This is the one thing in favor of the 7950-x3d. Though I have seen indications that idle power is still higher than Intel. Of course, if your PC is idle all the time, ,maybe you should turn it off. LOLOne thing that Steve failed to test for is power consumption. GN shows that the AMD Ryzen 9 7950X3D is about twice as power efficient as the i9-13900K, it's quite remarkable. Total power consumption is about half as much as the i9-13900K. Now that is a CPU I could get interested in. I will be waiting for the 7800X3D to see how that CPU performs.
According to Wendel if your application needs more than 1 ccd or 8 cores it does scale and wake up those coresThis is the one thing in favor of the 7950-x3d. Though I have seen indications that idle power is still higher than Intel. Of course, if your PC is idle all the time, ,maybe you should turn it off. LOL
To me, the issue is that you really have 2 CPUs on one die here. One with 3d V-cache and one without. I haven't seen any specifics as to whether this CPU will use both dies if the workload calls for it. Right now it seems like it uses one for games and the other for non-gaming workloads. In that regard it seems a disappointment. I like to use my PC for everything I need to do, gaming being one workload type. If I were building just a gaming machine, I wouldn't look at this CPU, I'd wait for the 7800-x3d. If I wanted a good gaming machine with good productivity performance, I'd look at the 13900K as it's $120 cheaper.
It prefers to use one chiplet for gaming and another for other workloads. Of course you can disable that behaviour if you want and then it just works as 7950X except that other chiplet has much faster cache.To me, the issue is that you really have 2 CPUs on one die here. One with 3d V-cache and one without. I haven't seen any specifics as to whether this CPU will use both dies if the workload calls for it. Right now it seems like it uses one for games and the other for non-gaming workloads. In that regard it seems a disappointment. I like to use my PC for everything I need to do, gaming being one workload type. If I were building just a gaming machine, I wouldn't look at this CPU, I'd wait for the 7800-x3d. If I wanted a good gaming machine with good productivity performance, I'd look at the 13900K as it's $120 cheaper.
Not sure what you're basing that conclusion on. According to Tom's Hardware about the only productivity workload that favors the 7950-3dx is Adobe Premiere Pro. Per the articleIt prefers to use one chiplet for gaming and another for other workloads. Of course you can disable that behaviour if you want and then it just works as 7950X except that other chiplet has much faster cache.
13900K is pretty bad for productivity tasks as it lacks AVX-512 support and power consumption is much bigger. Remember that Intel itself promoted AVX-512 for productivity so even Intel considers 13900K to suck on productivity workloads.
Problem here is that single software benchmark is very far from productivity. Basically when doing productivity you never have just one software running same time. And when running more than one software at one time stress for cache and/or memory will get higher. The becomes problem: E-cores suck when it comes to cache and/or memory access. They probably can run one software at a time quite well but putting more software running at same time E cores will be much more in trouble than P cores.Not sure what you're basing that conclusion on. According to Tom's Hardware about the only productivity workload that favors the 7950-3dx is Adobe Premiere Pro. Per the article
"The Core i9-13900K is 17% faster than the standard Ryzen 9 7950X3D configuration in single-threaded work and 4% faster in threaded applications. We have two overclocked configs here for the 7950X3D — the auto-overclocked PBO, and PBO combined with undervolting (PBO UV). Notably, the PBO UV config narrows the 13900K’s lead in single-threaded work to 11% and reduces the disparity in threaded work to a mere 1%. That’s incredible, given the 7950X3D’s much lower power consumption level (you can see the 7950X3D’s incredible power efficiency metrics on the power page). Regardless of whether or not the chip is undervolted, the 7950X3D consumes far less power and will generate much less heat than the 13900K, but that doesn’t come with a huge performance loss."
Power is definitely a win for AMD, which I stated above.
I believe this is what multi-core benchmarks are for. They simulate apps using multiple cores, which would happen in a multi-app scenario. The 13900 handily beat the x3d in Cinebench. That is some indication that it would fair better in multi-app workloads.Problem here is that single software benchmark is very far from productivity. Basically when doing productivity you never have just one software running same time. And when running more than one software at one time stress for cache and/or memory will get higher. The becomes problem: E-cores suck when it comes to cache and/or memory access. They probably can run one software at a time quite well but putting more software running at same time E cores will be much more in trouble than P cores.
