Whokeys Black Friday Sale 25% code: RGT Windows 11 Pro $23,2:biitt.ly/581eD Windows 10 Pro $17,6: biitt.ly/9f0ie Windows 10 Home $15:biitt.ly/XYm9o Office 2016 $28,8:biitt.ly/cy7Vb Office 2019 $48.6:biitt.ly/YInvw #blackfriday#Windows11pro#whokeys whokeys.com
The problem with increasing CPU core count is not the CPU's themselves, it's the software running on them holding the CPU back by being poorly optimised for parallel processing. Even today, most games are still very poorly optimised for even 8 cores. [Edit] Most non-gaming software is also, still poorly multi-core optimised.
It's linear with the cpus in consoles. If ps6 were to get faster cpu, games for it will optimise differently and so it will affect pc performance as well.
That's because games are not a great workload for parallel processing. Don't expect CPU optimization to change much. Sure, some games can "utilize" a CPU a lot, like The Last Of Us Part 1 on PC. But it wastes a lot of that CPU time doing nothing. You don't need multiple CPU cores to render a person or 2 standing in an empty house doing nothing, and yet that is constantly the case in that game.
Games is not a great example. The parallel processing where possible is offloaded to the GPU because it’s superior in parallel processing. In fact more software should utilise the GPU.
PCs do much more than gaming. Try looking at it from non gaming lens. There's a big market for high core count CPUs. Intel's i9 and ultra 9 processors as well as AMDs Ryzen 9s are NOT gaming CPUs so we shouldn't be judging them solely by their gaming performance or the fact that games aren't optimized for them.
Cpu bandwidth is a big problem if not the biggest problem. DDR5 doesn't provide enough bandwidth...DDR6 IS badly needed if performance is to improve by a bigger than usual amount
More cores per ccd would be nice. But a complete redesign for the IOD and placement of it needs to be done. The IF is the biggest limiting factor for zen.
@@shifu-mn6py That'll literally make ryzen opposite of of what it successful in the first place, lots of cores for cheap. It is literally the most idiotic thing to do.
@DragonOfTheMortalKombat except right now ryzen CPUs and builds are more expensive than the counterpart You can absolutely do some basic tinkering on a U7 chip and hang out with or beat a 9900/9950x You can also do some basic tinkering and have a raptor lake chip keep up with or beat a 7800x3d for less. The second they paid off the tech tubers to contort the benchmarks to show the 7800 as the affordable David beating the expensive Goliath they jacked the pice past an i9 and to this day people act like half a cpu for more money is a better deal. Mark my words man amd is going to be such an expensive platform to buy into here in the future cause of this.
For a company that came back from the brink of bankruptcy to obliterate Intel with Ryzen technology, AMD's "AI Max+ 395" branding simply gives its supporters goosebumps.
I'd love to see one of those HEDT features brought back; 3-4 channel memory. It's no secret cpus are bandwidth-starved since zen2 and the situation is not getting better, no matter the new tech and all, it's just 2 channels, double data rate blablabla... Would be cool to have the same package but different boards/chipsets offering different memory channel configurations, a bit like lga1366 back in the day.
Nah^ Skip this, give us a NEW IO Die, that can gear 1 the 8000mhz so we see a doubling in bandwidth. As an ITX user, I don't wanna have to buy a new Motherboard, I wanna update my bios, slot in the CPU, and get better performance from my 48GB 8000 kit.
Fully agree, 4 channel should be the norm, If they put 4 sticks it means it should be 4 channel not 2 channel. Anyway 4 sticks do not work for high memory speed sticks because of 2 channel scam.
@@melheno yes. 4 sticks generally runs 10% lower clocks. The reason is ddr5 has hundreds of pins. 4 channel setup would have to have something like 30% more pins on CPU and motherboard as well as a decent but more silicon on the die dedicated to memory controller. Even if you want 2 channels.. the platform would be more expensive because of these upgrades.
@@christophermullins7163 sometimes it is much more than 10%, and buying a motherboard becomes a lottery. 90% of the motherboards support 4 sticks, every advise you get use 2 sticks to get normal speed. I am not asking OC. 4 Channels 6400. It is not that expensive or difficult to do it is just the same shit what Apple does with its RAM and SSD upgrade. You want more bandwidth more ram, pay 5 times more to get Threadripper when you just need more bandwidth not 64 core cpu.
Yes! I remember this as well. They even had versions of the regular Pentium 4s that had more cache, they just weren't really pushed and were really expensive. The larger cache 3.0 GHz version was more expensive than the lower cache 3.4 GHz version. I managed to grab one for a build I did in the early 2000s and that extra cache made a huge difference! Was able to overclock the 3.0 to over 3.4GHz anyway, it was a great setup. I don't know why that concept didn't progress
Arctic Wolf? No, you're not alone. Nova Lake also reminds me of Nova Prospect from HL2. I guess we're dating ourselves with these older gaming references haha.
Point 1. IPC gains are about the total package because the only way that both Intel and AMD consider IPC is a percentage of uplift in applications/games from gen to gen. IPC is NOT something that is 100% limited to gains in changes to the cores. It's the ENTIRE CPU. So, architectural changes can improve IPC even if the cores are the same. Point 2. AMD isn't worried about core count for Intel, they're worried about thread count. With the ability for Intel to mash in lots of e cores into a processor AND moving to a better process node, that could allow them to get above 32t which is basically where Intel had gotten to and AMD has been at since Zen 2. For as much as AMD has avoided mixing cores, they may be mixing cores and making a chiplet that's 8x2 or 6x4, or make multiple types of chiplets for desktop/server. I could see them getting up to 40 - 48t for desktop which is something I've said they needed to do to stay ahead of Intel, but they can't do it with their current architecture because it doesn't have the bandwidth between the chiplets. AMD is rumored to be moving to direct connects between chiplets, which would get rid of the Infinity Fabric in the form it is now though they could keep the name even though there is no more multiplexer and no need to convert between serial and parallel data (serial transfers between chiplets which is slower, but necessary because the data moves across the PCB). AMD has said that DDR5 isn't fast enough to allow them to move up in core counts. THIS doesn't mean it's NEVER fast enough. It means that in memory intensive workloads there isn't the bandwidth. So, with direct connects between chiplets, that may allow AMD to use the fastest DDR5 kits which might be enough for 40t, but I highly doubt it would be fast enough to feed 48t. And to be clear the issue with faster memory not benefiting Zen 4 - 5, as in faster than about 6400 Zen 5 and 6000 with Zen 4 is because of the IF, not because of the cores. You can't get data across the IF fast enough. So, the Infinity Fabric gets in the way of making Zen CPUs more capable than they have gotten to with Zen 5. MAYBE they could squeeze a little more out, but as has been constantly said on different channels, especially Level1Tech, you need the bandwidth to feed the cores. It's possible that AMD moves to the direct connects that are used with RDNA 3. They aren't great, but it's a balance of bandwidth and cost and it's certainly better than needing a die just to move data between chiplets, as in the multiplexer on the IOD. That should be fast enough for 40 - 48t.
