Hardware discussion/analysis relating to software playback

[chros]

nVidia officially cancelled the launch of 3070 on 15th of October and pushed it back, just one day after the D-Day, the announcement of Big Navi.

The Earth has already started shaking for the upcoming release...
Hmm, "interesting", I wonder why this happened.

Jim talks about performance/watt/price of Ampere (https://www.youtube.com/watch?v=VjOnWT9U96g) by comparing top dies, he's not impressed :) Conclusion?
- performance: Ampere *and* Turing together achieved the same thing as Pascal did (last time)
- price: crap/worse than Turing was
- watt: maybe this generation should be called Watt instead of Ampere! :D

This gen has "only" 2 advantages: HDMI 2.1 and AV1 decoder ...

[NikosD]

This gen has "only" 2 advantages: HDMI 2.1 and AV1 decoder ... Both of them supported by AMD RX 6000 series and AV1 HW decoder is supported by Intel's Xe too.
Probably Rocket Lake will support in its Xe HDMI v2.1 too.
But AV1 HW decoder is hardware accelerated playback, not software.

[Alexkral]

This gen has "only" 2 advantages: HDMI 2.1 and AV1 decoder ...

And PCIe 4.0, I wonder if it can be useful for AviSynth AiUpscale and AviSynth Shader in general as the bottleneck is memory transfers from the GPU to the CPU.

[huhn]

good point!

@chros plz that guy is an insult on journalism.

comparing the generation performance by using 1 card that as an example. i don't even know anymore... is even not even aware of the low number of 2080 TI out there?
does he even realise that AMD did 100 % the same thing with navi.
RX 480 200-240
RX 5700 XT 400+ 96-100% performance increase for 100 % more money.

there is a reason the 2000 series didn't sell well that not competition from AMD by doing the same.

[SamuriHL]

Because the price was ridiculous. LOL I still have no regrets about my 2070 or this 2080 laptop. However, I have irritation about this "launch" of 3080 series and the inability to find one. I'm only looking at the ASUS lineup at this point. Even if that means it'll be next year before I get one. Oh well.

[NikosD]

...by comparing top dies, he's not impressed :) There was a reason that the marketing guy running nVidia, also known as leather-jacket-man, referred to 3080 as the flagship card of this generation and not 3090, although he presented them both at the same show from his kitchen.
He wanted to avoid such comparisons (performance per watt, performance per price etc) with 3090.

[huhn]

yes the 2000 series is over priced i claimed the same until AMD did the same then i was disarmed.

and why should you feel bad about buying the cheapest and fastest GPU you could at the time? there was no alternative except no GPU/laptop.

it this journalist even taking into account Vram price hike just blame solo on nvidia for price "increases" what about the loss of a microchip manufactures like global foundries and the rise of SSD and and smart phone needs. every one was going to TSMC and there was no way they ask for more too just like Vram?

and taking into account AMD did the same thing odd isn't it?
because the 2080 ti was 13% more expensive performance/dollar then the 1080 TI doesn't mean other cards where but by doing thsi you can't do a propaganda chart like this charts so you don'T do that.

that's what user care about: https://youtu.be/csSmiaR3RVE?t=1299
and the 3080 delivers here and if AMD will beat it or be in a similar range people will care about the RDNA 2 too. but until tehy are released i will not predict what will happen i don'T care about the shader count yet i don'T care about the memory interface yet.

and even AV1 decode isn't finished yet or do i needed to remind user here about the 970/980 hybrid garbage or the VP9 "decoder" in polaris.

[SamuriHL]

Agreed on all that. I've been quite pleased with my 20x0 cards. But I'm looking forward to the raw speed bump that a 3080 will provide over my 2070 for both gaming and for madvr especially. But I can wait until nVidia decides to get their s*** together. LOL

[el Filou]

and even AV1 decode isn't finished yetDo you mean on the GeForce, Radeon, or both? I've seen one review actually test the AV1 decode with Chrome and 8K YouTube, and Task Manager was showing 50% video decode with hardly any GPU 3D at all, so at least on the NVIDIA it's working.

