AMD Ryzen 9 7950X Allegedly 40% Faster than 5950X in CPU-Z Bench Multi-Threaded

[Daven]

Well it is a 12900K with higher clocks, more cache and architectural tweaks.

AMD is going to suffer on marketing, becuase they are releasing the same processors just in Zen 4 form, but Intel has increased core counts effectively pushing everything up a tier. 7600X will be expected to compete against 13600K but now is a lower end cpu same with 7700X vs 13700K. And Intel is going to keep increasing e-core counts with Arrow Lake hitting 40 IIRC. AMD does not appear to have an answer to this unless they also go hybrid in Zen 5 and use Bergamo cores as e-cores. AMD luckily does have v-cache coming though.

While I have no real interest in Raptor Lake, and will repalce my old Zen 1700X system with a 7900X (hopefully with v-cache) come Arrow Lake I might be a lot more inclined to go that way if AMD keeps to current core counts. No way a regular 16 core 8950X would compete with a 32 core (8P + 24 E) 14900K say. There are rumours AMD is going hybrid with Zen 5 and IMO they have no other option. I don't having 32 full cores is wise in the desktop market especially from power use terms.

AMD can also add 2 more chiplets for a total of 32 cores and increase TDP up to 300W.

I think they should do big, medium, little.
4 big superfast cores, 4 medium duty all round cores, 8 little slow cores.
Then have a robust ability to pin software threads to particular core types.

I really like your idea:
BIG = Zen 5
MEDium = best of Zen 3/3+/4
little = Zen 2

 

[zlobby]

Hi,
I'd rather have real cores and threads than thermal defective e cores any day.

This guy gets it! Also, no one wants toothpaste TIM.

That's not how silicon works. Different architecture works in its own frequencies and voltage to frequency curve. Nothing here is "defective".
Silicon isn't run on X frequency universally. E cores and P cores are fabricated entirely differently. They are not the same.

In case you never bothered knowing how the CPU cores look in the SOC before calling them defective, this is ADL-S, with the 8 P cores and 8 E-cores to the right.

These "toy cores" and "defective cores" are built to be very silicon space efficient, and carry an IPC similar to what people met on Skylake \ Zen2

I agree with his sentiment, though. Although not defective per se, these gimped space wasters are being sold at the same price as their proper (hence the 'P' moniker) cores.
The 'E' here should stay for 'e-waste'.

 

[InVasMani]

I just hope we get fair analysis with boards of the same VRM quality and same memory standard between companies. We don't need variables like oops the Intel board has a additional 4 VRM phases to muddy the waters.

 

[zlobby]

Sadly

I just hope we get fair analysis with boards of the same VRM quality and same memory standard between companies. We don't need variables like oops the Intel board has a additional 4 VRM phases to muddy the waters.

Sadly, that's often the case. I wonder who's to blame? /r

 

[AM4isGOD]

Hi,
Think the real term is thermal defective cores
If e cores can't do the same speed as p cores at the same temperature then they are defects which at one time were binned out.

Why would they even be able to do the same clocks when the P cores are completely different? you don't know what you're on about

That's not how silicon works. Different architecture works in its own frequencies and voltage to frequency curve. Nothing here is "defective".
Silicon isn't run on X frequency universally. E cores and P cores are fabricated entirely differently. They are not the same.

In case you never bothered knowing how the CPU cores look in the SOC before calling them defective, this is ADL-S, with the 8 P cores and 8 E-cores to the right.

These "toy cores" and "defective cores" are built to be very silicon space efficient, and carry an IPC similar to what people met on Skylake \ Zen2

Don't waste your time trying to explaining to dumb....sorry deaf ears.

 

[Daven]

I really don’t know where the idea came from that e-cores are thermally defective anything. They are fabbed within the same monolithic die as the rest of the chip. They are a completely different architecture than the p-cores as countless commenters have shown with die shots.

I mean is Intel making a monolithic Alder Lake chip, testing all the parts and then throwing out only the chips with the highest quality e-cores while only selling chips with defective e-cores. That makes no sense.

 

[AM4isGOD]

I really don’t know where the idea came from that e-cores are thermally defective anything. They are fabbed within the same monolithic die as the rest of the chip. They are a completely different architecture than the p-cores as countless commenters have shown with die shots.