You're correct about Intel and cache -- the Itanium had a (for the time) enormous cache, taking advantage of the fact that the cores were remarkably compact. It never hit its stride thanks to a series of missteps within, but before it died it was a leader in floating point performance in spite of a seemingly mediocre clock speed. I do expect AMD to increase its core counts in Zen6, though you're probably right about the Ryzen series remaining capped at 16 cores since that's made for consumers. My guess is that we'll see the core counts rise -- performance core counts, that is -- in the HEDT and server lines, because the HEDT market is a sweet spot for workstation users that aren't big VFX studios. :) On the server side of things we'll probably see a big surge in core counts to head off the ARM server processors that are already reaching 192 cores.
I just hope NVL will plop into LGA1851. I think the reason LGA 1851 socket has more pins in the first place is to possibly to accommodate those future chips.
At least for gaming, more cores is pointless. There's usually one or two main threads being the limiting factor, with some other smaller threads and handling the OS being nice to have. I'd rather have fewer, high power cores without SMT to make the absolute most out of them, then some additional cores with SMT to handle multitasking better.
Apparently Intel was working on something like this where they had 4 cores that could run as 1 huge core if needed but they dropped the idea. What a shame.
The i7-3970X might have been 1000 USD, but already back in 2014 there was the 5820K with 6 cores for 389 USD. The pricing for 6-core processors had reached an acceptable level when Ryzen launched, however 8-core CPUs were still at 999 USD list price.
The new packaging methods that would allow lower power consumption, increase bandwidth and reduce latency all have something in common. They also allow AMD to put chiplets far closer to the IO die. The package has enough space for having three core chiplets if components in the package could be placed way closer to one another.
A Ryzen with a 8core+X3D ("gaming") die and a second 16 core compact die for a total of 24c/48t would be nice. Plus improved IO die for better memory bandwith (triple oder quad channel). But thread scheduling would probably be a minefield for the first iteration as with 7900X3D.
I don't think more cores is the answer at the moment. We are already maxing out the dual channel memory setups with 16 cores. I can even max out quad channel memory with 32 cores. What we need is a consumer gaming platform with more memory channels and more PCI-E lanes.
Exactly, there are simply not enough PCIE lanes in any consumer CPUs nowadays to built cost effective workstations that can double as gaming rigs. If every creator needs a fat wallet and a rich mom to buy a Threadripper that can't be the answer. There aren't even enough slots that come out as x8 to support a good 40 or 100Gbit NIC or some extra SAS controllers or NVME expander cards. That's why so many people build extra machines based on older gen EPYCs or XEONs none of the later are great for gaming of course...
Facts just give us 16 cores on one CCD, faster memory controller and greatly improved IO and we are all good. The architecture of Zen 5 is already the best.
23:12 Bro, Intel had 128MB of additional eDRAM on their 5th gen Core lineup - Broadwell architecture. It was called Core i7 - 5775C. It went unnoticed because it wasn't marketed well, and few weeks later intel released rather famous 6th generation (skylake) with DDR4 support. This i7 - 5775C was a beast for gaming, it was faster then successor 6th gen and in some games even faster than 7th gen! Especially after overclocking. So intel had it before AMD and they new what wonders it can give. It looks like intel was too confident with their new 12th gen architecture and thought additional memory inside CPU will not be necessary. But they got caught off guard by AMD with their x3D chips, and it was too late changing 12th, 13th and 14th gen architecture.
Intel had big caches definitely before AMD, however it was L4, not L3 and it was not stacked, but integrated. But you are correct, it was a beast for certain type of workloads.
My sources have said slightly differently to this, firstly AMD might not be even making 8 core chiplets anymore, they will all be either 3nm 16-cores Zen 6 or 2nm 32-core Zen 6c but these are exclusive to the server market. Obviously 8-core SKUs will exist but these will simply be cut down chips for the lowest end SKUs as AMD wants to push on the core count and performance so Intel can't find their feet. IPC (or performance in general) for Zen 6 will actually be very decent, this is for several reasons, firstly the performance gains for Zen 5 didn't work properly because of the IMC issues so we will be getting Zen 5 + Zen 6 improvements. Secondly, having 16 cores all on the same chiplet will help performance of applications that are boosted by more than 8 cores but would normally be impacted by cross chiplet latency. The only caveat to all this performance is memory bandwidth, AMD did toy with the idea of having Zen 6 on both AM5 (dual DDR5) and a high end version on an early release of AM6 with DDR6, even quad channel was discussed to make sure 32c/64t CPUs were kept adequately fed. I'm not sure if they are still going ahead with this as of now or if they are pulling the launch of Zen 6 in because Zen 5 missed the mark so badly and so AM6 and DDR6 don't be ready in time. On the Intel side, 100% Intel is aiming for 16p + 32e, they feel they have to to keep up, they need a win and they see multitasking as the easiest option right now. Plus this gives them a lot of room for SKU options which Intel always likes to have. I also hear that one of the main reasons Nova Lake isn't on LGA-1851 is because of the same memory bandwidth issues as AMD is facing, dual channel just doesn't cut it anymore, so high end Nova Lake will have quad channel controllers and lower core count SKUs will just have the cheaper dual channel controllers, I also believe both controllers are DDR5 and DDR6 compatibility so that Intel has cheaper low end DDR5 options and higher performance DDR6 options. This was also the reason for separating the memory controller back out of I/O die again so there doesn't need to be larger more expensive quad channel parts on the low end SKUs. Obviously motherboards will be sold in cheaper dual channel options or expensive quad channel variants. Again Nova Lake should see a really big increase in performance because it's a full redesign from the ground up with a lot of new ideas, plus Intel won't have moved the needle forward for about 6 years by that point! I just hope both companies can deliver on their initial aims, time will tell I guess.
@@corey2232 if you can tell a CPUs performance by its ghz.. that means you are seeing no IPC gains on newer architecture. That is a failure.. as it should be. Why would you want to be in a time where there are no IPC gains?
@@christophermullins7163 Because now you can't tell a CPU's performance either by its clock speeds or its core count. I don't care either way, but I'm assuming the OP meant it used to be easier to identify X CPU > Y CPU.
Yeah Zen 7 for the desktop would be interesting if it had a core count increase. Even from 16 to 24 although I would prefer 32 cores on the desktop even if it were a mix of Zen 7 standard and Zen 7 dense and utilizing 3DV cache. Sadly we just have to wait and see.
I think that general systems should really become more like consoles. Design a system of three levels of processing power. Run a CPU, GPU, and NPU. Optimise the absolute crap out of the OS and applications to use the platform. I would buy a PS5 Pro if it could run a parallel NVME with a Windows/Linux desktop natively and I could switch between "gaming" and "work" OSs. Look at people buying a 4090 for $1800, the PS5 Pro is going to be $700 I think. If there was a PS5 Ultra which cost $2000 I would think it would quite a beast, and then with all the optimisations it would be really good.