[huhn]

so am i the only one that is pretty disappointed in zen 3?

i mean the 12 core looks ok and the 16 core too but 450 for an 8 core CPU i can easily get an 10 series i7 for 300, an 9 series i7 for 250 also an zen 2 3700x for 270 that's quite a difference in price i mean they compare then to the 3800x an CPU with a bad price performance ratio and even that CPU is available for 300. the 3900x cost around 400 so ehh well...

i mean they claim quite some performance uplift but i have my doubt they are aware of the current street prices...

this has a kinda a turing feeling with just similar price per performance.

looks like i end up with 2 zen 2 systems...

[Asmodian]

If you are beating the competition you charge more. AMD wants average prices of their CPUs sold to go up and they finally have the performance lead to get away with it.

I was not expecting competitive prices and I was pleasantly surprised by the amount of IPC increase claimed, including the clock speed bump it could be pretty significant. I definitely want to see benchmarks and where most all core OCs top out.

[SamuriHL]

I've been stuck in meetings all day so I haven't had a chance to catch up on Zen 3 news yet, but that's disappointing on price. I was considering a rebuild of my system while I wait for a 3080 to materialize, but not at those kinds of prices.

[butterw2]

a 19% IPC increase over a 16 month period is nothing to scoff at. But if you are not overly concerned with performance (IN GAMES !), zen2 still looks like a good option.

Zen3 is pretty much a drop in replacement for zen2: same process (slightly improved boost freq), same total L3 cache, same IO-die. They are just changing the CCX architecture to 8 cores (was 2*4 cores).

[huhn]

yes production cost clearly didn't increase the price.

edit: the clock speeds they have written down are pointless zen 2 is unable to reach these number without extrem overclocking boost clocks are simply pointless with AMD.

[Manni]

Definitely waiting for Rocket Lake here, especially if it meets its Q1 2021 delivery date (March). My position might be different because I need thunderbolt and a stable bios, but I don’t see anything at this stage that would make me change my plan if Rocket Lake delivers what I expect at a competitive price by April 2021. Nice to see AMD making that kind of progress though, but at that kind of price there is even less incentive for me to switch.

[NikosD]

People buying 3090 with 1500€ for no good reason and others buying Intel all these years with astronomical prices like Core i9 HEDT series which is a joke, are accusing AMD for the price of Zen 3.

So, you want to pay 10% to 20% more performance of 3090 vs 3080 with a price difference of 1500€ vs 700€ or you want to pay 2000$ for the two models of 18core HEDT Intel Core i9 7980XE/9980XE but you don't want to pay for better performance per price the Zen 3 processors.

Because they are AMD products.

Hilarious and very funny way to start your day reading your posts.

[el Filou]

Like many are saying, the biggest competitor to Zen 3 on release, with the prices that were announced, may not be Intel but Zen 2 itself. 5800X might not beat 3700X on performance/price.

It's interesting about HEDT, over here 9980XE can be found for 1109 eur (https://www.amazon.fr/Intel-i9-9980XE-processeur-Bo%C3%AEte-Smart/dp/B07JGCMQY8/), while 3960X is at 1409 eur.
I still think AMD should've released a 16-core Zen 2 Threadripper for people who need more PCIe lanes and/or higher memory bandwidth but not necessarily 24 cores.

[NikosD]

Tom from MILD has a new video and he is like reading my comments.
He thinks Zen 3 is the new Sandybridge for AMD this time aka a fully dominant architecture over the rival.
Zen 3 is going to be faster on everything than Intel.
Also, as he says in the video no HEDT for Intel across all 2021.
Now, regarding RX 6000 benchmarks he also thinks that the Big Navi shown at the presentation will not be the fastest one, but the second.
More here (and regarding availability)
https://youtu.be/SE5eYTIwoT0

[NikosD]

On the other hand a very respectful site like techpowerup.com says that AMD probably cherry-picked games for its presentation and the performance of Top Big Navi will be comparable to 2080 Ti level of performance.
Quote:
We know this is a very coarse and unscientific way to compare AMD's numbers to ours, and AMD has probably cherry-picked games that are most optimized to its GPUs, but it lends plausibility to a theory that Big Navi may end up comparable to the RTX 2080 Ti, and trade blows with the upcoming RTX 3070, which NVIDIA claims outperforms the RTX 2080 Ti.
https://www.techpowerup.com/273150/amd-big-navi-performance-claims-compared-to-tpus-own-benchmark-numbers-of-comparable-gpus

The nVidiot tech journalist responsible for the above article is called "btarunr"

[videoh]

The nVidiot tech journalist responsible for the above article is called "btarunr" Why do you need to toss around insults? Your technical arguments are not persuasive enough? Or you are just a jerk?

