Do you have nostalgia for PC building DIY master race? Do you like to twiddle with every single screw? Are you bored to death by all the new fancy automatic overclocking and lack of headroom, taking your freedom away? Do you want to experience again how all the possibilities are opening up to you when nothing works out of the box and you put in your sweat and blood to make stuff just run? Get ARC now, make your hard work count for something again!These cards are not for games, they ARE the game.
#honest infomercialsDo you have nostalgia for PC building DIY master race? Do you like to twiddle with every single screw? Are you bored to death by all the new fancy automatic overclocking and lack of headroom, taking your freedom away? Do you want to experience again how all the possibilities are opening up to you when nothing works out of the box and you put in your sweat and blood to make stuff just run? Get ARC now, make your hard work count for something again!
Linus actually used charts from this PCWorld article conveniently without the CPU lines. SVT always encodes at a higher quality for the same bitrate, so I am really not sure at all their claims of needing a beefy CPU to keep up with a380 are based in reality. Perhaps its impossible to CPU encode at a speed faster than 9, but given the distinct difference in image quality this would suggest the performance difference is from the software implementation and not necessarily the dedicated encoder being 'so much faster'. I'd really like to see some actual CPU load data outside of a quick cut to a seemingly random screenshot of task manager. Openbenchmark data for SVT shows it scaling with core count well, so opting for more cores instead of another GPU seems viable.Can you do comparable AV1 encoding with a CPU? I'm wondering if just throwing more cores at the problem makes more sense, since they're also good at more than one thing and don't have 20watts of idle power burn even when they aren't doing anything.
Linus actually used charts from this PCWorld article conveniently without the CPU lines. SVT always encodes at a higher quality for the same bitrate, so I am really not sure at all their claims of needing a beefy CPU to keep up with a380 are based in reality. Perhaps its impossible to CPU encode at a speed faster than 9, but given the distinct difference in image quality this would suggest the performance difference is from the software implementation and not necessarily the dedicated encoder being 'so much faster'. I'd really like to see some actual CPU load data outside of a quick cut to a seemingly random screenshot of task manager. Openbenchmark data for SVT shows it scaling with core count well, so opting for more cores instead of another GPU seems viable.
View attachment 68979
Yes 32 cores to run a much much higher image quality. Why not turn the quality down more to match the GPU encoder?The PCworld article says he needed 32 core threadripper to encode SVA-AV1 8 or 9 in real time. So unless you have a beast of a CPU the HW encoder will be significantly faster.
Yes 32 cores to run a much much higher image quality. Why not turn the quality down more to match the GPU encoder?
Reminds me of the "Poor Volta ad" time at AMD. Vega was supposed to be a lot faster than it ended as. Same guy, once, poor luck, twice, do I see a pattern?
Brutal.
To be fair, the rx6000 was also Raja.Reminds me of the "Poor Volta ad" time at AMD. Vega was supposed to be a lot faster than it ended as. Same guy, once, poor luck, twice, do I see a pattern?
I was just thinking about that argument and I don't accept that belief anymore.To be fair, the rx6000 was also Raja, so I do not buy that.
Raja created something from nothing, and that is not to be laughed at. The some of the stupid ideas (use iGPU software!) likely came from clueless corperate management. That seems exactly like the typical "cost saving" decision made by upper level management would look like.
I think everyone can agree Raja was a bad voice for AMD, setting unreal expectations left and right. This is 100% on him. When it comes to his actual job though, we don't get to arbitrarily pick his level of involvement from one project to another.If he was responsible for navi2, would he have not tried to include in ARC designs some sort of IF cache or more L2 as Nvidia is doing now. I now don't think he was, except maybe, in laying out very broad & high level design goals. No technical wizardry at all from him.
IF cache is fundamental to Navi 2 and later AMD GPUs. This is not some small design add-on. If he was anywhere close to being deeply involved then it would exist in ARC.I think everyone can agree Raja was a bad voice for AMD, setting unreal expectations left and right. This is 100% on him. When it comes to his actual job though, we don't get to arbitrarily pick his level of involvement from one project to another.