16 core per ccd would make latency slower in overall ccd since you need make bigger L2 cache to connect everything, it not worth it until software/gaming catch up
At 17:30 four cores in the background are system (SIP) management processors. At 20:30 u mean the E cores are finally really going to be used as a SIMD parallel array Intel has been working on the compiling tools since APO but this is much more than APO. At 23:15 right Intel technically true many first cache attempts (implementations) and Intel produced MCM for 30 years for industrial, military and aerospace applications. There were forebears that are cache designs Intel stole and others stole from DEC. AMD bought those DEC patents in 1996 that became Hyper-transport. DEC did most of the x-way associative write back cache development. Way back when at 24:00 1990s NexGen invented, well implemented, the first processor direct L2 that Intel borrowed at Pentium ll. Nx586 is the first L2 direct processor implementation and with IBM boards is my product marketing and platform validation responsibility before AMD K6. mb
No ZEN 6 nor Nova Lake until Q4 2026 is pretty bad news imo... It might be good for consumers because they'll keep their hardware longer, but on the other hand it means that CPU performance is not going to increase as much over the next 2 years and then we're going to have the RTX 60 and RDNA 5/6 GPUs that could be highly bottlenecked by CPUs! Let's hope that ZEN5 will at least get a Refresh with a better I/O Die !
I would love to see more cores on each CCD even if its just to 10 on each to see if "only need 8 core for gaming " is true LOL I do know that when I had the 7950x3d games loaded faster than 7800x3d but thinking of 32 cores on 2 CCD's with x3d on both OMG yes I dont care if I need it I just want it LOL
I wonder if a single CCD with 16 full cores calls for AIO cooling as standard. I've got a 9800X3D with an air cooler and it runs cool. I just wonder at the heat output though if 16 cores are within one CCD.
The problem with Adding more cores is the power each requires & heat each core puts out affect clocks and make cooling harder, cores can't clock as high as a result then add in size of chip affecting room on boards needed and you have to ask does average user with Windows, a few productive software uses and games need more than 16 cores at present mostly no but for those that do AMD have thread ripper. That said with latest improvement in efficiency AMD are positioning to up core count i feel not Zen 6 but with Zen7 that could be on new socket as a result, also as AMD kept multithreading they don't need as many cores to beat Intel most of the time who ditched it.
AMD said that Zen5 is foundational so SMT should stay the same. Zen6 should be a redesign of the cache hierarchy. Since Zen5 X3D shows that the old cache system cannot feed the cores without significant increase in L3. Maybe they can finally make the dual fetch decode work for a single thread but that's a big challenge and I expect a halfway designed where dual fetch decode works on some single thread workload but not every workload.
If ZEN6 is really to be released in late 2026 or even early 2027, than this would mean, AMD has a room for easy ZEN5 refresh with new IO die. It's almost as if AMD planed to release ZEN5 with old IO die dragging down the performance - releasing infamous "ZEN5%" since new Intel Core Ultra is not a threat. A refresh in between with new IO die seems like a next easy (cash-grab) step.
@@madb132 Intel is still struggling against Intel, yes, but they did legitimately beat not just their last gen parts but even the current gen Ryzen 16 cores in some multithreaded tests…. At a thread count disadvantage no less.
Zen is better suited to chiplet design. Since Ryzen 1000 was infact chiplet design and the whole infinity cache and memory optimization AMD is doing since 2013 i.e when development of ryzen started has paid off. They now know how to make a chiplet design work. They have done many iteration of Ryzen of 4 ccx, 2ccx design. Large server chips. They have covered their basics.
I can dream of Intel having 8+32 with big cache for pcores. If AMD could fit 16 pcores under one CCD with 3D cache it would be also tempting proposition. Looks like I am going to stuck to 5800x3D till one of those scenarios become real.
I am on 9800X3D. This caught between a rock and hard place and the desire for more than 8 homogenous cores on one die has made me swap parts a lot the last 1-2 years. I was willing to compromise and go Raptor Lake and give hybrid a chance, but Raptor Lake stability/degradation issues I had turned me off and returned or sold some chips I tried. They all gave e stability problems eventually within a month, So settled on 9800X3D after Arrow Lake disaster which I sold my 7800X3D system for thinking Arrow aLake would be a better Raptor Lake and stable and have same latency. But its tile design atency crippled gaming performance badly so it was a no go. 8 cores and they are much faster and tunable it is for me now on 9800X3D.
Next generation should be good for both companies. AMD has the best architecture. However they are limited by their IO, memory controller and not as much PCIE lanes as Arrow Lake.
My guess for Zen6: Server Zen6 dense: 32 core CCD in N2 Server Zen6: 16 core CCD in N2 Desktop Zen6: 8 core CCD in N3 N2 is rumoured to be around 50% more expensive per wafer while only being 10-15% more dense so cost/transistor would be higher. It has better power consumption so it is orobably used for datacenter where this is a bigger factor. On desktop many systems doi not need more than 8 cores so I see 2 possibilities: a) They are just going 8 cores again (more likely) b) They justbdesign a 16 core CCD for high end and rebrand Zen5 for everything below Regarding Intel: If the dual CCD rumours are true Intel is dead for gaming as they will run into the same issues that AMD currently has with 7950X3D.
It sounds like we are reaching the current limit on modern transistor size/count performance improvements. I must be living under a rock as Ihad notheard Intel has dumped hyperthreading. Like, WHAT?
Arrow lake sorely need some optimization/fixes to be fully realized. There's been a handful of overclockers and such disabling all but 1 P-core and gotten some interesting results in games. I think by the time we get to CES in January and beyond there will need to be a lot of retesting done by reviewers. I'm also kinda curious if the 9950X3D will play better this time round vs 7950X3D.
AMD Zen6 Ryzen 11950X has 16 cores and 32 threads and 64 MB of L3 cache and 16 MB of L2 cache and supports 6400 MHz DDR5 RAM instead of 6000 MHz. it will be 1 to 3 percent faster in games and 5 percent faster in multi threaded apps. Nova lake will have 8P cores, 16E cores and possibly 4 LPE cores. Novalake is 5 percent faster than Arrowlake in games so Novalake 385K is as fast as 14900KS in games and it will be 5 percent faster than 11950X in multithreaded apps. it runs at 5.6 GHz and supports DDR5 10000 MHz memory. it will work for 2 years until it degrades and breaks. it consumes 225 watts of power. 11800X3D will beat 385K in gaming by 20 percent. the reason AMD does not use 16 core CCDs for gaming CPUs is because dual channel DDR5 is extremely slow and can not feed 32 cores. quad channel DDR5 or dual channel DDR6 or dual channel QDR5 (Quad data rate) is required to feed 32 cores with data. AM6 socket will have 32 core flagships because it can handle more cores because it has better VRMs and power delivery system and will not bottle neck 32 cores because it will support DDR6. adding 32 cores to AM5 is useless because AM5 socket can only max out at 230 watts and can not handle feeding data to 32 5+ GHz cores. Novalake will not be able to beat Zen5 X3D let alone Zen6 X3D in gaming. dont forget that AMD DDR5 6400 MHz is equal to Intel DDR5 8400 MHz. AMDs memory timings are far tighter than Intels.
