k3vst3r:

Impressive, what's the cpu score on timespy?

CPU Score 17 260 https://www.3dmark.com/spy/13103810

Um, so the use of the term client means server stuff like Epyc? That normally goes first, doesn't it? So Zen 3 on the desktop is perhaps unlikely this year. Maybe. Dunno!

Undying:

Will it be enough to finally beat intel in gaming?

Maybe not but that´s not a problem as long as they can stay close enough to Intel. I´m mildly curious about Zen 3 although i´m trying to hold on until Zen 4.

nizzen:

CPU Score 17 260 https://www.3dmark.com/spy/13103810

Absolutely creams my 3900x then, best I've scored is 13,900

nizzen:

1440p is the new 1080p with the new gen gpu's 😉 Beat Intel in "cpu bound" gaming? First they have to dig hard to beat this in gaming: https://www.diskusjon.no/uploads/monthly_2020_07/1184977183_CB205600-4700c17-52001.42vbiosllc8.PNG.c28d30ce54cd92b42bef1cb5553ad307.PNG

Your are an anomaly who posts your ridiculous charts all all over the place. Your are less than a 1% user. For the vast majority your charts me nothing other than "good for you". Try and understand that most are talking about normal users.

Embra:

Your are an anomaly who posts your ridiculous charts all all over the place. Your are less than a 1% user. For the vast majority your charts me nothing other than "good for you". Try and understand that most are talking about normal users.

This forum WAS an enthusiast forum, and now it's just "normal" users left?

kings:

Intel's problem is not the design of the architectures. They already have architectures much faster than the current ones, they just cannot use them to the full, because they were designed for the new manufacturing processes. As 10nm did poorly and 7nm was also delayed, Intel found itself in a situation where the only chance is to continue to squeeze 14nm. When Intel has 10nm working properly or even 7nm, we will see good products from Intel again.

Well, a CPU you cannot manufacture is no CPU at all...;) About the best you can say for that is to call it vaporware, imo. I think that so far the market verdict on AMD vs. Intel is convincingly in favor of the AMD products. Although it's been far too long in coming, I note that the other day the AMD share price finally exceeded Intel's. Intel's past good will is expiring rapidly.

Seeing the rather poor reception of the XT refresh, I think AMD will (try to) go for a proper launch sooner than later. And yes, some are saying pricing could go higher vs what we've seen till the 3000 series...but that would be 'a la Intel' who dictated for over a decade. While on the subject..I was personally considering a desktop Renoir to tie me over the initial back and forth between Navi2 and Ampere, but I'm most probably be going to throw out the window 30 bucks for a 'new' videocard in order not to end up with a weak cpu when that time comes.

nizzen:

No one else is posting any results, just claims with no proof 😛 Like 3900x beating 9900k in games 😛 PS: I want Zen 3 to pop off and crush Intel in both rendering and gaming. Only this will push the "singlecore" performance forward. Now the latency/singlecore performance for zen 2 is 4-5 years old. Singlecore performance done right: https://www.diskusjon.no/uploads/monthly_2020_07/1732469762_Aida644700c175600-5200cacheRTL64IOL7.PNG.8631b043e82ace5b05c1510b1ba5820a.PNG

AMD is never going to be able to hit anywhere near that kind of latency, as long as the memory controller is located off die. Intel's monolithic style architecture is about as optimal for reducing latency as you can get. Even my 6900K at 4.2ghz with DDR4 3400 can hit in the low 50s for memory latency. But that HUGE unified L3 cache and smart prefetchers means that Zen 3 won't have to take as many trips to main memory as an equivalent Intel CPU, so they will definitely gain some ground when combined with a solid IPC increase. Also another factor to consider is that a large amount of AAA games are being designed with DX12 or Vulkan these days, which reduces the reliance on single threaded low latency performance due to the engine being more parallelized.

Here's a good example of how increased parallelization with APIs like DX12/Vulkan makes single threaded and low latency performance less important. The 3900x is actually ahead of the 10900K here: https://imagizer.imageshack.com/img924/5430/j2IR0c.jpg

schmidtbag:

Hey AMD: if you're tired of people questioning Zen3, maybe you should have a less stupid naming convention for APUs (including mobile). The whole reason people are doubting Zen3 this time around is because the 4000 series APUs are based on Zen2.

