News DDR5 Boosts Raptor Lake CPU Multi-Core Performance By 20 Percent In New Benchmark

I am expecting, more or less, this to be pretty accurate on release. The new IMC should yield better DDR5 scaling and now they must have figured out how to improve the ring BUS a bit more so it works better with DDR5 (gear 2 vs 1).

As with other leaks: cautiously optimistic.

Regards.
 
Reactions: KyaraM and Why_Me

Sleepy_Hollowed

Honorable
Jan 1, 2017
251
58
10,770
1
It'd be a bit more interesting to see this same test with a DDR4 3600 or 3800, but that's still impressive enough to warrant using DDR5 on those builds.

I'd love to see all next gen mainboards get support for ECC buffered RAM too, but that's just wishful really. Some AMD boards implemented and certified some DIMMs, but they were not fast enough to use outside of integrity or reliability required use cases.
 
Reactions: shady28 and Why_Me
The results showed that DDR5 didn't do much for the Core i7-13700K in terms of single-core performance. The DDR5 system was 1% slower than the DDR4 system, but it's within the margin of error. However, the DDR5 system delivered up to 20% higher multi-core performance. That's a pretty significant performance delta.
These results look extremely questionable. Single core performance is shown to be virtually identical between the two systems, yet multi-core performance gets an almost 20% boost? How do we know the system running DDR5 wasn't given a substantial all-core overclock? It seems unlikely that the different memory type would make that much of a performance difference, seeing as it didn't affect the single-core results in any meaningful way. I suppose the higher bandwidth could theoretically reduce congestion with certain heavily-multithreaded workloads, and they are apparently using different algorithms for the single and multi-core tests, but it still seems questionable.

Leaks seem to indicate that Raptor Lake's architecture is very similar to Alder Lake's, with the main changes being limited to the number of low-power cores, along with clock rates and the amount of cache, and while I don't know if it still holds true, it was rumored a while back that the memory controller would be the same. So I have some doubts that the performance characteristics between DDR4 and DDR5 will change much going from Alder Lake to Raptor Lake.

Of course, Geekbench arguably isn't an ideal choice of benchmark, and features synthetic workloads, so even if a difference is there, it may not translate to most real-world applications. I suspect the actual performance difference isn't going to be nearly that large, and we will also likely continue to see very similar, or in some cases better performance with DDR4 in latency-sensitive workloads like games.
 

KyaraM

Notable
Mar 11, 2022
931
329
890
41
Man..12th gen hasn't even completed a year and we're already starting at 13th gen?
Intel releases a new generation every year for ages now. Where is the news here? These are leaks, and they floated around for a while now in different forms. They always predate the finished product, and always have to be taken with a grain of salt since they are using CPUs that are not final and still on unoptimized platforms. Actual release is within a year of Alder Lake.
 
Intel releases a new generation every year for ages now. Where is the news here?
Sure but it would be 5-10% due to IPC in one gen, and then 5-10% due to the better arch giving better clocks the next year.
Both of which would be irrelevant because you could overclock more than that so you where ok comparing your OC system against a stock new system for 2-3 gens.
Now they are doubling the core count...of the e-cores at least, causing a big enough difference that is beyond any sensible overclock.
 

KyaraM

Notable
Mar 11, 2022
931
329
890
41
Sure but it would be 5-10% due to IPC in one gen, and then 5-10% due to the better arch giving better clocks the next year.
Both of which would be irrelevant because you could overclock more than that so you where ok comparing your OC system against a stock new system for 2-3 gens.
Now they are doubling the core count...of the e-cores at least, causing a big enough difference that is beyond any sensible overclock.
I'm... not sure what you are replying to? I was strictly replying to the poster lamenting about it not even being a year since Alder Lake, in that exact content only. I said nothing in regards to, nor was any of it intended to be applied to, the article at hand.
 

InvalidError

Titan
Moderator
Single core performance is shown to be virtually identical between the two systems, yet multi-core performance gets an almost 20% boost?
Single-threaded benchmarks can let the active core hog all of the L3 cache and mostly eliminate the memory controller from the equation. Multi-core can only dedicate ~1/24th of the L3 to each core and would be hammering the memory controller much harder.