144 MB of L4 cache is slower than 3D VCache because 3D VCache is L3 cache. L3 cache is faster than L4 cache. if Intel moves L3 cache to a separate die latencies will increase and performance will decrease.
Oh lol Intel 2 Tiles with 8 + 16. Will be same latency penalty as AMD on Zen 3 through 5 dual 8 core CCDs lol. Or worse as shown by Arrow Lake tile packaged design. Zen 5 X3D 9800X3D is where it is at for now for gamers. Games yet to use more than 8 cores except Cities Skylines 2 extremely large populations or some select few games that get marginal benefits from them, Dual CCD 9950X3D if both CCDs have the 96MB may be interesting as no hybrid. Yes still dual CCD latency, but at least both good cache and no. Maybe Zen 6 even if they keep 8 core CCDs the IO die they will configure some ladder cache that will completely ax any latency between CCDs so crossing CCDs to talk to a core is just as fast as taking intra-CCD on Zen 6. One can hope!!!
A lot of comments below station that games/software is not optimized for more the 8 cores this will always be the case, but certain games do take advantage. For example "Eveonline" - I play this game using a 7950X, now most players switch on max graphics and play as a solo player in full screen just like they would play any other game but because the game has many different avenues of play then open up one way of playing using multiple accounts at the same time eg; when mining asteroids as a group of ships under one persons control. Now the games devs were thinking greedily at the time so as not to limit the game to just one person at a time (one player playing two accounts means double the monthly income and why limit it to 2 accounts why not make this a no limit option). Now yes this is extremely unusual to allow multiple games to run at the same time but this is made possible because you can play in windowed mode and I actually play 13 accounts at the same time and I am CPU/memory bandwidlth limited now and even find no advantage using the 7950X3D. If the 9950X3D comes with double 3DV cache then I will get one and try it. But higher core count for gamer's is probably of no use because the GPU is the bottleneck which is why I reduce the GPU load in Eveonline because for me the pretty graphics is not what is important but instead the amount of accounts that I can use at any one time. BTW many players do this (This is called multi-boxing) because to increase one's game monetary income means either play more time (only so many hours per day you can play) or increase the amount of accounts that you play at anyone time. Many solo miners don't like Multi-boxer miners but also many will ask for boosts from our main ships. Which is how using multiple accounts help us benefit. For people that have never heard of Eveonline, its a multi-faceted Space game with different game styles available for each individual from Mining asteroids, Pirating, Trading, Building or even just playing the markets like a stock broker. It has been in existence since 2003 and has evolved over time with especially with the graphics as PC tech has improved.
@@jackthatmonkey8994 Threadripper is too expensive - new CPU, MB I wish they did dual socket AM5 M/b's but 7950X's and are not dual socket compatible !
I feel like you answered yourself with your own example - dual-CCD SKUs like the 7950 amply cater for the niche use-case of being a 13-account multiboxer.
Yikes! I play X4 (a game where you can command fleets of hundreds of ships) in 4K and my Radeon 6950XT bottlenecks my Ryzen 3 3300x (quad-core) while listening to music on youtube in the background. Almost every game I play is gpu bottlenecked at 4k. I'll upgrade to a R5 9600X when the RTX 5090 releases. I think big core cpus are overrated for 4K gaming.
Bro AM5 has already been delided we clearly know the size of the chip on the package... That said both AM5 and LGA1851 is a hard pass tor anyone with decent setup on AM4/LGA1700...
so?....are intel trying to save money on r&d by using design's for mobile cpu's in desktops??? or what is this about , cause i don't know anyone that are interested in e-cores for desktop pc's apart from intel themselves
Clustered Multi Threading (CMT) is the exact opposite of Simultaneous Multi Threading (SMT). AMD experimented with CMT on AMD Bulldozer, Piledriver, Streamroller and Excavator microarchitectures and it was a total failure. if Intel uses CMT then AMD wins.
"Gamers only need 4 cores" - Intel. That did not age so well huh? Bulldozer had 8 cores? That's perhaps more legit than Intel saying they have 20 or so. Intel markets a high core count, but they only disclose in small print how most of the cores are P cores - P for puny. A case for class action perhaps: 'You thought you were getting ~20 cores and then found your new CPU only has 8 proper cores and the rest were puny cores? Better Call Saul!'. 🤔
@@FinneousPJ1 And still many games cannot really use more than 6-8 cores. There are definitely games out there which can use more cores, e.g. Factorio, but those are not the majority. In the end it's always: think first, buy afterwards.
Whokeys Black Friday Sale 25% code: RGT
Windows 11 Pro $23,2:biitt.ly/581eD
Windows 10 Pro $17,6: biitt.ly/9f0ie
Windows 10 Home $15:biitt.ly/XYm9o
Office 2016 $28,8:biitt.ly/cy7Vb
Office 2019 $48.6:biitt.ly/YInvw
#blackfriday#Windows11pro#whokeys
whokeys.com
Where is the derbauer link?
Oops, sorry about that. I added it now, thanks!
Video starts at 5:08
Video starts at 25:55
Thank you
Video ends at 29:30
@@JYMBO ^^^ This. there was just too much rambling just to say core counts are going up. really, a 30 minute video to just say that
@@JYMBO lol
The problem with increasing CPU core count is not the CPU's themselves, it's the software running on them holding the CPU back by being poorly optimised for parallel processing. Even today, most games are still very poorly optimised for even 8 cores.
[Edit] Most non-gaming software is also, still poorly multi-core optimised.
It's linear with the cpus in consoles. If ps6 were to get faster cpu, games for it will optimise differently and so it will affect pc performance as well.
That's because games are not a great workload for parallel processing. Don't expect CPU optimization to change much.
Sure, some games can "utilize" a CPU a lot, like The Last Of Us Part 1 on PC. But it wastes a lot of that CPU time doing nothing. You don't need multiple CPU cores to render a person or 2 standing in an empty house doing nothing, and yet that is constantly the case in that game.
Games is not a great example. The parallel processing where possible is offloaded to the GPU because it’s superior in parallel processing. In fact more software should utilise the GPU.
PCs do much more than gaming. Try looking at it from non gaming lens. There's a big market for high core count CPUs. Intel's i9 and ultra 9 processors as well as AMDs Ryzen 9s are NOT gaming CPUs so we shouldn't be judging them solely by their gaming performance or the fact that games aren't optimized for them.
Cpu bandwidth is a big problem if not the biggest problem. DDR5 doesn't provide enough bandwidth...DDR6 IS badly needed if performance is to improve by a bigger than usual amount
More cores per ccd would be nice. But a complete redesign for the IOD and placement of it needs to be done. The IF is the biggest limiting factor for zen.
i actually think it is
No duel CCD.
More ccd and memory controller
@@shifu-mn6py That'll literally make ryzen opposite of of what it successful in the first place, lots of cores for cheap. It is literally the most idiotic thing to do.
@DragonOfTheMortalKombat except right now ryzen CPUs and builds are more expensive than the counterpart
You can absolutely do some basic tinkering on a U7 chip and hang out with or beat a 9900/9950x
You can also do some basic tinkering and have a raptor lake chip keep up with or beat a 7800x3d for less.
