Raptor Lake Leaks + Intel 4 developments

[g67575]

Based on the Geekbench 5 scores, it looks like the Ryzen 7700X and 13700K are gonna be pretty close in single core performance, comparing the integer and floating point workload scores:

7700X: https://browser.geekbench.com/v5/cpu/17122041

13700K: https://browser.geekbench.com/v5/cpu/17106570

7600X ST scores looking competitive too:
https://browser.geekbench.com/v5/cpu/16966070

Worst case, the 7700X will match the 13700 (non K) in single core performance. According to the official specs, the power draw on the 7700X should be significantly lower than the 13700K.

i9s are generally ahead of Zen 4 though, at least in Geekbench 5.

 

[g67575]

Geekbench does indeed seem to be AMDs preferred benchmark this time around but CPU-Z and cinebench are both showing a 10% advantage for Intel in single core, while the 13600k looks to to have a lead of 65% in MT for just £50 more if rumours in pricing are to be believed.

Geekbench 5 measures a range of tasks (shown in the breakdown), with the 13700K coming out on top in some, and the 7700X a bit ahead in others.

Cinebench measures measures 2D tile rendering performance, which doesn't correlate to other CPU tasks.

 

[g67575]

I'll be crying myself to sleep over the lost MT score points.

 

[g67575]

The problem with CPUz (v2017 x64 benchmark), is that again, it just measures one thing, as follows:
"The new benchmark computes a 2-dimensional noise function, that could typically be used in a game to generate a procedural map."
Info from here:

CPU-Z 1.79 : new benchmark, new scores. | News | CPUID

The latest version of CPU-Z introduced a new version of the benchmark (2017.01), that succeeds to the previous version introduced in 2015. We've received a lot of emails concerning the new scores, and we'll try to answer to the most frequently asked

www.cpuid.com

and apparently it still uses just that one algorithm to generate the score.

Apparently, the CPU instructions used are 'scalar SSE/SSE2 instructions', maybe Intel has an advantage here?

 

[g67575]

Cause more core make CPU go Brum? Would buy a console with 24 E-cores, totally worth it.

 

[g67575]

If the ST performance of a CPU is overall about the same, and there are fewer cores than the competition, then I would have thought total MT performance will always be lower. But people are interested in Zen 4, because of it's apparent ability to compete with 2nd gen. Goldencove CPUs in most tasks, at a much higher base clock, using less power.

The problem with E-cores, is that they don't make up for the inherent high power usage of Goldencove's P-cores at higher frequencies (all they can do is contribute to total power usage, at a much reduced amount). The total number of high performance cores kept scaling up with the 9th and 10th Intel generations (I think many expected this scaling up to continue), but basically it has been difficult to scale beyond 8-10 cores, while bringing in the IPC advantages of microarchitectures like Goldencove on 10nm (Intel 7).

The main reason I'm considering Zen 4 over the 13th gen, is that AMD is offering the ability to switch to a DDR5 (offering higher bandwidths) platform, that they have confirmed is going to be used for many more years to come. I also tend to think that AMD's design on TSMC's 5nm process, scales better above 8 high performance cores. For example, I think the 7900X (12 Zen 4 cores) should offer more than enough performance (in heavily multithreaded tasks), for what most people need. To some extent, it gives a clue about where next gen Zen CPUs are headed - On the other hand Meteor Lake is still going to be using E-cores /hybrid core design.

Purely for gaming, it looks like either an 8 or 16 core V-cache CPU is likely to give the best performance per $$$. Particularly if offered at $500 for an 8 core model (likely considering the $449 MSRP of the Zen 3 based 5800X3D).

I can see why people want to stick with Intel though, if they already own a LGA1700 board. The 13th gen might actually be more appealing to some without the E-cores, especially if Intel charged £50 less across the board (not everyone is concerned with future upgradability, right now).

 

[g67575]

Higher core counts should hopefully come down in price over time, it's not really mainstream still. This has a lot to do with Intel, I think.

V-cache seems to be the thing to spend money on, at least on Zen 4. It might be that Zen 4 is only really competitive in games, but that might be all a lot of people are interested in anyway. I'm expecting 9-10% higher performance in games, compared to what the 7700X will offer.