[NikosD]

Why do you need to toss around insults? Your technical arguments are not persuasive enough? Or you are just a jerk? I don't need to insult anyone.

I like doing so, especially for Intel and nVidia shills like "professional" tech journalists as these who are working in trusted and respectful sites to write garbage, most of the time paid articles.

Or simply jerks like you for example.

Are you ?

[Manni]

Why do you need to toss around insults? Your technical arguments are not persuasive enough? Or you are just a jerk?

Don’t feed the troll :)

The only thing he knows is insults and personal attacks.

Sadly the moderators don’t seem to care.

[NikosD]

Don’t feed the troll :)



The only thing he knows is insults and personal attacks.



Sadly the moderators don’t seem to care. Who is trolling and who is attacking others personally ?

You probably mean that I'm the victim of personal attacks.

Or do you mean that videoh is the tech journalist who wrote that article in techpowerup ?

Why are you calling moderators ?

Friends of you and videoh probably ?

[huhn]

Like many are saying, the biggest competitor to Zen 3 on release, with the prices that were announced, may not be Intel but Zen 2 itself. 5800X might not beat 3700X on performance/price.

It's interesting about HEDT, over here 9980XE can be found for 1109 eur (https://www.amazon.fr/Intel-i9-9980XE-processeur-Bo%C3%AEte-Smart/dp/B07JGCMQY8/), while 3960X is at 1409 eur.
I still think AMD should've released a 16-core Zen 2 Threadripper for people who need more PCIe lanes and/or higher memory bandwidth but not necessarily 24 cores.

yes for some odd reasons AMD gave up on the "budget" massive PCIe lanes option the discontinued 1900x was very useful for these cases.

what so ever i currently have a hard time seeing a 5800x ~450 been 50 % faster then a 3800x ~300 and don't forget the 3700x.

the reason zen 2 even got popular for gaming wasn't the fact they are faster it was the fact they where cheaper and the difference between intel was often "meaningless".

is the 9900k faster in gaming then a 3700x? yes without a doubt but it's just massively better to put the ~150 difference in a better GPU with the slightly less powerful gaming CPU.

if the 5600x is beating or even to the 3700x in everything i may still get one but else the values is questionable. i'm not going to buy a 450 euro CPU the money is better spend on other parts.

[QBhd]

...

edit: the clock speeds they have written down are pointless zen 2 is unable to reach these number without extrem overclocking boost clocks are simply pointless with AMD.

Boost clocks work perfectly fine with AMD if you know what you are doing. My 3700X (which is far from a Golden sample) will hit 4400 on single core and usually boosts up to 4100 all core... If I crank my AIO fans to max I can get that all core boost up to 4200.

As for the prices, I am more than happy to let AMD tack on $50 per skew if the performance claims are true.

QB

[SamuriHL]

See I struggle cause CPU performance doesn't mean nearly as much to me anymore as GPU does. Yes, you need a decent CPU, I'm not suggesting otherwise, but, decent...doesn't have to be melt your face off performance. What entices me to the AMD camp is not just the raw performance edge over Intel, which is pretty vast at this point, but the fact that they've had a leg up on PCIe 4.0. We're getting into what, 2nd, 3rd gen boards now with PCIe 4.0 on AMD? Intel's starting to maybe think about possibly ok we just might for real this time release their PCIe 4.0 stuff in spring of 2021. Yeaaaaaaa. I really can't say that for me with a 9900k and 32 gigs of pc3200 ddr4 that it's really TIME to upgrade. PCIe 4.0 would be nice, though, for more than just a 3080.