The second they paid off the tech tubers to contort the benchmarks to show the 7800 as the affordable David beating the expensive Goliath they jacked the pice past an i9 and to this day people act like half a cpu for more money is a better deal.
Mark my words man amd is going to be such an expensive platform to buy into here in the future cause of this.
I'm excited for Nova Lake.
2026 or 2027 is when I'll be upgrading my PC.
I look forward to the age of 18A and beyond.
For a company that came back from the brink of bankruptcy to obliterate Intel with Ryzen technology, AMD's "AI Max+ 395" branding simply gives its supporters goosebumps.
Obliterate? hahahaha youre high af. look at market share. intel shits on amd
I'd love to see one of those HEDT features brought back; 3-4 channel memory. It's no secret cpus are bandwidth-starved since zen2 and the situation is not getting better, no matter the new tech and all, it's just 2 channels, double data rate blablabla... Would be cool to have the same package but different boards/chipsets offering different memory channel configurations, a bit like lga1366 back in the day.
Nah^ Skip this, give us a NEW IO Die, that can gear 1 the 8000mhz so we see a doubling in bandwidth. As an ITX user, I don't wanna have to buy a new Motherboard, I wanna update my bios, slot in the CPU, and get better performance from my 48GB 8000 kit.
Vcache does mitigate the reliance on fast memory to a degree.
Fully agree, 4 channel should be the norm, If they put 4 sticks it means it should be 4 channel not 2 channel. Anyway 4 sticks do not work for high memory speed sticks because of 2 channel scam.
@@melheno yes. 4 sticks generally runs 10% lower clocks. The reason is ddr5 has hundreds of pins. 4 channel setup would have to have something like 30% more pins on CPU and motherboard as well as a decent but more silicon on the die dedicated to memory controller. Even if you want 2 channels.. the platform would be more expensive because of these upgrades.
@@christophermullins7163 sometimes it is much more than 10%, and buying a motherboard becomes a lottery. 90% of the motherboards support 4 sticks, every advise you get use 2 sticks to get normal speed. I am not asking OC. 4 Channels 6400. It is not that expensive or difficult to do it is just the same shit what Apple does with its RAM and SSD upgrade. You want more bandwidth more ram, pay 5 times more to get Threadripper when you just need more bandwidth not 64 core cpu.
I remember when Pentium 4 Extreme Edition came out, and it had a HUGE 2MB L3 cache, which was like 4x the regular Pentium 4's.
Yes! I remember this as well. They even had versions of the regular Pentium 4s that had more cache, they just weren't really pushed and were really expensive. The larger cache 3.0 GHz version was more expensive than the lower cache 3.4 GHz version. I managed to grab one for a build I did in the early 2000s and that extra cache made a huge difference! Was able to overclock the 3.0 to over 3.4GHz anyway, it was a great setup. I don't know why that concept didn't progress
Arctic Wolf? No, you're not alone. Nova Lake also reminds me of Nova Prospect from HL2. I guess we're dating ourselves with these older gaming references haha.
Point 1. IPC gains are about the total package because the only way that both Intel and AMD consider IPC is a percentage of uplift in applications/games from gen to gen. IPC is NOT something that is 100% limited to gains in changes to the cores. It's the ENTIRE CPU. So, architectural changes can improve IPC even if the cores are the same.
Point 2.
AMD isn't worried about core count for Intel, they're worried about thread count. With the ability for Intel to mash in lots of e cores into a processor AND moving to a better process node, that could allow them to get above 32t which is basically where Intel had gotten to and AMD has been at since Zen 2.
For as much as AMD has avoided mixing cores, they may be mixing cores and making a chiplet that's 8x2 or 6x4, or make multiple types of chiplets for desktop/server. I could see them getting up to 40 - 48t for desktop which is something I've said they needed to do to stay ahead of Intel, but they can't do it with their current architecture because it doesn't have the bandwidth between the chiplets.
AMD is rumored to be moving to direct connects between chiplets, which would get rid of the Infinity Fabric in the form it is now though they could keep the name even though there is no more multiplexer and no need to convert between serial and parallel data (serial transfers between chiplets which is slower, but necessary because the data moves across the PCB).
AMD has said that DDR5 isn't fast enough to allow them to move up in core counts. THIS doesn't mean it's NEVER fast enough. It means that in memory intensive workloads there isn't the bandwidth.
So, with direct connects between chiplets, that may allow AMD to use the fastest DDR5 kits which might be enough for 40t, but I highly doubt it would be fast enough to feed 48t. And to be clear the issue with faster memory not benefiting Zen 4 - 5, as in faster than about 6400 Zen 5 and 6000 with Zen 4 is because of the IF, not because of the cores. You can't get data across the IF fast enough.
So, the Infinity Fabric gets in the way of making Zen CPUs more capable than they have gotten to with Zen 5. MAYBE they could squeeze a little more out, but as has been constantly said on different channels, especially Level1Tech, you need the bandwidth to feed the cores.
It's possible that AMD moves to the direct connects that are used with RDNA 3. They aren't great, but it's a balance of bandwidth and cost and it's certainly better than needing a die just to move data between chiplets, as in the multiplexer on the IOD. That should be fast enough for 40 - 48t.
You're correct about Intel and cache -- the Itanium had a (for the time) enormous cache, taking advantage of the fact that the cores were remarkably compact. It never hit its stride thanks to a series of missteps within, but before it died it was a leader in floating point performance in spite of a seemingly mediocre clock speed.
I do expect AMD to increase its core counts in Zen6, though you're probably right about the Ryzen series remaining capped at 16 cores since that's made for consumers. My guess is that we'll see the core counts rise -- performance core counts, that is -- in the HEDT and server lines, because the HEDT market is a sweet spot for workstation users that aren't big VFX studios. :)
On the server side of things we'll probably see a big surge in core counts to head off the ARM server processors that are already reaching 192 cores.
I just hope NVL will plop into LGA1851. I think the reason LGA 1851 socket has more pins in the first place is to possibly to accommodate those future chips.
Pretty sure it's already confirmed LGA1851 is dead. NVL will use a new socket.
@@1idd0kun From what source? I heard mixed speculations. Nothing has been confirmed by Intel.
Nova lake is for lga1900+
Lga1851 is probably dead platform unless intel decides to keep it alive after panther lake
At least for gaming, more cores is pointless. There's usually one or two main threads being the limiting factor, with some other smaller threads and handling the OS being nice to have. I'd rather have fewer, high power cores without SMT to make the absolute most out of them, then some additional cores with SMT to handle multitasking better.
Apparently Intel was working on something like this where they had 4 cores that could run as 1 huge core if needed but they dropped the idea. What a shame.
Not be pedantic, but how could you forget the Phenom II X6? I wouldn't consider that HEDT. I wanted one so bad back in the day.
I had one, it outperformed the trash FX cpus that came out after it!
It's pre-bulldozer and had no multi threading
The i7-3970X might have been 1000 USD, but already back in 2014 there was the 5820K with 6 cores for 389 USD. The pricing for 6-core processors had reached an acceptable level when Ryzen launched, however 8-core CPUs were still at 999 USD list price.