I mean is Intel making a monolithic Alder Lake chip, testing all the parts and then throwing out only the chips with the highest quality e-cores while only selling chips with defective e-cores. That makes no sense.

usual anti intel crap on TPU. Buy AMD and join the bashing, or ignore it and move on.

 

[dj-electric]

I agree with his sentiment, though. Although not defective per se, these gimped space wasters are being sold at the same price as their proper (hence the 'P' moniker) cores.
The 'E' here should stay for 'e-waste'.

Why fault the E cores as gimped and not give credit to P cores as buffed? This coin has two sides. Once again, I'd send everyone bashing this kind of CPU architecture to look at its results in compute and value. While not the fastest thing in any kind of workload, Intel has proven that its hybrid design mostly get its goals done. Otherwise ADL-S wouldn't get the relative success it did get.

If you want to bash stuff, bash Intel's aggressive frequency curve sending its CPUs to consume dozens of additional watts in the name of 3-5% added performance (igor's lab and HU have articles about how 12900K performs at 105W).

There's something quite juvenile in the common act of armchair engineering online, and frankly has been for 20 odd years now. The double standards are very clear, and people who spent a lot of time in technology circles are willingly ignoring how things work to serve some none important nerrative they have. These are rich companies, drop off your swords. none of them need your help. Criticize where important.

A few months prior to ADL-S release yours truly confronted the very minds behind it asking how "the hell" are they planning fighting monsters like the 5950X with 8 big cores and 8 small ones. I was told to wait and see, being pessimistic about any kind of remote return to market relativity. This time, it just seemed like most of my worries were shut off, possibly besides general power consumption which im very much not a fan of with ADL-S

 

[InVasMani]

Sadly

Sadly, that's often the case. I wonder who's to blame? /r

I mean do you mean in testing or board designs. There is fault to go around I'd say. The testers don't always have the best control over the situation in whole unfortunately. I'd say that's especially true when their supplied boards by the AIB's and it just paints a bit of a skewed and lopsided scenario in favor of the newer emerging hardware.

 

[Divide Overflow]

So great to see such strong competition! Consumers get to reap the benefits.

 

[zlobby]

Why fault the E cores as gimped and not give credit to P cores as buffed? This coin has two sides. Once again, I'd send everyone bashing this kind of CPU architecture to look at its results in compute and value. While not the fastest thing in any kind of workload, Intel has proven that its hybrid design mostly get its goals done. Otherwise ADL-S wouldn't get the relative success it did get.

If you want to bash stuff, bash Intel's aggressive frequency curve sending its CPUs to consume dozens of additional watts in the name of 3-5% added performance (igor's lab and HU have articles about how 12900K performs at 105W).

There's something quite juvenile in the common act of armchair engineering online, and frankly has been for 20 odd years now. The double standards are very clear, and people who spent a lot of time in technology circles are willingly ignoring how things work to serve some none important nerrative they have. These are rich companies, drop off your swords. none of them need your help. Criticize where important.

A few months prior to ADL-S release yours truly confronted the very minds behind it asking how "the hell" are they planning fighting monsters like the 5950X with 8 big cores and 8 small ones. I was told to wait and see, being pessimistic about any kind of remote return to market relativity. This time, it just seemed like most of my worries were shut off, possibly besides general power consumption which im very much not a fan of with ADL-S

I'll bash what I feel like to, thank you very much. All I'm supposedly 'ignoring' has been factored in, I promise.

Bottom line is I don' shill or intentionally troll. Everyone is still free to buy whatever they feel like it. No harm, no foul?

I mean do you mean in testing or board designs. There is fault to go around I'd say. The testers don't always have the best control over the situation in whole unfortunately. I'd say that's especially true when their supplied boards by the AIB's and it just paints a bit of a skewed and lopsided scenario in favor of the newer emerging hardware.

While mobo producers sort of caught up in terms of design parity for both AMD and intel, OEM like Dell, HP, Lenovo, etc. still don't pay the same ammoumt of attention to both chip makers.
You can get intel designs with rich feature sets, beefy cooling, etc. while AMD designs gets the leftovers.That's mostly my issue.

