raptor lake Intel claims Core i9-13900K will be 11% faster on average than AMD Ryzen 9 7950X in gaming

[TetraSky]

Did they once again use slow ass memory for Ryzen while using fast RAM for their CPU?
Like when they used DDR4-3200 on the 5800X3D vs DDR5 5600 for their side?

In any case, as always, wait for independent reviews and don't believe first party "reviews".

[Shzzit]

2 hours ago, CHICKSLAYA said:

There's gonna be a 7950x3d and 7800x3d for sure. Also rumored to be a perhaps a 7900x3d. All 3 are gonna be juicy. Apparently they fixed the voltage issues with second gen 3d cache so you will be able to run PBO etc unlike the 5800x3d. If they can keep the clock speed the same it'll be absolutely naughty

Sounds amazing, uggggg I hate waiting lol.

 

2 hours ago, TetraSky said:

Did they once again use slow ass memory for Ryzen while using fast RAM for their CPU?
Like when they used DDR4-3200 on the 5800X3D vs DDR5 5600 for their side?

In any case, as always, wait for independent reviews and don't believe first party "reviews".

They are using rated ram speeds from amd and intel.  Not our fault amd was only rated at 3200mhz.

 

2 hours ago, Shimmy Gummi said:

We say this now

 

but wait until there's a sale or other promo

 

it will be just like the 10900k all over again LOL

I’m kinda waiting for a bundle deal, where you get like 3-4 free games, worth at least 150 bucks.  That might sway my purchase lol

Edited October 20, 2022 by SansVarnic
Merged - Please learn to use the Multi-quote function.

[TetraSky]

11 minutes ago, Shzzit said:

They are using rated ram speeds from amd and intel.  Not our fault amd was only rated at 3200mhz.

Meaning they are likely using DDR5-5200 for AMD and 5600 for theirs, instead of using the same speed for a fair benchmark with as few variables as possible between the two. Got it.
Last time, they used G.Skill DDR5 CL 28-34-34-89, 2X 16GB DDR5-5600MHz for their tests.

I can't wait for them update their website with these benchmarks and see if they at least used some comparable low timing RAM or some CL 40 DDR5-5200 for AMD.

https://edc.intel.com/content/www/us/en/products/performance/benchmarks/desktop/

[starsmine]

3 minutes ago, TetraSky said:

Meaning they are likely using DDR5-5200 for AMD and 5600 for theirs, instead of using the same speed for a fair benchmark with as few variables as possible between the two. Got it.

Using the specified ram IS fair. its literally their spec.
Thats like saying its not fair to compare the two because the nodes allow different clocks.

[Mister Woof]

7 minutes ago, Shzzit said:

I’m kinda waiting for a bundle deal, where you get like 3-4 free games, worth at least 150 bucks.  That might sway my purchase lol

Haven't seen this for Intel really. Mostly AMD. I've gotten a LOT of free games from AMD over the years.

[TetraSky]

4 minutes ago, starsmine said:

Using the specified ram IS fair. its literally their spec.
Thats like saying its not fair to compare the two because the nodes allow different clocks.

Lets agree to disagree on that one. While it IS the spec, for benchmarks they should reuse as much as possible the same parts across both platform, with only the motherboard and CPU changing to remove as much chance for errors in numbers between the two and actually put them head to head on the same starting line.
I would much rather see benchmarks using DDR5-6000 on both platform, meaning both are "overclock" for it, than have them use different type of RAM that may provide wrong results.

[porina]

1 hour ago, KaitouX said:

Important to note that this Techspot/HUB chart goes against to what Intel themselves claimed and that ComputerBase confirmed, that 13900K@65W is almost on par with 12900K@Stock on average, Intel used SPEC, ComputerBase used their own test suite. DerBauer CB R20 results are close to ComputerBase ones too.

Thanks. The ComputerBase results for R23 for the two top CPUs look much closer than HUB, but the advantage still goes to 7950X as it scores higher on equivalent power limit. Still, it is good info for context and I'll keep looking for more/wider testing in this style. Makes me wonder if HUB used a mobo that set excessive voltage by default or something. Think I heard of that happening somewhere in the past.

[LAwLz]

10 minutes ago, TetraSky said:

Lets agree to disagree on that one. While it IS the spec, for benchmarks they should reuse as much as possible the same parts across both platform, with only the motherboard and CPU changing to remove as much chance for errors in numbers between the two and actually put them head to head on the same starting line.
I would much rather see benchmarks using DDR5-6000 on both platform, meaning both are "overclock" for it, than have them use different type of RAM that may provide wrong results.

I disagree with you too.