The new packaging methods that would allow lower power consumption, increase bandwidth and reduce latency all have something in common. They also allow AMD to put chiplets far closer to the IO die. The package has enough space for having three core chiplets if components in the package could be placed way closer to one another.
A Ryzen with a 8core+X3D ("gaming") die and a second 16 core compact die for a total of 24c/48t would be nice. Plus improved IO die for better memory bandwith (triple oder quad channel).
But thread scheduling would probably be a minefield for the first iteration as with 7900X3D.
I don't think more cores is the answer at the moment.
We are already maxing out the dual channel memory setups with 16 cores.
I can even max out quad channel memory with 32 cores.
What we need is a consumer gaming platform with more memory channels and more PCI-E lanes.
Exactly, there are simply not enough PCIE lanes in any consumer CPUs nowadays to built cost effective workstations that can double as gaming rigs. If every creator needs a fat wallet and a rich mom to buy a Threadripper that can't be the answer. There aren't even enough slots that come out as x8 to support a good 40 or 100Gbit NIC or some extra SAS controllers or NVME expander cards. That's why so many people build extra machines based on older gen EPYCs or XEONs none of the later are great for gaming of course...
hsyy... old xeons 😊😊
they might be slow as a single. but as a group, (and quad {6th idk} channel memory) they can be up to date
Facts just give us 16 cores on one CCD, faster memory controller and greatly improved IO and we are all good. The architecture of Zen 5 is already the best.
23:12 Bro, Intel had 128MB of additional eDRAM on their 5th gen Core lineup - Broadwell architecture. It was called Core i7 - 5775C. It went unnoticed because it wasn't marketed well, and few weeks later intel released rather famous 6th generation (skylake) with DDR4 support. This i7 - 5775C was a beast for gaming, it was faster then successor 6th gen and in some games even faster than 7th gen! Especially after overclocking. So intel had it before AMD and they new what wonders it can give. It looks like intel was too confident with their new 12th gen architecture and thought additional memory inside CPU will not be necessary. But they got caught off guard by AMD with their x3D chips, and it was too late changing 12th, 13th and 14th gen architecture.
Intel had big caches definitely before AMD, however it was L4, not L3 and it was not stacked, but integrated.
But you are correct, it was a beast for certain type of workloads.
My sources have said slightly differently to this, firstly AMD might not be even making 8 core chiplets anymore, they will all be either 3nm 16-cores Zen 6 or 2nm 32-core Zen 6c but these are exclusive to the server market. Obviously 8-core SKUs will exist but these will simply be cut down chips for the lowest end SKUs as AMD wants to push on the core count and performance so Intel can't find their feet. IPC (or performance in general) for Zen 6 will actually be very decent, this is for several reasons, firstly the performance gains for Zen 5 didn't work properly because of the IMC issues so we will be getting Zen 5 + Zen 6 improvements. Secondly, having 16 cores all on the same chiplet will help performance of applications that are boosted by more than 8 cores but would normally be impacted by cross chiplet latency. The only caveat to all this performance is memory bandwidth, AMD did toy with the idea of having Zen 6 on both AM5 (dual DDR5) and a high end version on an early release of AM6 with DDR6, even quad channel was discussed to make sure 32c/64t CPUs were kept adequately fed. I'm not sure if they are still going ahead with this as of now or if they are pulling the launch of Zen 6 in because Zen 5 missed the mark so badly and so AM6 and DDR6 don't be ready in time.
On the Intel side, 100% Intel is aiming for 16p + 32e, they feel they have to to keep up, they need a win and they see multitasking as the easiest option right now. Plus this gives them a lot of room for SKU options which Intel always likes to have. I also hear that one of the main reasons Nova Lake isn't on LGA-1851 is because of the same memory bandwidth issues as AMD is facing, dual channel just doesn't cut it anymore, so high end Nova Lake will have quad channel controllers and lower core count SKUs will just have the cheaper dual channel controllers, I also believe both controllers are DDR5 and DDR6 compatibility so that Intel has cheaper low end DDR5 options and higher performance DDR6 options. This was also the reason for separating the memory controller back out of I/O die again so there doesn't need to be larger more expensive quad channel parts on the low end SKUs. Obviously motherboards will be sold in cheaper dual channel options or expensive quad channel variants. Again Nova Lake should see a really big increase in performance because it's a full redesign from the ground up with a lot of new ideas, plus Intel won't have moved the needle forward for about 6 years by that point!
I just hope both companies can deliver on their initial aims, time will tell I guess.
I remember when gigahertz meant something...
Why do you care about ghz?.. it means nothing
@@christophermullins7163 those days yeah in the past it was.
@@christophermullins7163 He literally said he remembers "when it meant something..."
Meaning right now, it doesn't.
@@corey2232 if you can tell a CPUs performance by its ghz.. that means you are seeing no IPC gains on newer architecture. That is a failure.. as it should be. Why would you want to be in a time where there are no IPC gains?
@@christophermullins7163 Because now you can't tell a CPU's performance either by its clock speeds or its core count. I don't care either way, but I'm assuming the OP meant it used to be easier to identify X CPU > Y CPU.
I am calling it
New io-die + npu that barely beats Intel and faster igpu.
I just want amd to make all zen 6 x3d chips. And stop the non-x3d chips.
Definitely the R9 X950X3D.
R9 X950X3DX for dual X3D chiplets, X950XXX for dual Zen 6C chiplets.
To be honest i would prefer a few extra pcie lanes over an increase in core count.
Yeah Zen 7 for the desktop would be interesting if it had a core count increase. Even from 16 to 24 although I would prefer 32 cores on the desktop even if it were a mix of Zen 7 standard and Zen 7 dense and utilizing 3DV cache. Sadly we just have to wait and see.
I think that general systems should really become more like consoles. Design a system of three levels of processing power. Run a CPU, GPU, and NPU. Optimise the absolute crap out of the OS and applications to use the platform. I would buy a PS5 Pro if it could run a parallel NVME with a Windows/Linux desktop natively and I could switch between "gaming" and "work" OSs. Look at people buying a 4090 for $1800, the PS5 Pro is going to be $700 I think. If there was a PS5 Ultra which cost $2000 I would think it would quite a beast, and then with all the optimisations it would be really good.
Why would they go for a 16c CCD and not a 12c one? It's a 50% increase and matches intel's core count.
AMD is not doing BigLittle so their complexes are optimized in 2ⁿ counts.
@@kazedcat except that there are already 12c CCXs
@vyor8837 That's a 16core with 4 cores disabled. It would be a waste if AMD builds a 16core CPU and only sells 12core.
@@kazedcat ... there are 12c CCXs for their Zen4c chips.
Make better memory controller and more V5 PCI lanes first!!!!