With testers getting their hands on yet unreleased hardware it's common sense - they will have to make do with what they have at hand at the time. Hardly any choice that early before mass launch.

 

[Garrus]

Did you really read the article? In the article you showed, "12900K E-core", i.e. 8xE-core, were compared.

And have you never seen the Alder lake die photos, Intel would have distributed a lot of them in PR. Including cache and corresponding ring bus area, each P-core is about 10.7 mm² and each E-core is about 13.4 mm² in a cluster of four. 24 E-cores are placed in the same area of 8 P-cores. Intel itself treats a cluster of four E-cores as if they were a single P-core. Have you ever seen an illustration like this? (I have directly taken these from Intel Newsroom Press Kit: 12th Gen Intel Core)

View attachment 259568View attachment 259569

In the 12900K, comparison between 4xE cores and 1xP core become:
- 25% more area
- Equal power consumption
- 50% more MT performance
The relationship is 1.2x area efficiency and 1.5x power efficiency.

The entire point of the article is that they are comparing 4 Skylake Cores against 8 E cores, NOT AGAINST 4. It never ceases to amaze me, the power of marketing.

First of all, notice Intel's trick in talking about Skylake at all. The modern CPU is an Alder Lake CPU. What we need to know is how do E cores compare against Alder Lake P cores. Bringing up Skylake in the first place was intentional, to mislead people and get them confused.

Secondly, you know E cores are half as powerful as you suggest. Why? Because the 6 P cores in the 12600k are slightly faster than 6 Ryzen cores, and if the 4 E cores were like 4 Zen2 cores, then how come the 6+4 ends up about the same speed as the 8 Ryzen cores in the 5700X? Voila, proof right there, 4 E cores gives you 2 core boost, about. You only see Skylake style IPC with one core comparisons. Why? Because E cores are nerfed as I mentioned before and don't work together well. One of the reasons people get confused is because P cores are so strong at Cinebench, people imagine the 12900k's strength comes from the 8 E cores. No. Out of 4 CPU tests, AMD actually wins 3 with the 5950X. I can't stand the overuse of Cinebench. The 5950X is still faster with Octane, Blender, and Compiling benchmarks. Loses in Cinebench. AMD is still ahead today 5950X versus 12900k. Not at Cinebench that's all.

I don't want any E cores. It's cheap for Intel, good for marketing. It is the same cellphone nonsense coming to desktop. Apple uses 2 P cores, I don't care if Android wants to use 8 horrible ones. Apple added E cores because they needed to for marketing. Also the nonsense about thread direction ticks me off. In no way can a computer use E cores to spread the load and save the P cores. That's not how CPUs work, at all. And the fact that it is a total lie was shown 100 percent when Alder Lake launched and every ADL laptop is at the bottom in battery life. OOPS. E cores don't give you battery life, which should be obvious, unless the P cores are off, which would make your laptop slow and terrible. IF you have the P core running for responsiveness as needed, the E cores are always idle anyways in low usage situations, and using the E cores makes no sense, they are not efficient, you can just run a P core at lower voltages and frequency and rush to complete faster.

NO E CORES. Thank you. Useless nonsense to save Intel money and create good marketing. There was nothing stopping CPU manufacturers from making strong + weak cores for years, until they figured out it helps for marketing. I'll take 8 P cores over 6+4 every time.

 

[AM4isGOD]

All this arguing just to play a fucking game faster, sheesh what is happening to humanity. No wonder this planet is so fucked up.

 

[Nanochip]

Zen 2 or Skylake, total nonsense. You bought the marketing. Anandtech tested E core performance exhaustively. The 12900K has 8 E cores and it often loses to 4 Skylake cores. So half the speed of Skylake or Zen 2 is more like it.

"Having a full eight E-cores compared to Skylake's 4C/8T arrangement helps in a lot of scenarios that are compute limited. When we move to more memory limited environments, or with cross-talk, then the E-cores are a bit more limited due to the cache structure and the long core-to-core latencies."

The issue is one single E core can appear to equal a Skylake core, all by itself. But when you run 8 of them together, the core performance doesn't scale right because of their limited design. Look at the MT tests and you'll get almost half the performance as what you expected.

E cores are small for a reason. Lacking cache, high latency, lacking hyper threading, don't expect much from 16 of them. CPUs are more than the raw compute core. I don't want any E cores at all, thank you.