I don't think overclocking one platform and running the other at stock is fair. Likewise, it would not be fair to underclock Ryzen 7000 chips because "we need to keep frequency the same between Intel and AMD to make the comparison fair".

 

I rarely agree with Linus, but I 100% agree when he said that saying "don't use this feature that company X has that company Y don't because it's unfair!" is what fanboys does when they don't want their favourite company to get beat.

You don't cripple one company just because they have a feature that the other company don't have. In this case, that feature is support for higher frequency RAM. Likewise, you don't overclock one platform while keeping the other at stock just because one platform has a frequency advantage. 

 

 

Asking to overclock on AMD and keep Intel at stock just to keep the frequency matched is just as bizarre to me as if someone thought it was unfair to let the 7950X run with all its cores, and a "fair" comparison would be to disable 8 out of the 16 cores and run benchmarks that way, because the i9-13900K only has 8 large cores. 

You don't remove advantages from one company just because the other company has feature parity. If AMD wants reviewers to run higher frequency memory then they need to make sure their memory controller can actually handle it, not rely and hope that reviewers will overclock and tweak their memory to do something Intel does out of the box.

 

 

Anyway, TechPowerUp ran their Zen 4 system with overclocked RAM. All of their DDR5 systems, Zen4, Alder Lake and Raptor Lake were running with 6000MHz RAM, CL36.

Their results align pretty well with the other results, including the first party Intel results.

[Brooksie359]

2 hours ago, Shzzit said:

Aww I get it, only group think in here ? No one can have an opinion that they don’t like gamers nexus?  Did I shade ur favorite YouTuber?  
 

Grow up lol.  
 

For one if the cpu was at intel stock settings then it wouldent go over 252 watts.  If it’s hitting 300 plus watts then it’s motherboard has changed it’s settings to u limited power.

 

And a thumbnail with a dumb face and giant words that make you think it uses massive power to do anything.

 

Then you watch them video and it’s the opposite.

 

Yoou don’t like my opinion that’s fine, telling me I can’t have one and should stop talking is not ok. 

My grip was clear, I think gamers nexus issues clickbait titles and thumb nails.

So again not sure what the hell ur crying about.

 

 

 

Getting upset about youtube channels for click bait titles and thumbnails is like getting mad at a realistate agent for staging a house to make it look good before putting it on the market. People do it because it's stupid not to as it objectively increases your chances of success. Don't hate the player hate the game as it's just how youtube works. I am not a fan of click bait just as much as the next person but I am not going to get mad at a channel for doing so as it would be stupid not to. You are basically handicapping yourself if you don't create interesting titles and thumbnails. Also at the end of the day gamers nexus does some really good work and do really indepth informative videos so it's not like I am mad at the content of the video itself. Personally I only get mad at clickbait if the video sucks. 

[porina]

I've plotted the power scaling for Cinebench R23 from Hardware Unboxed and ComputerBase. I got lazy, sorry, power on horizontal axis, score on vertical. I used same scaling on the two carts. I wanted to overlay them but I couldn't figure out how on Google Sheets. Anyway, it's obvious their results are very different, although 7950X remains on top in both cases. HUB's results make it look like are Intel even trying. ComputerBase's results are pretty close to each other. A little more digging may be an idea. Maybe tomorrow I'll see if I can get a different sheet, and also add more R23 scores/powers from other sources to see how they stack up.

 

Edit: Better chart in later post. Scroll down to find it.

Spoiler

 

 

[KaitouX]

55 minutes ago, porina said:

Thanks. The ComputerBase results for R23 for the two top CPUs look much closer than HUB, but the advantage still goes to 7950X as it scores higher on equivalent power limit. Still, it is good info for context and I'll keep looking for more/wider testing in this style. Makes me wonder if HUB used a mobo that set excessive voltage by default or something. Think I heard of that happening somewhere in the past.

Techpowerup 12900K power limit tests had a similar issue as HUB, other people that tested the 12900K at lower power limits got way better results. Likely some similar issue, seemingly related to BIOS settings messing things up, possibly misreporting the power consumption or messing up the voltage.

[Shzzit]

1 hour ago, Brooksie359 said:

Getting upset about youtube channels for click bait titles and thumbnails is like getting mad at a realistate agent for staging a house to make it look good before putting it on the market. People do it because it's stupid not to as it objectively increases your chances of success. Don't hate the player hate the game as it's just how youtube works. I am not a fan of click bait just as much as the next person but I am not going to get mad at a channel for doing so as it would be stupid not to. You are basically handicapping yourself if you don't create interesting titles and thumbnails. Also at the end of the day gamers nexus does some really good work and do really indepth informative videos so it's not like I am mad at the content of the video itself. Personally I only get mad at clickbait if the video sucks. 