16 core per ccd would make latency slower in overall ccd since you need make bigger L2 cache to connect everything, it not worth it until software/gaming catch up
At 17:30 four cores in the background are system (SIP) management processors. At 20:30 u mean the E cores are finally really going to be used as a SIMD parallel array Intel has been working on the compiling tools since APO but this is much more than APO. At 23:15 right Intel technically true many first cache attempts (implementations) and Intel produced MCM for 30 years for industrial, military and aerospace applications. There were forebears that are cache designs Intel stole and others stole from DEC. AMD bought those DEC patents in 1996 that became Hyper-transport. DEC did most of the x-way associative write back cache development. Way back when at 24:00 1990s NexGen invented, well implemented, the first processor direct L2 that Intel borrowed at Pentium ll. Nx586 is the first L2 direct processor implementation and with IBM boards is my product marketing and platform validation responsibility before AMD K6. mb
No ZEN 6 nor Nova Lake until Q4 2026 is pretty bad news imo... It might be good for consumers because they'll keep their hardware longer, but on the other hand it means that CPU performance is not going to increase as much over the next 2 years and then we're going to have the RTX 60 and RDNA 5/6 GPUs that could be highly bottlenecked by CPUs! Let's hope that ZEN5 will at least get a Refresh with a better I/O Die !
I would love to see more cores on each CCD even if its just to 10 on each to see if "only need 8 core for gaming " is true LOL I do know that when I had the 7950x3d games loaded faster than 7800x3d but thinking of 32 cores on 2 CCD's with x3d on both OMG yes I dont care if I need it I just want it LOL
I wonder if a single CCD with 16 full cores calls for AIO cooling as standard. I've got a 9800X3D with an air cooler and it runs cool. I just wonder at the heat output though if 16 cores are within one CCD.
The problem with Adding more cores is the power each requires & heat each core puts out affect clocks and make cooling harder, cores can't clock as high as a result then add in size of chip affecting room on boards needed and you have to ask does average user with Windows, a few productive software uses and games need more than 16 cores at present mostly no but for those that do AMD have thread ripper. That said with latest improvement in efficiency AMD are positioning to up core count i feel not Zen 6 but with Zen7 that could be on new socket as a result, also as AMD kept multithreading they don't need as many cores to beat Intel most of the time who ditched it.
Do not need more cores. Do need HEDT with quad channel UDIMM and 48 PCIE lanes though.
We need a duel cpu motherboard 😊
No, with today's core counts there is no point
AMD should have a 16C CCD for Zen 6 with NPUs in the IOD... I don't think they will add more SMT but they can add more L1, L2 and other features...
AMD said that Zen5 is foundational so SMT should stay the same. Zen6 should be a redesign of the cache hierarchy. Since Zen5 X3D shows that the old cache system cannot feed the cores without significant increase in L3. Maybe they can finally make the dual fetch decode work for a single thread but that's a big challenge and I expect a halfway designed where dual fetch decode works on some single thread workload but not every workload.
I want a cpu with 1 million cores
If ZEN6 is really to be released in late 2026 or even early 2027, than this would mean, AMD has a room for easy ZEN5 refresh with new IO die. It's almost as if AMD planed to release ZEN5 with old IO die dragging down the performance - releasing infamous "ZEN5%" since new Intel Core Ultra is not a threat. A refresh in between with new IO die seems like a next easy (cash-grab) step.
Core Ultra not a threat? wtf are you smoking? Intel still destroys AMD in multitasking workloads.
@@iequalsnoob WTF you smoking? intel only destroying it's self.🤣
@@madb132 Intel is still struggling against Intel, yes, but they did legitimately beat not just their last gen parts but even the current gen Ryzen 16 cores in some multithreaded tests…. At a thread count disadvantage no less.
Zen is better suited to chiplet design. Since Ryzen 1000 was infact chiplet design and the whole infinity cache and memory optimization AMD is doing since 2013 i.e when development of ryzen started has paid off. They now know how to make a chiplet design work. They have done many iteration of Ryzen of 4 ccx, 2ccx design. Large server chips. They have covered their basics.
I wouldn't mind picking up arrow lake if Intel would confirm whether or not it will be supported for a while like up to Nova lake...
I can dream of Intel having 8+32 with big cache for pcores. If AMD could fit 16 pcores under one CCD with 3D cache it would be also tempting proposition. Looks like I am going to stuck to 5800x3D till one of those scenarios become real.
I am on 9800X3D. This caught between a rock and hard place and the desire for more than 8 homogenous cores on one die has made me swap parts a lot the last 1-2 years. I was willing to compromise and go Raptor Lake and give hybrid a chance, but Raptor Lake stability/degradation issues I had turned me off and returned or sold some chips I tried. They all gave e stability problems eventually within a month,
So settled on 9800X3D after Arrow Lake disaster which I sold my 7800X3D system for thinking Arrow aLake would be a better Raptor Lake and stable and have same latency. But its tile design atency crippled gaming performance badly so it was a no go.
8 cores and they are much faster and tunable it is for me now on 9800X3D.
Nah P cores are going up
I'd love to see a zen6 --800X3D with with a single ccd even with 10 or 12 cores (if not 16).
Amd norm 2 and a half years to launch a new generation of CPU?
I will curse amd and intel if they still stick with 16 cores for AMD and 8 P-cores for intel
You sound like Richmond from IT Crowd
Will it have 40% IPC increase like you said Zen 5 would have?
Next generation should be good for both companies. AMD has the best architecture. However they are limited by their IO, memory controller and not as much PCIE lanes as Arrow Lake.
My guess for Zen6:
Server Zen6 dense: 32 core CCD in N2
Server Zen6: 16 core CCD in N2
Desktop Zen6: 8 core CCD in N3
N2 is rumoured to be around 50% more expensive per wafer while only being 10-15% more dense so cost/transistor would be higher. It has better power consumption so it is orobably used for datacenter where this is a bigger factor.
On desktop many systems doi not need more than 8 cores so I see 2 possibilities:
a) They are just going 8 cores again (more likely)
b) They justbdesign a 16 core CCD for high end and rebrand Zen5 for everything below
Regarding Intel: If the dual CCD rumours are true Intel is dead for gaming as they will run into the same issues that AMD currently has with 7950X3D.
Forgot about AMD phenom ii x6
It sounds like we are reaching the current limit on modern transistor size/count performance improvements. I must be living under a rock as Ihad notheard Intel has dumped hyperthreading. Like, WHAT?
It’s disabled. Lion cove in xeons still uses HT
Arrow lake sorely need some optimization/fixes to be fully realized. There's been a handful of overclockers and such disabling all but 1 P-core and gotten some interesting results in games. I think by the time we get to CES in January and beyond there will need to be a lot of retesting done by reviewers. I'm also kinda curious if the 9950X3D will play better this time round vs 7950X3D.
"2 nm" is pronounced 2 nanometers , dammit !
AI MAX+
Man what is this world becoming lol
Hoping zen6 is good, planning to get 9800x3d and then sell it if zen6x3d is much better
There must be 3 minutes of info in this 30 minute video. That has to be some kind of record
Is this a Temu ad?
