recently reached 10% from near zero pre Zen). And then Genoa in 2022 should help increase that even further. But to what level? Forget beating Intel, to even reach 50% is a herculean task and will take till at least say 2025, IF AMD executes flawlessly and Intel keeps screwing up. That said, I can see AMD maintaining their CPU lead till Intel 7nm in 2023, though AMD will have access to TSMC 3nm by then so it could still be in AMD's favour.
Point being, and as stated by others, the current prioritization of wafers towards the server market where the ASPs are the in thousands of dollars (and their margins huge thanks to the chiplet strategy) makes a lot of sense over the dGPU market where they have been weak in recent times. If well received and with access to the additional capacity vacated by Huawei, they can ramp production if necessary.
Great explanation. I have similar thoughts.
Not many would remember that the origins of Conroe and Intel's future success story for the last 15 years came from Dothan.
Why would anything be hit if RT isn't being used? RT being shared with TMU's wouldn't hamper their efficiency I'd think. Just that you couldn't use them concurrently.
Bandwidth is definitely a large reason? The 2080Ti has 50% more, and correspondingly more ROPs.
I'm not following your logic here. If 2080Ti is 50% faster than a 5700XT, then simple math says double of 5700XT is 33% faster than 2080 Ti (i.e. 200% vs 150%). Definitely more than just "barely faster"
Bondrewd
Veteran
Not the idea.
Ideal x86 market is 50/50 swinging into 60/40 per specific product cadence.
AMD roadmap is ultra-agressive and one should never ever write them off even assuming Intel 7nm rollout happens flawlessly in '23.
It chugs more than comparable class nV GPUs.
Why?
No idea.
neckthrough
Regular
What's the point? Once all the numbers are available it ain't fun no more. The fun is all in armchair extrapolation from scraps of news. Carry on folks!!!
Tarkin1977
Newcomer
2080Ti is more than 50% faster than 5700XT @4K. Doubling 5700XT will barely get you above 2080Ti @4K.
excuse me, but... wouldn't a doubling of 5700XT performance be 200% ??? - thats about 30% faster than a 2080TI - or am I a complete idiot?
Bondrewd
Veteran
In my humble experience in dealing more so on the console side of things: intentionally vague commentary, added to short one liners and a lot of 'you'll see' has generally been a road map to their demise ultimately.
We are all capable of looking at trend data and making fairly reasonable assumptions of where things are headed.
The true hallmark of an insider isn't predicting that trend, it's predicting the outliers on that trend, because those are the things that no one expects. It's about knowing the data points that no one has information about.
It's okay to be a fanboy or whatever, I don't mind optimism, and I don't have a problem with people asking others to stay as objective as possible, but I generally prefer that if someone doesn't know something, it's okay to say you don't know. I don't know a lot of stuff. And that's okay.
And even if you do know something, ti's okay to even say you don't know if it's still coming, because plans change and is all too common in this industry.
I frankly find his posts confrontational and unnecessarily forceful - and that's usually where I feel like I'm being taken for a ride if I trust this person.
Bondrewd
Veteran
Trend data says AMD has no chance at mobile yet they're smashing it.
Just write everything off and patiently watch; it's very fun that way.
Well, I guess if I have to guess, I'm going to say navi21, or whatever the hell it's called is going to be about 285W and roughly double the performance of a 5700xt, which should put it in 3080 territory. We'll have to see where 3080 benchmarks vs the 2080ti. I'm going to guess that frequency is not going to see a big boost from the 5700xt because it's an 80CU part. Whatever the smaller RDNA2 parts are called will probably see higher clocks pushing game clocks into 1850MHz or 1950MHz, but game clocks for this 80C part will stay reasonably close to 1755 MHz. My gut still tells me that Nvidia will be well ahead on ray tracing performance because they have many dedicated (2nd gen) units vs shared functionality with TMUs.
I'll be frank, I have tons of money invested in AMD, and it's absolutely in my favour for them to perform amazing and to watch the ticker go up. And they have absolutely been doing that since I bought them at their low early 2019.
But I can't help the fact that when I read your comments, none of it makes me want to invest in AMD more. You talk about watching and trends and this and that. But you haven't been able to write any sort of coherent post that takes 1 idea or concept and lead it to the next; not a single person can follow you, but you've been highly effective at stopping all sorts of discussion from proceeding under the guise of knowing things.
I get that people don't like concern trolling, and that probably happens a lot, and I'm sure I've done my fair share so I'm certainly not innocent from doing it. But if you're not going to let that discourse occur, there's no technical discussion here and this forum may as well not exist. We aren't a marketing arm for any single company, so having conversations loaded with heavy pessimism as a counter to over enthusiasm is how we improve the probability around the accuracy of our assertions.
Don't worry, I will sit back and watch.
Bondrewd
Veteran
Great, it's too late anyway.
There is nothing to discuss because nothing concrete is known.
Okay, AquariusZi on ppt said N21 is great but they're not allocating much into it.
Cool enough?
There's nothing to discuss, all the flavours of napkin math have been done three times over since the FAD announcement.
Unless we of course want to wank nVidia, but there's plenty of that shit all over the internet now.
Nice; not much time left to wait anymore.
I want AMD to compete or even best NV, its in a pc gamers intrest. Its almost certain amd will offer in the same performance range as nv (20 to 30TF), and thats all they need to do for now i think. A 3090 variant as their highest end will most likely also come, but has a smaller market.
DegustatoR
Legend
As I've said already this and next year will be fun. Probably not in a way which some posters here are expecting but fun nevertheless.
Says the man who is pointing to XSX APU at all times.
Bondrewd
Veteran
Hell yeah gimme those SoIC GPUs.
Yeah it's napkin math that has been done ten times over and no one is interested in it because XSX announcement was ages ago.
This thread is just raw shitlording from both sides, and while fun it is fundamentally pointless.
But I like simple fun.
DavidGraham
Veteran
To reiterate, NVIDIA is calling RDNA2 a minimalistic RT approach:
https://www.techpowerup.com/review/...ture-board-design-gaming-tech-software/4.html
Yes I'm expecting Rocket Lake to be a turning point for Intel. Sandybridge was awesome in it's day but they've stuck with it for too long. I love Zen 2 (I have one) and I'm expecting Zen 3 and 4 to deliver the goods too, but in gaming at least, core for core, Comet Lake is still a decent amount faster. It's just that no-one cares at this point because they're stuck on PCIe 3.0, they're more expensive, and they have fewer cores in the top end halo products. I'm expecting Rocket Lake to bring a significant enough IPC uplift to make the gap between it and Zen 3 much more obvious, while obviously bringing Intel up to the times on the PCIe front. I'm not really expecting them to progress on cores counts though. That may be AMD's saving grace.
Bondrewd
Veteran
We need to call IMG to continue the conga line.
Oooooh god no.
It has no chances in hell against Zen3.
At all.
0.
You have benchmarks to prove this do you?
Bondrewd
Veteran
Technically yes.
Is SPECint enough?
- Status