[huhn]

holding 4.4 for a short time is just "pointless" measuring tools are not accurate enough to show the real deal they do this only for a very very short time so they can say it's not "wrong" and there is the issue with precision boost... nearly as much of a mass as MCE on intel. AMD is so advanced with there boost clock they switch in such an fast interval that thee shown number is not represent of the true clock.
pre abba: https://www.youtube.com/watch?v=DgSoZAdk_E8

because AMD say it is 50 doesn't mean it is 50 a 3700x cost 270 and a 3800x 300 the 5800x is planned for 450 that not a delta of 50! a 5800x for 350 and the shown claims no question count me in.

edit: AMD has only one mainstream mainboard series with PCIe 4 that the 500 series and with only two chipsets with PCIe 4. don't care enough just to fact check if the 550 can do PCIe 4 at all but as far as i know it's no the PCIe 4 is coming directly from the CPU.

[Asmodian]

B550 motherboards do support PCIe 4.0 lanes from the CPU but the lanes to and from the chipset are 3.0.

Boost clocks work perfectly fine with AMD if you know what you are doing. My 3700X (which is far from a Golden sample) will hit 4400 on single core and usually boosts up to 4100 all core... If I crank my AIO fans to max I can get that all core boost up to 4200.

What control do you have over it? Keeping it cool? You get a +0.25x step in multiplier for every 4-7 degrees C drop in temperature but I am not sure what else "know what you are doing" could mean. Are you maxing out PBO or something? My 3950X under water will "hit" 4.675 Ghz for maybe a few milliseconds (I did see 4.7 Ghz max clock reported once) and it will boost to ~4.15 Ghz under all core load. A manual OC to 4.4/4.3 CCD0/CCD1 is faster for everything but pure single threaded workloads and not much slower even then. What really annoys me is how it boosts with middling loads, as soon as there is more then one thread doing anything I never see any core go above ~4.45 GHz even for a few milliseconds while temps stay very low.

I am not sure what "perfectly fine" means to you but AMD's boost algorithm is still temperamental at best. If you want to look into really freaking it out research the Ryzen PBO EDC bug (https://www.overclock.net/threads/edc-1-pbo-turbo-boost.1741052/) for more information. :p

[huhn]

so great he even he even needed a new nick name.

[SamuriHL]

That's depressing about pcie 4 support. I was under the impression they'd been doing it a while and that it was full support. Guess I will hold off that till it matures a bit more. I've not followed the amd stuff too closely but a couple guys I know have been playing with overclocking their 3800x and seem to be having fun. No idea how stable it really is. Not something I'll be doing. Lol

Sent from my SM-G975U using Tapatalk

[huhn]

why should there be something wrong with the PCIe 4 implementation? the only thing related to issue with PCIe 4.0 are NAVI GPUs where the cure for blackscreens is >everything<.

[nevcairiel]

There is nothing wrong with PCIe4, its just market segmentation. B550 is cheaper because the chipset doesn't have an expensive PCIe 4.0 controller (boards still have PCIe 4.0 for the GPU, straight from the CPU). If you want PCIe 4.0 everywhere, there is X570, and likely going to be plenty board refreshes coming out the next month until the Ryzen 5000 release.

[Manni]

There is nothing wrong with PCIe4, its just market segmentation. B550 is cheaper because the chipset doesn't have an expensive PCIe 4.0 controller (boards still have PCIe 4.0 for the GPU, straight from the CPU). If you want PCIe 4.0 everywhere, there is X570, and likely going to be plenty board refreshes coming out the next month until the Ryzen 5000 release.

Are you aware of any AMD mobo that will have full PCIE 4 and full Thunderbolt 4 implementation? I need a robust TB4 for my specific application (I’m using TB3 currently with many peripherals) and that’s one of the last things that ties me to Intel for now and why I’ll try to wait for Rocket Late if it does show up in Q1. I’d happily try an AMD MB fitting my needs, though I’m a bit concerned with bios stability.

I usually keep a build at least 5-10 years (current one is 8 years old, with a few CPU/GPU/Mem upgrades along the way) so I really build with room to grow. I don’t expect any significant performance improvement with PCI4 to start with, even with the 3090, though there might be a minimal difference.