If we want to judge him for detailed architectural fails at Intel, we cannot deny him the same level of contribution at AMD. For that we would need insider info, which we lack so far.
Very much agree. Someone more technically competent than him was either hired or promoted after he left AMD and they probably changed or tweaked the architecture from the ground up. Raja's brilliance only seems to show up in his excuses, that almost always follow after huge promises and millions or even billions sunk in R&D costs.My point is, I do not accept anymore the claim that he had that much to do with the Navi 2 design from a technological standpoint.
Early in it's design cycle, but with a fundamental emphasis on IF cache as you just underlined. You don't see the paradox?I'm saying that when he left AMD, Navi2 was still early in it's design cycle.
My impression is indeed that while Raja's was still involved to an unknown degree in RDNA1 and CDNA1, the gens after that already have too much of distinct touches from the Zen side already. It's even likely that like with Zen the first gen was in preparation for all the changes afterward. Raja may have contributed, but the overall plan likely was set from above. Integration of IF, IC etc. may all have been demands from above and Raja may have considered that as meddling with his area of expertise. Nowadays CDNA is where AMD pushes packaging tech, wouldn't be surprised if the team behind Zen 2's packaging layout moved onto that next.IF cache is fundamental to Navi 2 and later AMD GPUs. This is not some small design add-on. If he was anywhere close to being deeply involved then it would exist in ARC.
I'm saying that when he left AMD, Navi2 was still early in it's design cycle. After all, he left 5+ years ago, and rumors were swirling long before that time of deep internal dissent.
Are you telling me that in the 3 years between him leaving and the release of Navi 2, all design factors, or even most, were finalized?
My point is, I do not accept anymore the claim that he had that much to do with the Navi 2 design from a technological standpoint. He was involved fully with Vega, and this played exactly as ARC. Too big and power hungry for the performance, plus the PR spin almost identical. Sure, it's all by chance it followed him to Intel.
Two diametrically opposite views here. I say because it was fundamental & because ARC is deficient in this, then, he had no big input in Navi 2. A simple if/then.Early in it's design cycle, but with a fundamental emphasis on IF cache as you just underlined. You don't see the paradox?
I'm not on a quest to defend Raja, not a fan of him to say the least, which is ironic considering the hill I chose today. I'm only arguing we need to weigh his contribution at Intel the same way we do at AMD, in the absence of any other insight. If we consider his work at AMD to be more high level oriented, then we have little reason to consider his architectural input at Intel was any different. If we choose to believe he single-handedly messed up Alchemist and/or Vega, then we need to acknowledge some of his merits with Navi and maybe even Navi 2 groundwork.
But... isn't the guy who had ambitions to become Intel CEO very much upper management?!To be fair, the rx6000 was also Raja.
Raja created something from nothing, and that is not to be laughed at. Some of the stupid ideas like use the iGPU software likely came from clueless corperate management. That seems exactly like the typical "cost saving" decision made by upper level management would look like.
Everything bad with arc feels like upper management stupidity. Shoe in a licensed memory controller. Glue it together. Reuse iGPU drivers. They will write their software for us. Lets add more pointless features!
IF cache is fundamental to Navi 2 and later AMD GPUs. This is not some small design add-on. If he was anywhere close to being deeply involved then it would exist in ARC.
I'm saying that when he left AMD, Navi2 was still early in it's design cycle. After all, he left 5+ years ago, and rumors were swirling long before that time of deep internal dissent.
Are you telling me that in the 3 years between him leaving and the release of Navi 2, all design factors, or even most, were finalized?
My point is, I do not accept anymore the claim that he had that much to do with the Navi 2 design from a technological standpoint. He was involved fully with Vega, and this played exactly as ARC. Too big and power hungry for the performance, plus the PR spin almost identical. Sure, it's all by chance it followed him to Intel.
There was an old interview (it would be great if someone could find it), where asked, if GCN can have more than 4 shader engines, Raja says it could have, but RTG are trying to solve the issue in another way.
In the end, that way did not work- according to HardOCP- because Raja made designs in hardware and then demanded impossible from software. So, we got Vega with the same underutilisation issue as Fiji, and offering the same perf/clock.