This time around? Every APU released has been then next generational naming with previous generation technology. 3000 series APUs are zen+, etc. I'm not saying that it shouldn't be this way, because it is surely confusing, but i don't understand what you mean by "this time around"

kings:

Intel's problem is not the design of the architectures. They already have architectures much faster than the current ones, they just cannot use them to the full, because they were designed for the new manufacturing processes. As 10nm did poorly and 7nm was also delayed, Intel found itself in a situation where the only chance is to continue to squeeze 14nm. When Intel has 10nm working properly or even 7nm, we will see good products from Intel again.

The first part of your message is wrong. Intel do have problems with their architecture, in fact, their whole architecture is broken. Intel is struggling with numerous, serious security vulnerabilities from which their whole brand is affected. And before you say, that you have to be physically connected to the PC to exploit these vulnerabilities: no you don't anymore. This is required only for Spectre and some vatiants of Meltdown. But there is more vulnerabilities by now that can be exploited through backdoors and malware already. (MDS, Foreshadow, ZombieLoad etc. just to name a few) AMD did manage to fix all the known vulnerabilities on hardware basis with their ZEN2 Architecture, and they were even able to make !full! software fixes for previous generations. Intel did roll out !some! software fixes to !some! types of Spectre, yet Spectre is still a threat, not to mention the other dozens of vulnerabilities that Intel can't even fix !due to! the fact that their architecture design isn't capable of fixing those problems on hardware AND software level. The only way for intel is to make a whole new design to roll out. But they aren't ready yet as you said. And this won't change for quite a while (years to come probably). By then however, I think AMD will have much better improvements with their ZEN4 architecture and above.

JamesSneed:

If the rumors about the 8-core CCX, unified cache, 15% IPC improvements are true then yes AMD should beat Intel at everything at least until Intel can make a CPU on 10nm+. The 8-core CCX should make all 8-core or less CPU's similar to Intel monolithic cores and as such have similar latencies which is the main issue with the very high FPS gaming. All speculation and rumors right now so have to wait and see but I hope the rumors are true as we need AMD fighting it out with Intel for years to come.

In particular with my explanation from my quote from above, I think Intel's "new" architecture won't be able to beat AMD that easily. Considering the fact that AMD actually was only affected by Meltdown/LVI vulnerabilities, which they managed to fully fix, Intel's new architecture should aim to fix way more vulnerabilities than AMD ever had to, while many of these are affecting performance. If say, Intel were able to fix it with current generation, then you could expect - round about - at the very least 10-15% performance loss compared to AMD's ZEN2 (that's just my personal guess, considering the way how the vulnerability like Spectre and Meltdown actually work). So at some points, Intel manages a better score in single core performance mode because they have so many open doors to insecurely access the CPU workloads, which is not good but a big disaster. Also on long term usage, intel currently has much higher TDP, which you should also consider as an issue, because intel is not just more expensive to buy, they're also more expensive to maintain over time.

Aura89:

I'm not saying that it shouldn't be this way, because it is surely confusing, but i don't understand what you mean by "this time around"

Fair enough, I wasn't very clear on that: This isn't the first or even the second time people have questioned if Zen3 will make it to socket AM4, hence the first sentence of the article saying "fifth time this year [is] confirming that Zen3 is launching this year". So, my point was saying that the reason this time people are questioning if Zen3 will be available is because AMD is releasing Zen2 products on the 4000 series, which is confusing people who don't know that APUs and mobile CPUs "skip a generation". Does that make more sense? So if AMD's product naming wasn't so confusing, they wouldn't have to yet again reassure people that yes, Zen3 is being released this year on AM4 [EDIT] and that the 4000 series isn't going to be a wholly Zen2 rebrand.