AMD Zen6 Ryzen 11950X has 16 cores and 32 threads and 64 MB of L3 cache and 16 MB of L2 cache and supports 6400 MHz DDR5 RAM instead of 6000 MHz. it will be 1 to 3 percent faster in games and 5 percent faster in multi threaded apps. Nova lake will have 8P cores, 16E cores and possibly 4 LPE cores. Novalake is 5 percent faster than Arrowlake in games so Novalake 385K is as fast as 14900KS in games and it will be 5 percent faster than 11950X in multithreaded apps. it runs at 5.6 GHz and supports DDR5 10000 MHz memory. it will work for 2 years until it degrades and breaks. it consumes 225 watts of power. 11800X3D will beat 385K in gaming by 20 percent. the reason AMD does not use 16 core CCDs for gaming CPUs is because dual channel DDR5 is extremely slow and can not feed 32 cores. quad channel DDR5 or dual channel DDR6 or dual channel QDR5 (Quad data rate) is required to feed 32 cores with data. AM6 socket will have 32 core flagships because it can handle more cores because it has better VRMs and power delivery system and will not bottle neck 32 cores because it will support DDR6. adding 32 cores to AM5 is useless because AM5 socket can only max out at 230 watts and can not handle feeding data to 32 5+ GHz cores. Novalake will not be able to beat Zen5 X3D let alone Zen6 X3D in gaming. dont forget that AMD DDR5 6400 MHz is equal to Intel DDR5 8400 MHz. AMDs memory timings are far tighter than Intels.
No 4 thread count cpus.
So no new CPUs next year? Q4 2026 for both intel and amd?
16core single CCD, 3d vcashe, a better I/o die, and a better igpu would be banger at $350 or lower
very interesting, intel have some challenges ahead indeed
144 MB of L4 cache is slower than 3D VCache because 3D VCache is L3 cache. L3 cache is faster than L4 cache. if Intel moves L3 cache to a separate die latencies will increase and performance will decrease.
Oh lol Intel 2 Tiles with 8 + 16. Will be same latency penalty as AMD on Zen 3 through 5 dual 8 core CCDs lol. Or worse as shown by Arrow Lake tile packaged design. Zen 5 X3D 9800X3D is where it is at for now for gamers. Games yet to use more than 8 cores except Cities Skylines 2 extremely large populations or some select few games that get marginal benefits from them, Dual CCD 9950X3D if both CCDs have the 96MB may be interesting as no hybrid. Yes still dual CCD latency, but at least both good cache and no.
Maybe Zen 6 even if they keep 8 core CCDs the IO die they will configure some ladder cache that will completely ax any latency between CCDs so crossing CCDs to talk to a core is just as fast as taking intra-CCD on Zen 6. One can hope!!!
A lot of comments below station that games/software is not optimized for more the 8 cores this will always be the case, but certain games do take advantage. For example "Eveonline" - I play this game using a 7950X, now most players switch on max graphics and play as a solo player in full screen just like they would play any other game but because the game has many different avenues of play then open up one way of playing using multiple accounts at the same time eg; when mining asteroids as a group of ships under one persons control. Now the games devs were thinking greedily at the time so as not to limit the game to just one person at a time (one player playing two accounts means double the monthly income and why limit it to 2 accounts why not make this a no limit option). Now yes this is extremely unusual to allow multiple games to run at the same time but this is made possible because you can play in windowed mode and I actually play 13 accounts at the same time and I am CPU/memory bandwidlth limited now and even find no advantage using the 7950X3D. If the 9950X3D comes with double 3DV cache then I will get one and try it. But higher core count for gamer's is probably of no use because the GPU is the bottleneck which is why I reduce the GPU load in Eveonline because for me the pretty graphics is not what is important but instead the amount of accounts that I can use at any one time. BTW many players do this (This is called multi-boxing) because to increase one's game monetary income means either play more time (only so many hours per day you can play) or increase the amount of accounts that you play at anyone time. Many solo miners don't like Multi-boxer miners but also many will ask for boosts from our main ships. Which is how using multiple accounts help us benefit. For people that have never heard of Eveonline, its a multi-faceted Space game with different game styles available for each individual from Mining asteroids, Pirating, Trading, Building or even just playing the markets like a stock broker. It has been in existence since 2003 and has evolved over time with especially with the graphics as PC tech has improved.
13 acc's damn lol. Dont think Ill join EVE myself. At this point you might as well have 2 PC's. Or a "plain old" threadripper.
@@jackthatmonkey8994 Threadripper is too expensive - new CPU, MB
I wish they did dual socket AM5 M/b's but 7950X's and are not dual socket compatible !
I feel like you answered yourself with your own example - dual-CCD SKUs like the 7950 amply cater for the niche use-case of being a 13-account multiboxer.
Yikes! I play X4 (a game where you can command fleets of hundreds of ships) in 4K and my Radeon 6950XT bottlenecks my Ryzen 3 3300x (quad-core) while listening to music on youtube in the background. Almost every game I play is gpu bottlenecked at 4k. I'll upgrade to a R5 9600X when the RTX 5090 releases. I think big core cpus are overrated for 4K gaming.
@@malcaniscsm5184 Hahahaha yeah but i wanna run more acc's !
Well Paul found a shirt.
Bro AM5 has already been delided we clearly know the size of the chip on the package... That said both AM5 and LGA1851 is a hard pass tor anyone with decent setup on AM4/LGA1700...
so?....are intel trying to save money on r&d by using design's for mobile cpu's in desktops???
or what is this about , cause i don't know anyone that are interested in e-cores for desktop pc's apart from intel themselves
shared L2 cache??? remember Bulldozer
Kehna kya chahte ho Sombhir Bhai😂
Clustered Multi Threading (CMT) is the exact opposite of Simultaneous Multi Threading (SMT). AMD experimented with CMT on AMD Bulldozer, Piledriver, Streamroller and Excavator microarchitectures and it was a total failure. if Intel uses CMT then AMD wins.
AMD is going to go to at least 32 cores.
ZEN 6 WILL HAVE A 32 CORE 64 THREAD 9950X ..... THIS WAS LEAKED A LONG WHILE AGO BY M.L.I.D.
the game that ported need 8 cores since the dev are too lazy to limit thread for 6 cores..
thank god, no more gym clothing...
so my 9800x3d is already dead?
five minutes before any of his usual bs starts
"Gamers only need 4 cores" - Intel. That did not age so well huh?
Bulldozer had 8 cores? That's perhaps more legit than Intel saying they have 20 or so. Intel markets a high core count, but they only disclose in small print how most of the cores are P cores - P for puny. A case for class action perhaps: 'You thought you were getting ~20 cores and then found your new CPU only has 8 proper cores and the rest were puny cores? Better Call Saul!'. 🤔
The 4 core intel chips absolutely shat on Bulldozer for gaming. come on fanboy use your brain
Does Zen 6 need more freds? Comment below.
More cores? No one will care in gaming.
My Factorio save cares 😂
That's what they said when we had 2 or 4 cores lol
@@FinneousPJ1 I'm a dev too, gimme moar cores! :D
@@FinneousPJ1 And still many games cannot really use more than 6-8 cores. There are definitely games out there which can use more cores, e.g. Factorio, but those are not the majority.
In the end it's always: think first, buy afterwards.
Get a hairbrush
Already talking about next gen products 🥱
To save on cringe value, never let geeks name anything...