CPU Benchmark Performance: E-Core - The Intel 12th Gen Core i9-12900K Review: Hybrid Performance Brings Hybrid Complexity (anandtech.com)

But the E-cores also beat Skylake in many tasks according to the same article. But the key point and consideration that seems to escape you is: the power consumption of the 8 E-cores vs the 4 Skylake cores... for less power and die area, you get roughly the same performance. As intel upgrades its E-core architectures with crestmont, skymont, and darkmont, they'll maximize performance per watt per die area. Which is a good thing to do. If you were to take the 8 Alder Lake P-cores and add 4 Skylake Cores (rather than 8 Gracemont E-Cores), power consumption would go through the roof.

 

[Chrispy_]

That's good news for Zen4 but how much of that is down to power limit changes? With Zen4 defaulting to 170W (230W boost)

The 5950X is a beast of a processor if you unchain it from its "105W" TDP that has a boost limit of 142W. Overclocked and permitted to draw 230W from the socket, it's going to make up some of that 40% the 7950X is claiming - possibly more than half!

 

[dj-electric]

That's good news for Zen4 but how much of that is down to power limit changes? With Zen4 defaulting to 170W (230W boost)

Generally, the trend is to use those higher TDP numbers in order to reach particularly aggressive frequencies. Since we know Zen4 will do whatever it can to maximize its silicon potential, I wouldn't be surprised if we are going to see post 200W numbers on some types of loads.

I do expect gaming power to be mild, Zen3 is particularly efficient during gaming.

 

[InVasMani]

The entire point of the article is that they are comparing 4 Skylake Cores against 8 E cores, NOT AGAINST 4. It never ceases to amaze me, the power of marketing.

First of all, notice Intel's trick in talking about Skylake at all. The modern CPU is an Alder Lake CPU. What we need to know is how do E cores compare against Alder Lake P cores. Bringing up Skylake in the first place was intentional, to mislead people and get them confused.

Secondly, you know E cores are half as powerful as you suggest. Why? Because the 6 P cores in the 12600k are slightly faster than 6 Ryzen cores, and if the 4 E cores were like 4 Zen2 cores, then how come the 6+4 ends up about the same speed as the 8 Ryzen cores in the 5700X? Voila, proof right there, 4 E cores gives you 2 core boost, about. You only see Skylake style IPC with one core comparisons. Why? Because E cores are nerfed as I mentioned before and don't work together well. One of the reasons people get confused is because P cores are so strong at Cinebench, people imagine the 12900k's strength comes from the 8 E cores. No. Out of 4 CPU tests, AMD actually wins 3 with the 5950X. I can't stand the overuse of Cinebench. The 5950X is still faster with Octane, Blender, and Compiling benchmarks. Loses in Cinebench. AMD is still ahead today 5950X versus 12900k. Not at Cinebench that's all.

I don't want any E cores. It's cheap for Intel, good for marketing. It is the same cellphone nonsense coming to desktop. Apple uses 2 P cores, I don't care if Android wants to use 8 horrible ones. Apple added E cores because they needed to for marketing. Also the nonsense about thread direction ticks me off. In no way can a computer use E cores to spread the load and save the P cores. That's not how CPUs work, at all. And the fact that it is a total lie was shown 100 percent when Alder Lake launched and every ADL laptop is at the bottom in battery life. OOPS. E cores don't give you battery life, which should be obvious, unless the P cores are off, which would make your laptop slow and terrible. IF you have the P core running for responsiveness as needed, the E cores are always idle anyways in low usage situations, and using the E cores makes no sense, they are not efficient, you can just run a P core at lower voltages and frequency and rush to complete faster.

NO E CORES. Thank you. Useless nonsense to save Intel money and create good marketing. There was nothing stopping CPU manufacturers from making strong + weak cores for years, until they figured out it helps for marketing. I'll take 8 P cores over 6+4 every time.

The E-cores are a stop gap solution to higher MT performance relative to AMD Ryzen. It's a little bit of a weaker pitch today relative to when Ryzen launched in terms of ST vs MT selling points due to the fact that MT has already caught up a great deal so doubling down on it further versus shifting towards balancing out and increasing ST performance is probably going to in general be of higher practicality to typical consumer usage like games and lighter weight multitasking. If true hard core MT you're not even looking at consumer chips in the first place over TR Pro and Epyc that utterly dominate high end MT work load scenario's with a lot more PCIE lanes to leverage as well.