Comparing DDR4-3200 which is basically bargain-basement stuff to DDR5-5200 which costs 50+% more doesn't seem particularly fair beyond being the maximum officially supported speeds for their respective standards. DDR4-4000 is relatively cheap and will work on just about any Intel board at this point, that would be a fairer match-up in terms of value.
 

spongiemaster

Reputable
Dec 12, 2019
2,010
1,072
4,560
0
Comparing DDR4-3200 which is basically bargain-basement stuff to DDR5-5200 which costs 50+% more doesn't seem particularly fair beyond being the maximum officially supported speeds for their respective standards. DDR4-4000 is relatively cheap and will work on just about any Intel board at this point, that would be a fairer match-up in terms of value.
How much of the market is using DDR-4000? People are more interested in seeing a comparison of the new to whatever they are using so they can see what kind of improvment to expect. 3200 is still likely the most used followed by maybe 3600.
 
Reactions: shady28 and Why_Me

InvalidError

Titan
Moderator
How much of the market is using DDR-4000?
How does this have anything to do with it? If you are buying 12th/13th-gen it doesn't matter what memory the world's install base uses since you will most likely be buying new RAM regardless of whether you go DDR4 or DDR5. You don't buy new RAM on the basis of what systems from the last five years have been using, you buy on the basis of price-performance benefits for the next 3-5 years.

With 32GB DDR4-4000 kits available for as low as $120, there is little reason to consider anything worse when building a new $1000+ PC when the alternative is spending $120+ extra just to make the jump to DDR5.
 

spongiemaster

Reputable
Dec 12, 2019
2,010
1,072
4,560
0
How does this have anything to do with it? If you are buying 12th/13th-gen it doesn't matter what memory the world's install base uses since you will most likely be buying new RAM regardless of whether you go DDR4 or DDR5. You don't buy new RAM on the basis of what systems from the last five years have been using, you buy on the basis of price-performance benefits for the next 3-5 years.

With 32GB DDR4-4000 kits available for as low as $120, there is little reason to consider anything worse when building a new $1000+ PC when the alternative is spending $120+ extra just to make the jump to DDR5.
People moving from non-Alder Lake systems are either going to save money choosing DDR4 and reusing their existing DDR4, again likely 3200/3600, or go all in and choose DDR5. So what the buyers need to see is what the performance difference is between their existing RAM and DDR5 so they can make a decision. Choosing a Raptor Lake DDR4 platform and then buying new DDR4 RAM would be stupid.
 
Reactions: KyaraM

InvalidError

Titan
Moderator
Choosing a Raptor Lake DDR4 platform and then buying new DDR4 RAM would be stupid.
DDR5 platforms cost $20-50 more than their DDR4 counterparts and the memory itself costs $120-200 more. Unless your typical uses actually benefit from DDR5, it is still cheaper to ditch your old DDR4 and buy new than go DDR5.

People reusing their old RAM in new builds is something that mostly happens when memory prices are at absurdly high levels or using formerly absurdly high-end memory that was way ahead of its time when new. Right now, DDR4 prices are near their all-time low for speed grades most people couldn't dream to touch a few years ago. The more likely scenario is people buying new CPU+MoBo+RAM and using their spare parts to upgrade a friend or family's older PC.
 

shady28

Distinguished
Jan 29, 2007
356
205
19,090
8
How does this have anything to do with it? If you are buying 12th/13th-gen it doesn't matter what memory the world's install base uses since you will most likely be buying new RAM regardless of whether you go DDR4 or DDR5. You don't buy new RAM on the basis of what systems from the last five years have been using, you buy on the basis of price-performance benefits for the next 3-5 years.

With 32GB DDR4-4000 kits available for as low as $120, there is little reason to consider anything worse when building a new $1000+ PC when the alternative is spending $120+ extra just to make the jump to DDR5.
It's actually possible for both of you to be correct from different perspectives.

They are doing the comparison using JEDEC standards. What you're neglecting here, is that limits both systems, not just the DDR4 system.

If you are trying to go all out, you wouldn't use DDR4-3200 is your contention, I agree.

What I notice is that you don't mention the obvious, if you are trying to go all out you wouldn't use DDR5-5200 either.

So to price such a comparison, I used two of (what I) consider the best specs for DDR4 and DDR5, DDR4-3600 C14 and DDR5-6000 C36. These are not the fastest for either type, but they are both plenty fast with a very high probability of working on any AMD or Intel platform, and within spitting distance of the fastest in terms of performance without price spiking way up into the 300+ range (for both types).

Guess what? They cost the same.