I think that really depends on the workload. Consider this, if x3d cache was universally good for productivity or non-gaming workloads why didn't AMD just include that in all CPUs? x3d has shown to be very favorable to gaming applications and that is the focus for it. Also consider that clock speeds are lower on the chiplet with v-cache. Will more v-cache overcome slower clock speeds? That doesn't seem to be the case for all workloads, hence why most benchmarks focus on gaming.I also call BS that "3D cache does not help with productivity". When I get 7950X3D I will definitely try that. I have no doubt that running more than one cache sensitive software on chiplet that has extra cache will help a lot.
Not sure what point you're making here. You compare SATA SSD to NVMe and then bring in Ram disk? Apples and oranges.Also this "NVMe drives aren't faster than SATA SSDs because benchmarks tell so". Let's just calculate: copy 50GB data into RAMdisk. 500MB/s (SATA) that takes 100 seconds. 5GB/s NVMe SSD that takes 10 seconds. That's tenfold improvement. For what? Simple task like install new virtual machine for example. RAMdisk has seek times hundreds times lower than NVMe SSD.
There's no reason you couldn't run multiple apps simultaneously in a benchmark and it can be repeatable. Benchmarks perform a series of actions or commands in a consistent fashion. You could, for example, kick off several "benchmarks" if you wanted to simulate multi-app workloads. I assume that there's a reason no one is doing that, much like using 1080P in CPU/Gaming benchmarks.You won't see any kind of those in benchmarks since benchmarks must be repeatable. You cannot put gazillion software running background and get consistent results. So they run one software at a time and I cannot blame for that because that's only way to do it. Still, I don't care what benchmarks say. I can guarantee that for my real life productivity usage 7950X3D is much faster than on 7950X and there is no need to even consider Intel CPU with crap cores.
There might be big differences on memory/cache usage, storage load etc. Since we know E-cores suck on cache/memory intensive tasks, it's far from that simple.I believe this is what multi-core benchmarks are for. They simulate apps using multiple cores, which would happen in a multi-app scenario. The 13900 handily beat the x3d in Cinebench. That is some indication that it would fair better in multi-app workloads.
Of course it depends on workloads. Point is that multi-app workload is much different than multi-core workload. Since multi-app workload is very hard to simulate, there are also very few benchmarks about it. AMD is not stupid, unless multi-core benchmarks show benefits of 3D cache, AMD surely won't put it everywhere.I think that really depends on the workload. Consider this, if x3d cache was universally good for productivity or non-gaming workloads why didn't AMD just include that in all CPUs? x3d has shown to be very favorable to gaming applications and that is the focus for it. Also consider that clock speeds are lower on the chiplet with v-cache. Will more v-cache overcome slower clock speeds? That doesn't seem to be the case for all workloads, hence why most benchmarks focus on gaming.
According to most benchmarks, SATA SSDs are about as fast as NVMe drives and RAMdisks are useless. And then there is "NVMe is useless, PCIe 5.0 NVMe drives are useless pick 3.0 instead" shouts all over internet. Also RAMdisk is useless because operating systems have good memory prefetch etc etc. Benchmarks...Not sure what point you're making here. You compare SATA SSD to NVMe and then bring in Ram disk? Apples and oranges.
You could but that will probably create more problems. Common benchmark practice is to discard results that are too far away from what they should be. Multiple software same time will probably cause more that kind of "strange" results. That's probably main reason for it.There's no reason you couldn't run multiple apps simultaneously in a benchmark and it can be repeatable. Benchmarks perform a series of actions or commands in a consistent fashion. You could, for example, kick off several "benchmarks" if you wanted to simulate multi-app workloads. I assume that there's a reason no one is doing that, much like using 1080P in CPU/Gaming benchmarks.
Go look at Gamers Nexus' review of the 7950X3D you will see that on non-gaming benchmarks the CPU runs just a bit slower than the 7950X. So you just have to decide, what is best for you. Faster gaming or better productivity, the option is yours to make.This is the one thing in favor of the 7950-x3d. Though I have seen indications that idle power is still higher than Intel. Of course, if your PC is idle all the time, ,maybe you should turn it off. LOL
To me, the issue is that you really have 2 CPUs on one die here. One with 3d V-cache and one without. I haven't seen any specifics as to whether this CPU will use both dies if the workload calls for it. Right now it seems like it uses one for games and the other for non-gaming workloads. In that regard it seems a disappointment. I like to use my PC for everything I need to do, gaming being one workload type. If I were building just a gaming machine, I wouldn't look at this CPU, I'd wait for the 7800-x3d. If I wanted a good gaming machine with good productivity performance, I'd look at the 13900K as it's $120 cheaper.