 

[g67575]

All we know so far is that Zen 5 will be:

• 'Optimised for scale'
• Built on TSMC's 4nm EUV process technology
• 'New Grounds up architecture' with enhanced performance + efficiency
• Similarly to Zen 4, there will be 'Front End' performance improvements

From these slides:

https://images.anandtech.com/doci/17439/Zen5_678x452.jpg

https://cdn.mos.cms.futurecdn.net/8ChX3WGdPNf2DFEd2G6FuY.jpg

Judging by the fact that they are using a denser fabrication process, and mention scale optimizations, it's probably safe to assume that the core count will increase.

The most advanced '4nm' process appears to be called 'N4X' and offers improvements to clock frequency and allows higher voltages to be used. It's estimated to go into volume manufacturing in 'H1 2024'. More info here:

TSMC Unveils N4X Node: Extreme High-Performance at High Voltages

www.anandtech.com

 

[g67575]

People will buy these CPUs because they like the AM5 platform, these CPUs also have a better memory controller than Intel's Goldencove CPUs. If the performance gains in games are accurate, AMD shouldn't have a problem selling Zen 4 CPUs. Intel's 13th generation prices won't be cheap either, so I think that's the main reason.

 

[g67575]

Because people don't care about MT performance, if the number of cores their CPU has already provide enough (for games, work or whatever). It just isn't important to most users. They want faster cores, not more cores.

E-Cores can actually be a disadvantage in some situations, it's very much a compromise in design.

Regarding Zen 4's IMC, it runs at full speed upto 3000mhz, with DDR5 6000 MT/s RAM.

This was something that bothered me a bit about Alder Lake (and the 11th generation too with DDR4 - So, I decided to go with the 10th generation instead), the memory controller has to be run in gear 2 with DDR5.

 

[g67575]

Did you learn nothing from your hundreds of posts in the CPU section over the last few months? Zen4 on AM5 demands (expensive) DDR5 and (expensive) AM5 motherboards. Sure, people will buy them - though I don't think it'll be anywhere near as popular as Zen3 was on release, due to the global recession, inflation and the very high initial investment cost.

I'm glad there's competition from AMD and new tech, you don't have to be happy about it.

 

[g67575]

In this DSO review, a 9900K was handling Spider-Man Remastered comfortably over 60 FPS:
https://www.dsogaming.com/pc-perfor...pider-man-remastered-pc-performance-analysis/

Even with 4 or 6 cores (and threads) active, still getting minimums over 60. CPU utilization doesn't tell the whole story.

Good framerates with this CPU, tested with various RT options enabled, as shown here:
https://www.dsogaming.com/pc-perfor...ss-fsr-2-0-benchmarks-comparison-screenshots/

Lots of people were pleased with the 5600X's performance in games, so there's clearly a market for a successor.

I'd generally agree that an 8 core is the way to go for games, but prices for 8 P cores /8 Ryzen cores (still) haven't come down enough to overtake 6 core CPUs in price/performance.

 

[g67575]

Basically, what I took from this review, is that you need ~6000 MT/s DDR5 to overtake DDR4 performance (running in gear 1) in latency sensitive tasks like games:
https://www.techpowerup.com/review/intel-core-i9-12900k-alder-lake-ddr4-vs-ddr5/3.html
https://www.techpowerup.com/review/ddr5-memory-performance-scaling/3.html

So, you lose some performance to running in gear 2 mode. It's still early days, so we'll see if this makes much difference for Zen 4.

The impression I get, is that you don't seem to think it matters if Zen 4's memory controller runs at a higher frequency, than Goldencove's?

 

[g67575]

<delete please>

 

[g67575]

Yet, somehow what you've said is even less enlightening, takes some doing

 

[g67575]

The results look pretty clear to me, gear 1 or gear 2, DDR4 @3600 MT/s is slower than DDR5 @6000MT/s in games. This seems consistent with other memory scaling reviews I've seen.

Another way to look at this, is that this is definitely the point where DDR5 becomes worth it for tasks like gaming.

Using DDR5 may not be worth it with new Intel builds on LGA1700, if the user already has DDR4 RAM (even if DDR5 does provide better performance at higher frequencies).