[nevcairiel]

I'm not aware of a single board, Intel or AMD, with Thunderbolt 4. It usually takes a while for adoption, TB4 was only finalized this year afterall (in July). I wouldn't expect that on boards until at least mid next year.
TB3 AMD boards exist, but they are rare, although may get more common now that other vendors are allowed to produce Thunderbolt silicon instead of having to buy Intels chips.

[SamuriHL]

I want pcie 4 for more than just gpu. I guess I'll see what mobo mfgs do with the zen 3 on their refresh models. Like i said I've not followed amd solutions all that closely but would defintely consider it now. Time to do some homework.

Sent from my SM-G975U using Tapatalk

[huhn]

if you really want TB 4 just do it with addon cards x570 has more then enough bandwidth for that and that the most PCIe lanes on consumer boards you can get don't forget they are over a year old and the just released intel z490 is still PCIe 3...

i'm actually puzzled what's so special about TB 4 the difference to TB 3 is well not that much. there is no USB 4 on the market there is no DP 2.0 card on the market and in "general" it has the same speed as TB 3.

[Manni]

I'm not aware of a single board, Intel or AMD, with Thunderbolt 4. It usually takes a while for adoption, TB4 was only finalized this year afterall (in July). I wouldn't expect that on boards until at least mid next year.
TB3 AMD boards exist, but they are rare, although may get more common now that other vendors are allowed to produce Thunderbolt silicon instead of having to buy Intels chips.

Thanks, I was talking about upcoming AMD boards as you were mentioning them. AFAIK Intel will need Rocket Lake next year for TB4 support. If it means mid-year to do a build, that's a bit of a bummer, but I can wait if needed. I was hoping that new Intel boards with TB4 would be released around RL launch in April.

I want pcie 4 for more than just gpu. I guess I'll see what mobo mfgs do with the zen 3 on their refresh models. Like i said I've not followed amd solutions all that closely but would defintely consider it now. Time to do some homework.

Sent from my SM-G975U using Tapatalk

I agree, I plan to use PCIE4 for my internal SSDs too.

In your case I wouldn't upgrade the CPU yet, with your new 3080 it's unlikely your current CPU will be a bottleneck.

if you really want TB 4 just do it with addon cards x570 has more then enough bandwidth for that and that the most PCIe lanes on consumer boards you can get don't forget they are over a year old and the just released intel z490 is still PCIe 3...

i'm actually puzzled what's so special about TB 4 the difference to TB 3 is well not that much. there is no USB 4 on the market there is no DP 2.0 card on the market and in "general" it has the same speed as TB 3.

I have limited space in my case so I'd rather have it integrated. This usually also helps with performance and stability.

There is no increase in bandwidth with TB4 vs TB3. Just added stability, which is a very important aspect for me.

As I said, when I do a build, I do it for 5-10 years, so I try to get the major techs on board. That means PCIE 4 and TB4 for my needs. As I explained, not massively significant today, but it will be within the next 2-3 years, and I'll still have that MB by then.

[SamuriHL]

I agree, I plan to use PCIE4 for my internal SSDs too.

In your case I wouldn't upgrade the CPU yet, with your new 3080 it's unlikely your current CPU will be a bottleneck.


Yea, that's exactly what I want. As for the GPU, I'm not as concerned about PCIe 4 there. But for NVMe I think it's going to make a nice difference. But I agree, I don't really want to upgrade the CPU either. However, it wouldn't be replaced so much as moved to a new HTPC for my other room with the 3D plasma.

[huhn]

I have limited space in my case so I'd rather have it integrated. This usually also helps with performance and stability.

There is no increase in bandwidth with TB4 vs TB3. Just added stability, which is a very important aspect for me.

As I said, when I do a build, I do it for 5-10 years, so I try to get the major techs on board. That means PCIE 4 and TB4 for my needs. As I explained, not massively significant today, but it will be within the next 2-3 years, and I'll still have that MB by then.

where do you get these stability and performance claims from?
TB 4 is a minor feature update. i mean that's about it: https://www.dpreview.com/files/p/articles/8495582933/USB_insert.jpeg

[chros]

So I think I'm going to cancel next week and wait until we know more about the 3080ti as it might be on TSMC 7nm according to rumours, which might make it a bit less power hungry than the 3090. I don't really believe they'll move the already released card to 7nm, but that's still a slight possibility.
Yep, just cancel it, this is ridiculous.
Do you have a link about the 7nm rumors?