I've seen that video. And, I understand why there's a difference due to the slower clock speeds of the x3D version of the CPU. But, I don't necessarily believe it's a one or the other choice. What I'm not seeing is a driving reason to buy the 7950x3d. It's good in some games but not as good in others. It's OK in productivity but not excessively better. The main attraction is power draw and (maybe) generational CPU upgrades on the same mobo platform. Get the price down $100 and the case becomes more compelling, but even then, if you're only looking for gaming performance the 7800x3d may be the better choice.Go look at Gamers Nexus' review of the 7950X3D you will see that on non-gaming benchmarks the CPU runs just a bit slower than the 7950X. So you just have to decide, what is best for you. Faster gaming or better productivity, the option is yours to make.
Benchmarking is a guide, not an absolute. There are inherent issues with most benchmarks, hence why most published benchmarks have a caveat of your mileage may vary.There might be big differences on memory/cache usage, storage load etc. Since we know E-cores suck on cache/memory intensive tasks, it's far from that simple.
Of course it depends on workloads. Point is that multi-app workload is much different than multi-core workload. Since multi-app workload is very hard to simulate, there are also very few benchmarks about it. AMD is not stupid, unless multi-core benchmarks show benefits of 3D cache, AMD surely won't put it everywhere.
Point is that multi-app workloads are productivity as well, it's not restricted to single multi-core app.
I've never seen benchmarks comparing ramdisk to SSD/NVMe or heard that NVMe is useless. I think everyone understands that ramdisk has limited value given that it's volatile storage. Used in the right place, it's great, but it's not general purpose storage.According to most benchmarks, SATA SSDs are about as fast as NVMe drives and RAMdisks are useless. And then there is "NVMe is useless, PCIe 5.0 NVMe drives are useless pick 3.0 instead" shouts all over internet. Also RAMdisk is useless because operating systems have good memory prefetch etc etc. Benchmarks...
I don't agree with this. I have 40 years in computer experience. I've sold and designed high-end, enterprise systems for major corporations. Benchmarking gives you a comparison point that you could otherwise not determine just by knowing specs.To put it short: More knowledge you have, less you need benchmarks.
Actually, there are benchmarks that address launch time for apps. Passmark has the App Timer which does exactly this.You could but that will probably create more problems. Common benchmark practice is to discard results that are too far away from what they should be. Multiple software same time will probably cause more that kind of "strange" results. That's probably main reason for it.
Another things benchmarks do not say: how usable is computer when software is running, how long it takes to launch software etc etc. Benchmark script is not interested about either of those things.
I'm interested IRL. That is very evident when looking early dual core benchmarks (around 2003). If we believe benchmarks, single core CPU was better in most cases. IRL, did anyone want to go back to "faster" single core after trying dual core?
Exactly, but some people take benchmark results as absolute facts and that has created problems quite long time.Benchmarking is a guide, not an absolute. There are inherent issues with most benchmarks, hence why most published benchmarks have a caveat of your mileage may vary.
Really? Web is full of comments like SATA SSD is enough or (insert here PCIe version) NVMe is overkill/useless etc.I've never seen benchmarks comparing ramdisk to SSD/NVMe or heard that NVMe is useless. I think everyone understands that ramdisk has limited value given that it's volatile storage. Used in the right place, it's great, but it's not general purpose storage.
Point is: if you know everything, and I mean everything, about system both hardware and software, then you really don't need benchmarks for anything. You can determine outcome without testing it. Problem of course is that it's almost impossible to know everything. But still, more you know, more you can determine about speed without benchmarks. It also makes much easier to spot fake benchmarks like this one without actual testing https://wccftech.com/amd-3d-v-cache...-ryzen-9-7950x3d-by-over-4x-versus-7950x-cpu/I don't agree with this. I have 40 years in computer experience. I've sold and designed high-end, enterprise systems for major corporations. Benchmarking gives you a comparison point that you could otherwise not determine just by knowing specs.
Of course there are but those are Very rarely tested with multiple software. Sometimes tested with single software. Additionally background processes are usually eliminated, something that rarely happens IRL.Actually, there are benchmarks that address launch time for apps. Passmark has the App Timer which does exactly this.
Basically most software were single threaded but still dual core made usage much nicer since single software does not make computer unresponsive. Benchmarks missed that.I can't speak to those benchmarks; however, I would speculate that early multi-core benchmarking suffered from lack of applications that could use multi-core CPUs or poor drivers that couldn't schedule apps and cores properly. Look at early gaming benchmarks that suggested you only needed a 4 core CPU. That's because games couldn't take advantage of more cores, even though there were plenty of higher core count CPUs available.