 

InvalidError

Titan
Moderator
So to price such a comparison, I used two of (what I) consider the best specs for DDR4 and DDR5, DDR4-3600 C14 and DDR5-6000 C36.
Doesn't make much sense to pay $230 for DDR4-3600 when DDR4-4000-18 starts from $120 and will be about as fast in most cases. It has basically become the new 3200-16: something most people can throw a few extra dollars at without much of a second thought and still get benefits comparable to ludicrously more expensive kits. From what few benchmarks I could find, there is only a 1-2% difference between 3600-14 and 4000-18 with 4000-18 sometimes winning productivity by 1-2% thanks to more bandwidth allowing cache lines to move in/out of memory faster than 1ns lower latency does.

Basically, the huge premium for super-special ultra-low-latency DDR4 is not worth it on newer platforms that can readily use dirt-cheap DDR4-4000-18.
 

shady28

Distinguished
Jan 29, 2007
356
205
19,090
8
Doesn't make much sense to pay $230 for DDR4-3600 when DDR4-4000-18 starts from $120 and will be about as fast in most cases.

It has basically become the new 3200-16: something most people can throw a few extra dollars at without much of a second thought and still get benefits comparable to ludicrously more expensive kits. From what few benchmarks I could find, there is only a 1-2% difference between 3600-14 and 4000-18 with 4000-18 sometimes winning productivity by 1-2% thanks to more bandwidth allowing cache lines to move in/out of memory faster than 1ns lower latency does.

Basically, the huge premium for super-special ultra-low-latency DDR4 is not worth it on newer platforms that can readily use dirt-cheap DDR4-4000-18.
It took me about 10 seconds to find this. That's 14FPS or about 7% loss on cheap DDR4-4000 C19 vs DDR5-6000.

Cheap DDR4-4000 is just that, cheap hot garbage. A lot of Hynix DDR4-3200 kits can be OC to 4000 if you loosen up the timing. Cheap DDR4-4000 is just that - slower ram with crap timings.

And good luck getting two 16GB sticks of DDR4-4000 C15 to work, there's a reason the heavily OC DDR4 is not sold in 32GB kits.

Lest we fail to mention, if you go with extremely high speed DDR4 you are stuck at 16GB or lower your speeds significantly. Not the case with DDR5.

The reason I chose to compare DDR4-3600 C14-15-15-35 is because it is some of the best RAM you can get for DDR4 that will work on almost any platform. This chart shows that too.


 

InvalidError

Titan
Moderator
It took me about 10 seconds to find this. That's 14FPS or about 7% loss on cheap DDR4-4000 C19 vs DDR5-6000.

Cheap DDR4-4000 is just that, cheap hot garbage. A lot of Hynix DDR4-3200 kits can be OC to 4000 if you loosen up the timing. Cheap DDR4-4000 is just that - slower ram with crap timings.
Normal people don't want to pay a $100+ premium on ultra-low-latency DDR4/5, they want the best they can get before the price curve shoots for the moon. It makes no sense to pay a $100+ premium for ultra-premium DDR4 or $200+ for ultra-premium DDR5 when going with cheap DDR4 4000-18 memory shrinks the performance penalty from 20% to 5% while saving enough for a one tier GPU upgrade and get a 20-40% performance bump from that.
 

shady28

Distinguished
Jan 29, 2007
356
205
19,090
8
Normal people don't want to pay a $100+ premium on ultra-low-latency DDR4/5, they want the best they can get before the price curve shoots for the moon. It makes no sense to pay a $100+ premium for ultra-premium DDR4 or $200+ for ultra-premium DDR5 when going with cheap DDR4 4000-18 memory shrinks the performance penalty from 20% to 5% while saving enough for a one tier GPU upgrade and get a 20-40% performance bump from that.
Well this is true, generally speaking normal people won't have a GPU or CPU that would benefit anyway. In fact, they won't spend a lot on much of anything on a PC. Nor will they buy or compare individual components.
 