[SamuriHL]

Firmly in the rumor no one has a clue category:

https://www.thefpsreview.com/2020/10/09/nvidia-refreshing-geforce-rtx-30-series-on-tsmcs-7-nm-process/

[chros]

I see, cheers, well, I guess we will see, but even if it will happen, it won't be tomorrow :)

[Manni]

Yep, just cancel it, this is ridiculous.
Do you have a link about the 7nm rumors?

https://www.hardwaretimes.com/nvidias-rtx-30-series-gpus-may-switch-to-tsmcs-7nm-node-in-2021-rumor/

https://www.thefpsreview.com/2020/10/09/nvidia-refreshing-geforce-rtx-30-series-on-tsmcs-7-nm-process/

No idea how trustworthy these rumors are...

[chros]

Thanks. I still don't understand the reason why the +100W difference between 3070 and 3080:
- is it gddr6x?
- and/or awful scaling of hardware?

[el Filou]

why should there be something wrong with the PCIe 4 implementation? the only thing related to issue with PCIe 4.0 are NAVI GPUs where the cure for blackscreens is >everything<.There have been issues on X570 with add-in cards designed for older PCIe revisions. Among those reported that I've read about are sound cards, NICs, and FireWire adapters, but maybe there are others. Some motherboards manufacturers have fixed them with new BIOS that enable a "compatibility mode" that disables newer features of the protocol added in 4.0, one of those being the "10-bit tag" explained here: https://blogs.synopsys.com/vip-central/2016/05/03/full-utilization-of-16-gts-pcie-gen-4-bandwidth/
This has not been reported with GPUs AFAIK, and some people have reported that putting those add-in cards in the slot connected to the CPU does not cause the issues (but then you have to put your GPU on the chipset, so not a fix), so it may be related to the lanes on the X570 chipset. Then again as it's the only chipset currently providing PCIe 4.0 outside of the CPU lanes, there is no way to verify this on another platform (of course the issues could also come from the design of those older adapters not 100% following the protocol spec and causing forward compatibility issues, or "undefined behaviour").
Anyway, yes there have been issues even if it's limited to a very small percentage of hardware and users.
If I had to buy a new system now I would buy a B550 just to avoid that because I'm using an old add-in soundcard and PCIe Wi-Fi.

[QBhd]

I have zero issues with my soundcard... I don't think the problems described above are widespread.

QB

[huhn]

the x1 slots are PCIe 3 not 4 on x570 only the CPU slot and the x4 slot are PCIe 4. i have tested quite some sounds cards with it no problem and the PCIe x1 slots should be identical between b550 and x570.

older PCIe had that problem multiply times (even GPUs) where they stopped working in a new slot and the mainboard bios had to find a workaround. what so ever the topic was a new card in a new spec. if you create a PCIe x4 4 card right now you create it for x570 (for the most part). there shouldn't be any problem.

[Asmodian]

There have been issues on X570 with add-in cards designed for older PCIe revisions.

I am not sure how much I would worry about this, I am sure there are issues with some older cards but most work fine.

My Mellanox MCX354A (ConnectX-3, ca. 2011) and CX455A (ConnectX-4, ca. 2014) work perfectly in my X570 Asus Crosshair VIII Hero with no workarounds. I have all legacy compatibility modes disabled and the slots all in "auto" generation. I don't think these PCIe 3.0 devices are ones Asus did a special BIOS workaround for.

the x1 slots are PCIe 3 not 4 on x570 only the CPU slot and the x4 slot are PCIe 4. i have tested quite some sounds cards with it no problem and the PCIe x1 slots should be identical between b550 and x570.

The PCIe x1 slots support 4.0 on X570. Support for PCIe 4.0 lanes from the chipset is the major difference between X570 and B550. All the PCIe slots on my motherboard support PCIe 4.0.

[huhn]

i just checked again on msi they use PCIe 3.i only checked like 10 boards. the interface between chipset and CPU is PCIe 4 x4.

[Asmodian]

Weird, I just totally assumed it was standard for all X570 boards. :o