I kinda get what ur saying, not really mad but more annoyed I guess. 
 

Not sure when that trend started where everyone makes a dumb face then enlarges it for the thumbnail lol.  I get why they do it, it gets a lot of views for some reason.  Alot of channels do it now.  

[TetraSky]

1 hour ago, LAwLz said:

I disagree with you too.

I don't think overclocking one platform and running the other at stock is fair. [...]

 

Asking to overclock on AMD and keep Intel at stock[...]

 

Anyway, TechPowerUp ran their Zen 4 system with overclocked RAM. All of their DDR5 systems, Zen4, Alder Lake and Raptor Lake were running with 6000MHz RAM, CL36.

Their results align pretty well with the other results, including the first party Intel results.

I did say to run both at 6000, meaning they'd both be overclocked, not just overclock AMD and keep intel stock. 
So TechPowerUp doing so is great. This is the kind of benchmark I want to see on the same playing field. Not ones limited to arbitrary lower tech specs that no one other than a cheaped out pre-built actually follows in the real world.

[porina]

33 minutes ago, KaitouX said:

Techpowerup 12900K power limit tests had a similar issue as HUB, other people that tested the 12900K at lower power limits got way better results. Likely some similar issue, seemingly related to BIOS settings messing things up, possibly misreporting the power consumption or messing up the voltage.

If you (or anyone else) knows of any more R23 results showing both score and power, I'll add them to the chart.

 

I found Onlyoffice which seems to chart a bit better than Google sheets, but has its own formatting quirks. I've managed to combine the two data sets at least.

 

 

The 7950X results from Computerbase and Hardware Unboxed track pretty closely. It's just the 13900k results that are far apart from each other. Need more data points, maybe look at mobos used too.

[KaitouX]

Spoiler

According to this person test(from this post on 3DCenter) the 12900K achieves 86% of the stock performance at 125W on CB R20, if we assume that the same applies to CB R23, the 12900K would score 23000 points on CB R23@125W, that is higher than HUB result for the 13900K, which doesn't really make sense.

 

There's also this from a user in the Techpowerup 12900K power limit test discussion showing their 12900K scoring 18700 points at 75W, which also is close to the number from the image above of ~67% of stock at 80W and almost the same as the 13900K in HUB test:

A 12900K matching the 13900K in performance when both are limited to the same amount of power doesn't seem right.

[Dr_Whom]

ICYMI HUB has acknowledged the discrepancies in their power scaling tests and have contributed it to the motherboard auto voltage they believe.

 

 

[jos]

I am confused with the reviews, which one amd or Intel which one is actually performing better, some reviewers talk about efficiency, some talk about at a given power, some talks about with what Intel/amd specified. I do not care of any of it. I want to know at a given ambient temperature, a comfortable one like 25°c and i put best hardware possible for the platform, without any over locking which perform better sustained. I think that is what most people want to know

[KaitouX]

5 minutes ago, jos said:

I am confused with the reviews, which one amd or Intel which one is actually performing better, some reviewers talk about efficiency, some talk about at a given power, some talks about with what Intel/amd specified. I do not care of any of it. I want to know at a given ambient temperature, a comfortable one like 25°c and i put best hardware possible for the platform, without any over locking which perform better sustained. I think that is what most people want to know

The 7950X is usually slightly better in heavy all-core while the 13900K is usually better in lighter workloads and gaming. The 13700K and 13600K are way better than what AMD has at similar prices.

[jos]

1 minute ago, KaitouX said:

The 7950X is usually slightly better in heavy all-core while the 13900K is usually better in lighter workloads and gaming. The 13700K and 13600K are way better than what AMD has at similar prices.

Thank you, I am a developer so it is usually heavy workload, and i need it sustained not some small burst of performance. Now a days even chrome feels like heavy application /s

 

 

[BiG StroOnZ]

Here's a 13th Gen “Raptor Lake” CPU Review Roundup:

 

https://videocardz.com/140948/intel-13th-gen-core-raptor-lake-desktop-cpu-review-roundup

[Shzzit]

56 minutes ago, BiG StroOnZ said:

Here's a 13th Gen “Raptor Lake” CPU Review Roundup:

 

https://videocardz.com/140948/intel-13th-gen-core-raptor-lake-desktop-cpu-review-roundup

Nice, thanks brother.

[BiG StroOnZ]

52 minutes ago, Shzzit said:

Nice, thanks brother.

 

No problem, enjoy.

 

Been going through them throughout the day. Seems that Raptor Lake has the edge. But that Zen4 trades blows in the high end, but in the "midrange" Raptor Lake takes the win.