But there's no reason why someone building a new PC shouldn't choose DDR5 for a new build (assuming your goal is the best performance) considering that the cost for 2x8 modules is ~£120 and 2x16 modules is ~£200.

I think some of you are overestimating how many people have a desktop CPU, with over 10 CPU cores in total or greater. Here's some data from Steam's August hardware survey:

• only ~1.22% of systems had a CPU with 10 cores (such as a 12600K)
• ~2.6% with 12 cores (e.g. 12700K)
• ~0.69% with 16 cores (e.g. 12900K)

From here: https://store.steampowered.com/hwsurvey/cpus/

In total, that's just 4.51% (approx.). Out of this figure, some of these will be AMD CPUs and some systems will be LGA1700 boards already using DDR5. There will also be a percentage of systems with locked 12th gen CPUs (with 4 or 6 cores), but it's harder to measure that. So, all the discussion of marketshare, somewhat misses the point (it's Intel willy waving, not relevant to consumers).

TL;DR
The vast majority (over 90%) of PC systems have 8 cores or less, and few of these are going to be locked 12th gen CPUs, using DDR4 RAM.

 

[g67575]

There's a lot of rubbish (especially on Twitter) at the moment, probably better off waiting for October for full details.

Other than that, the locked Intel CPUs and mid tier boards aren't coming until early 2023.

 

[g67575]

What kind of price increases are expected for the 13th gen, vs the 12th?

I looked back at the i9 prices from the 10th, 11th and 12th generations, and it looks like there was a ~10% price increase per gen.

So, I'd guess this would be a likely increase for the 13th gen i9s? Maybe less for the i7s?

There was a bit of info about this back in July:
"some of the price increases, inflationary increases, have turned out to be more permanent, where there’s a certain amount that we do need to pass on to the customers.”

Intel confirms it will raise CPU prices after losing $500 million this quarter

Intel said that it will raise prices on processors and other products and confirmed it is exiting the Optane and drone businesses as it reported an unexpected loss for the second quarter of 2022.

www.pcworld.com

 

[g67575]

Some good testing methodology for the games, in this review:
"In order to ensure a fair and unified test, the demo and frame number statistics that come with the game are used, and each game runs the demo five times, and the average is taken".

A decent performance improvement in CPU intensive games:
"In high-frame games that focus more on CPU, such as Ashes of Singularity, CSGO, etc., the improvement of 13900K compared to 12900K can be 10%+".

This is perhaps the most interesting bit:
"the Ring frequency of RPL is decoupled from Ecore, so there will no longer be the problem that the Ring slows down significantly when the ADL is loaded with small cores".

And this too:
"However, the RPL will still slow down. The 13900K will be downclocked from Auto 5000MHz to 4600MHz, but the improvement is much larger than that of the 12900K, but it means that you can still improve the game performance by turning off Ecore".
Sounds like it's fine to leave the E-cores on now, without impacting performance (when E-cores are under greater load) much.

 

[g67575]

All in all, not much of a gaming CPU given the price + power consumption. Pretty pathetic effort from Intel tbh, but I guess it was expected since it always looked like a tweaked ADL.

It's not that bad really, I was expecting the same architecture and fabrication technology. This is basically what we got, but the i9s appear to be using improved silicon to reach upto 6ghz (just 1 core) for the top SKU. Also optimization to the ring cache / E-core config. There seems to be (upto) 1% IPC improvement too, which apparently, Intel did not wish to advertise on their slides .

So, there is some benefit for premium CPUs, but for the i7s and i5s, I don't think there will be much in it, except if you care about the extra E-cores. The locked 13700 ought to provide the best value overall. It seems I was correct in thinking that another Golden Cove based generation was a largely unplanned move, that came about as a result of delays to Intel's 7nm CPUs (which is now known as 'Intel 4').

The main problem is that 13th gen unlocked CPUs will undoubtedly use more power than than their 12th gen i7 and i9 counterparts (which already ran very hot if overclocked). When comparing unlocked i9s (with board power unlocked, turbo boost enabled), the 13th gen 13900K uses significantly more power than the 12900K, as shown here:

https://i0.hdslb.com/bfs/article/ea7d422a6327e3109b4d87874fb9926e6e91d3e8.png@846w_507h_progressive.webp