Carfax:

Here's a good example of how increased parallelization with APIs like DX12/Vulkan makes single threaded and low latency performance less important. The 3900x is actually ahead of the 10900K here: https://imagizer.imageshack.com/img924/5430/j2IR0c.jpg

No it's not when both are overclocked 😉 In this test 10900k is nerfed to the max.... Try 10900k @ 5300mhz +, 5000mhz cache and 4400mhz memory. Nice try, but try harder next time 😀

nizzen:

No it's not when both are overclocked 😉 In this test 10900k is nerfed to the max.... Try 10900k @ 5300mhz +, 5000mhz cache and 4400mhz memory. Nice try, but try harder next time 😀

While you're correct that a tweaked 10900K would beat the 3900x, the context of my point still stands. 3900x with partitioned L3 cache, off die memory controller, chiplet based architecture and lower clock speed than the 10900K with its monolithic architecture, integrated memory controller and unified L3 still managed to edge what is currently the fastest stock gaming CPU. This is thanks to the huge L3 cache of the 3900x, combined with the Decima engine being able to render across multiple threads.

Carfax:

AMD is never going to be able to hit anywhere near that kind of latency, as long as the memory controller is located off die. Intel's monolithic style architecture is about as optimal for reducing latency as you can get. Even my 6900K at 4.2ghz with DDR4 3400 can hit in the low 50s for memory latency. But that HUGE unified L3 cache and smart prefetchers means that Zen 3 won't have to take as many trips to main memory as an equivalent Intel CPU, so they will definitely gain some ground when combined with a solid IPC increase. Also another factor to consider is that a large amount of AAA games are being designed with DX12 or Vulkan these days, which reduces the reliance on single threaded low latency performance due to the engine being more parallelized.

Don't forget moving to an 8-core CCX will remove all the higher latency cross CCX communications. So while I'm with you that AMD wont beat Intel on the latency front I fully expect for CPU's with 8-cores or less to get really close. Close enough the better IPC will end up with AMD having better performance.

Carfax:

While you're correct that a tweaked 10900K would beat the 3900x, the context of my point still stands. 3900x with partitioned L3 cache, off die memory controller, chiplet based architecture and lower clock speed than the 10900K with its monolithic architecture, integrated memory controller and unified L3 still managed to edge what is currently the fastest stock gaming CPU. This is thanks to the huge L3 cache of the 3900x, combined with the Decima engine being able to render across multiple threads.

Bend the rules: 10 core 10900k oc 14nm++++++++++++++++++++++++ (LOL), is faster than 7nm high tech 12 core 3900x cpu. 😛 PS: I have both 3900x and 10900k, and use them for different tasks. Now 10900k is for benchmarking ONLY 😀

JamesSneed:

Don't forget moving to an 8-core CCX will remove all the higher latency cross CCX communications. So while I'm with you that AMD wont beat Intel on the latency front I fully expect for CPU's with 8-cores or less to get really close. Close enough the better IPC will end up with AMD having better performance.

For 8 core CPUs, but what about the 12 and 16 core? Those will likely still use multiple CCXs. Speaking of which, it's quite possible that AMD will only have Zen 3 CPUs with multiples of 8 CCXs, so the lowest SKU would have 8 cores. That's just speculation on my part though. AMD will also hopefully increase the speed of the IF connect for Zen 3.

Carfax:

For 8 core CPUs, but what about the 12 and 16 core? Those will likely still use multiple CCXs. Speaking of which, it's quite possible that AMD will only have Zen 3 CPUs with multiples of 8 CCXs, so the lowest SKU would have 8 cores. That's just speculation on my part though. AMD will also hopefully increase the speed of the IF connect for Zen 3.

Right, over 8 cores I expect higher latencies. Of course Intel has a little higher latencies over 10 cores since they switch over to mesh instead of there ring bus but its still much better than AMD's cross CCX latencies. I think this is perfectly fine trade off that allows AMD to offer great gaming 8 core CPU's(or whatever else max single threaded performance is needed) that will hit very high FPS and if you really need more threads you know there will be a little sacrifice due to increased latencies. I think this fits much better in the real world since frankly if you want very high FPS you are looking at a 6-8 core CPU today due to how the thermals play out.

niz:Dzen:

Bend the rules: 10 core 10900k oc 14nm++++++++++++++++++++++++ (LOL), is faster than 7nm high tech 12 core 3900x cpu. 😛

It would be funny if Intel was just playing with nodes and $$ like Nvidia, and if they could move to 7nm at will. They're stuck with their foundries and litog. issues. That's not funny.