The E-cores are still better than you make them out to relative to P-cores when you factor in the die area involved between each and overall performance uplift, but with the power draw increase that comes along with it the additional peak MT and slight loss to ST and IPC driven MT isn't as ideal as P-cores from a core to core basis however on a chip die to die one one chip die represents 4 cores though performance is more like 2 cores either way that's obviously a step forward relative to silicone area. It's obvious how Intel will aim to market them pretty much the same as AMD did at Ryzen launch, but as explained above it's a much weaker case argument today to make when current chips already sufficiently can handle gaming and a good bit of background task duties.

I really see a bit of role reversal about to take place between Intel and AMD on CPU positioning in the consumer segment from what I can see from both camps thus far direction wise. The jury is still out on just how X3D will also further change the landscape of things to be honest I see that primarily just extending the gap between the two in terms of where Zen 4 will be at it's best. It's not mention to be ThreadRipper Pro or Epyc replacement with extreme MT and 16 cores should be more than enough for most consumer case scenario's so if you need more you're better off looking towards those other platforms squarely aimed more heavily at such usage better suited towards it.

Intel will position the E-cores as bit of a price advantage for additional peak MT at the expense of power draw and other relative disadvantages is how I see it. It's a tough sell for their marketing department in 2022 if you're not a staunch Intel fan if you ask me. How well each perform and/or the efficiency they do so for what types of scenario's at given price points will probably sway more consumers.

 

[zlobby]

The E-cores are a stop gap solution to higher MT performance relative to AMD Ryzen. It's a little bit of a weaker pitch today relative to when Ryzen launched in terms of ST vs MT selling points due to the fact that MT has already caught up a great deal so doubling down on it further versus shifting towards balancing out and increasing ST performance is probably going to in general be of higher practicality to typical consumer usage like games and lighter weight multitasking. If true hard core MT you're not even looking at consumer chips in the first place over TR Pro and Epyc that utterly dominate high end MT work load scenario's with a lot more PCIE lanes to leverage as well.

The E-cores are still better than you make them out to relative to P-cores when you factor in the die area involved between each and overall performance uplift, but with the power draw increase that comes along with it the additional peak MT and slight loss to ST and IPC driven MT isn't as ideal as P-cores from a core to core basis however on a chip die to die one one chip die represents 4 cores though performance is more like 2 cores either way that's obviously a step forward relative to silicone area. It's obvious how Intel will aim to market them pretty much the same as AMD did at Ryzen launch, but as explained above it's a much weaker case argument today to make when current chips already sufficiently can handle gaming and a good bit of background task duties.

I really see a bit of role reversal about to take place between Intel and AMD on CPU positioning in the consumer segment from what I can see from both camps thus far direction wise. The jury is still out on just how X3D will also further change the landscape of things to be honest I see that primarily just extending the gap between the two in terms of where Zen 4 will be at it's best. It's not mention to be ThreadRipper Pro or Epyc replacement with extreme MT and 16 cores should be more than enough for most consumer case scenario's so if you need more you're better off looking towards those other platforms squarely aimed more heavily at such usage better suited towards it.

Intel will position the E-cores as bit of a price advantage for additional peak MT at the expense of power draw and other relative disadvantages is how I see it. It's a tough sell for their marketing department in 2022 if you're not a staunch Intel fan if you ask me. How well each perform and/or the efficiency they do so for what types of scenario's at given price points will probably sway more consumers.

Landscape will remain ever-changing. CPU/GPU/NPU/SoC/whatever-dafuq-not-PU will overgo constant metamorphosis with no solution trully fitting all.

Quantum computing and AI may help us build monstrous designs where memory and compute units are one, sort of what HBM, X3D are starting to look like.

 

[jesdals]

I am looking forward to see Infinity settings and memory control on these new cpu/boards

I wonder if there will be the same 1:1 between memory and cpu or we will se DDR5 speeds takeing it to 1:2

 

[Butrus]

I think they should do big, medium, little.
4 big superfast cores, 4 medium duty all round cores, 8 little slow cores.
Then have a robust ability to pin software threads to particular core types.