Reactions: KyaraM
People reusing their old RAM in new builds is something that mostly happens when memory prices are at absurdly high levels or using formerly absurdly high-end memory that was way ahead of its time when new.
hmm i bought 3200MHz CL14 ram sticks in 2017 wasnt really absurdly high end, but it had decent latency, when i replaced CPU last year with better IMC, ram could overclock at 3733MHz CL16 1.35v which is even better than before (latency wise) :)
not everybody has to replace old ram when upgrading cpu (or mainboard)
i usual stick with one set of sticks for its entire lifetime (since DDR1), i did skip DDR2 tho as DDR1 was overclocked to 750MHz at CL3 which beated any DDR2 available and was somewhat keeping with startup DDR3 (7.8GB/s at 30ns) , then i sold them and got 8GB 1866 CL9 DDR3 for same price as i sold 2GB DDR1 xD
 

KyaraM

Notable
Mar 11, 2022
931
329
890
41
DDR5 platforms cost $20-50 more than their DDR4 counterparts and the memory itself costs $120-200 more. Unless your typical uses actually benefit from DDR5, it is still cheaper to ditch your old DDR4 and buy new than go DDR5.

People reusing their old RAM in new builds is something that mostly happens when memory prices are at absurdly high levels or using formerly absurdly high-end memory that was way ahead of its time when new. Right now, DDR4 prices are near their all-time low for speed grades most people couldn't dream to touch a few years ago. The more likely scenario is people buying new CPU+MoBo+RAM and using their spare parts to upgrade a friend or family's older PC.
I run my 12700k with brand new DDR4-3600, and have seen quite a few people on this website with the same config or 3200RAM. Just sayin'.

Also, no, DDR4-4000 is not "dirt cheap". 3200 is.
 
So to price such a comparison, I used two of (what I) consider the best specs for DDR4 and DDR5, DDR4-3600 C14 and DDR5-6000 C36. These are not the fastest for either type, but they are both plenty fast with a very high probability of working on any AMD or Intel platform, and within spitting distance of the fastest in terms of performance without price spiking way up into the 300+ range (for both types).

Guess what? They cost the same.
DDR4-3600 C14 is not what people are comparing DDR5 to. They're comparing it against more reasonably priced kits with common timings that cost $100+ less. Paying a big premium for either DDR5 or a kit of DDR4 with abnormally high frequencies or tight timings is not worthwhile, unless perhaps someone is using it specifically for a workload that can benefit from it, and doesn't have any other components that the money could be better spent on.

In the vast majority of real-world applications and games, a $250 kit of RAM will perform nearly indistinguishable from a $100 kit of RAM of the same capacity. Even in CPU-limited scenarios, the performance difference is generally going to be within a few percent, and in graphics-limited gaming scenarios, they will perform virtually identical. Unless the system is already using near-top-of-the-line components, that money would almost certainly be better put toward something like a higher-end CPU or GPU.

It took me about 10 seconds to find this. That's 14FPS or about 7% loss on cheap DDR4-4000 C19 vs DDR5-6000.

Cheap DDR4-4000 is just that, cheap hot garbage. A lot of Hynix DDR4-3200 kits can be OC to 4000 if you loosen up the timing. Cheap DDR4-4000 is just that - slower ram with crap timings.

And good luck getting two 16GB sticks of DDR4-4000 C15 to work, there's a reason the heavily OC DDR4 is not sold in 32GB kits.

Lest we fail to mention, if you go with extremely high speed DDR4 you are stuck at 16GB or lower your speeds significantly. Not the case with DDR5.

The reason I chose to compare DDR4-3600 C14-15-15-35 is because it is some of the best RAM you can get for DDR4 that will work on almost any platform. This chart shows that too.
That could potentially be relevant for someone buying a 3080 Ti just to run that one racing game at 1080p with medium settings. : P

I tracked down that review, and in another chart they show performance with the settings turned up, still at 1080p, where there is only a 1.5% difference between those two underlined RAM kits. At resolutions above 1080p, there would likely be no measurable difference even with a 3080 Ti.

And of course, the post you were responding to was comparing DDR4-4000 C18 against the DDR5-6000 C36 kit you suggested, whereas the underlined results in that chart are comparing DDR4-4000 C19 against DDR5-6000 C30. So, looser timings on the DDR4, and much tighter timings on the DDR5. Right now, a 32GB kit of DDR5 with those timings will set you back $350, while a 32GB kit of DDR4-4000 C18 (with faster timings) can be had for $120.

And you might also notice that a kit of DDR4-3600 C17 managed to perform within 3% of that $350 DDR5 kit in the chart you posted. I would not be surprised if a $140 kit of DDR4-3600 C16 managed to perform the same or better than DDR5-6000 C36 in that game. The 35% slower absolute latency of the DDR5 kit is going to often counter its higher transfer rate, at least in games.
 
Last edited:

ASK THE COMMUNITY