They won't need to do that.
AMD will have PAIRS of "BIG+little" cores sharing the L2 cache. Which means that migrating threads from a big core to it's little counterpart (or vice versa) will be quite fast.

 

[mclaren85]

You're not kidding, It feels great, been waiting for this moment since FX. Finally the two titans are in a proper arms race again, AMD pushing the hi-cache chiplet route and Intel hitting back with their split P+E architecture. Exciting times.

People need to take a step back from their brand loyalty for a second and just appreciate the pace of innovation we've been seeing here lately. So many are quick to forget (or even think fondly of) the stagnation we saw in the early to mid 2010s. AMD's weak heavy machinery architectures paving the way for Intel to begin resting on their laurels put the industry on its ass for way too long. Feels so good to see the two firing on all cylinders in the CPU space again. Here's to hoping we see a proper K6 vs. Pentium era once again!

Meanwhile apple M1 looks behind with a smiling face and waves

 

[Minus Infinity]

They won't need to do that.
AMD will have PAIRS of "BIG+little" cores sharing the L2 cache. Which means that migrating threads from a big core to it's little counterpart (or vice versa) will be quite fast.

I think given AMD has confirmed Bergamo 4c cores and there are rumours Zen 5 will indeed be hybrid with updated 5c cores it's going to happen. Zen 4c cores despite being weaker than full Zen 4 (10-30% on leaked estimates) cores will still be much stronger than Intel's E cores, and Zen 5c cores would probably be a bigger leap compared to Intel's E-cores in Meteor and even Arrow Lake AMD will be in a strong position to counter Intel's dramatic core count increase post Raptor Lake.

 

[InVasMani]

Honestly I'll be disappointed if big LITTLE isn't 3D stacked cores for AMD. Just stack a 4 sided and 5 sided dice in layers the inverted design and use the vacant space for tackling concentrated heat dispersion and/or cache insertion and allow the cores in layers stacked above or below to also access the inverted cache regions stacked above or below. In a 3 layer stack that middle layer which could likely be 5c's would scream due to being able to access the silicone space of about 14c's. That would be a TSV engineering Marvell hell.

AMD should probably work with SK Hynix on it with a bit of licensing and pair with HBM in the middle layer then it could have 10 stacks of CPU cache for the middle layer to access along with something like 4 HBM chips. The stacks above below then have like 5 regions of CPU cache with the 4 HBM chips to pool from above or below as well. Anyway that's just a obscure engineering idea that would be a real ***** of a time to design probably, but it worked as well as sounds at face value it would be a ripper o threads and/or other bits this and that. It would be fairly clever if it's feasible can be done I do believe at least until Quantum space jumps in technological advancements.

 

[dgianstefani]

I think they should do big, medium, little.
4 big superfast cores, 4 medium duty all round cores, 8 little slow cores.
Then have a robust ability to pin software threads to particular core types.

Not needed in desktop pc.

Area efficiency more important than a middle ground core imo.

Adds a 3rd dimension to a nice simple binary calculation for thread director/equivalent.

People seem to be forgetting that e cores are most likely smaller to zen c cores. It's likely that a zen c core won't be 1/4 the size of a full zen core.

40 e cores even if they're half the speed of a zen c core will still have more multithread oooomph than 16 c cores.

8+40 vs 16+16? Who knows. All of this is speculation.

 

[Frank_100]

I think given AMD has confirmed Bergamo 4c cores and there are rumours Zen 5 will indeed be hybrid with updated 5c cores it's going to happen. Zen 4c cores despite being weaker than full Zen 4 (10-30% on leaked estimates) cores will still be much stronger than Intel's E cores, and Zen 5c cores would probably be a bigger leap compared to Intel's E-cores in Meteor and even Arrow Lake AMD will be in a strong position to counter Intel's dramatic core count increase post Raptor Lake.

The Zen c cores are for servers. I remember an AMD stating that they were going to control power in each phase and section of processor to make full cores act like efficiency cores but would still be able to power up full core when required. It is looking to me that the 7950x is the HEDT chip, even if it is not as fast as intel. The only alternative is to pay $3000 for a Xeon.