[HenrySalayne]

1 hour ago, BiG StroOnZ said:

Been going through them throughout the day. Seems that Raptor Lake has the edge. But that Zen4 trades blows in the high end, but in the "midrange" Raptor Lake takes the win.

I'm pretty sure we'll see price drops on AMD's line-up in the near future. Zen 4 sales were pretty rough even before Raptor Lake came along.

It is really good to see some actual competition in the market.

[BiG StroOnZ]

6 minutes ago, HenrySalayne said:

I'm pretty sure we'll see prices drop through the next month. Zen 4 sales were pretty rough even before Raptor Lake came along.

It is really good to see some actual competition in the market.

 

Yeah, I saw this article the other day:

 

Quote

The report that cites AMD's internal management suggests that the red team is planning to lower its production of the Ryzen 7000 "Zen 4" CPUs amidst a decline in the PC market and the overall poor reception of the AM5 platform.

 

https://wccftech.com/amd-ryzen-7000-zen-4-cpu-lower-production-capacity-due-to-pc-decline-ryzen-9-7900x-best-seller/

 

Some other tech news sites echoed that article, as WCCFTech was the source. I'm not sure of the exact accuracy of it, however, but it does make sense given the AM5 platform's barriers to entry.

 

Regardless, this gen is shaping up to being as exciting as we expected many months ago. Competition is strong between the two. Hopefully AMD can provide equal competition in the GPU space with RDNA 3 against NVIDIA, and Intel eventually to catch up in the dGPU arena as well. I agree that prices should drop eventually, given Intel has backwards compatibility with motherboards and DDR4 support. Also, as stated in the WCCFTech article, we've still yet to see the promised $125 B650 motherboards. So that might become a thing.

[MageTank]

14 hours ago, TetraSky said:

I did say to run both at 6000, meaning they'd both be overclocked, not just overclock AMD and keep intel stock. 
So TechPowerUp doing so is great. This is the kind of benchmark I want to see on the same playing field. Not ones limited to arbitrary lower tech specs that no one other than a cheaped out pre-built actually follows in the real world.

I'd say don't run both at 6000, both platforms are having issues with DDR5 when pushing higher frequencies at the moment. XMP (DOCP, EXPO, whatever AMD decided to call it this year) has quirks beyond 5600Mhz on multiple board vendors. First and foremost, if you go beyond 5600Mhz, you lose sleep functionality on ASRock boards and if you attempt to attempt to "Reset this PC" from a recovery environment on ASUS, you BSOD and memory defaults back to 4800 despite keeping XMP loaded and not warning you that the overclocking safe mode was loaded, so you'll be none the wiser to memory speed changing.

 

Performance wise, I haven't noticed much scaling with memory this time on Ryzen. Here is my awful graph showing my findings in Metro Exodus:

Note: Red is average, green is minimum. Benchmark was configured on 1080p Extreme preset, 3 passes enabled and averaged. Results sorted by minimum framerates.

Testing Platform:

• Motherboard: ASRock X670E Steel Legend
• Cooler: Cooler Master ML240L V2 AIO (pump 100%, fans auto)
• Chassis: Cooler Master MasterFrame 700 Open Bench
• Memory: Varies (component under test)
• CPU: Ryzen 9 7900X
• GPU: RTX 3090 FE
• Ambient Room Temp: 74F

Looking at benchmarks from Guru3D, would appear they are seeing similar poor scaling: https://www.guru3d.com/articles_pages/ddr5_ryzen_7_7700_ddr5_memory_scaling_review,19.html. If anyone has any games they would like to see tested for memory scaling, I am open to suggestions.

 

Now I have a few theories as to why this is. I think first and foremost, the larger cache on AMD is helping mask the shortcomings of the DDR5 IMC and their unsynced fabric clock. Larger cache has always shown a benefit to gaming performance, even as far back as original Broadwell when comparing clock for clock performance and factoring in Broadwell's IPC increase over Haswell.

 

Secondly, I think because the fabric is unsynced and running at 2000Mhz on most boards by default, we are not seeing the fabric scaling uplift that we saw on Zen 3. We have to remember that when you overclocked memory frequency on Zen 3, you also overclocked the fabric and everything attached to it, including your PCIe subdomain and its various interconnects. The performance uplift wasn't coming from raw frequency increase, we know bandwidth doesn't work that way for traditional gaming applications (though latency improvements can certainly help).

 

Speaking of latency, this platforms stock memory speed is 5600Mhz, with a recommendation from AMD to run at 6000Mhz. If you run at 6000Mhz, you are running your fabric unsynced at a 3:2 ratio MEMCLK/FCLK. On Zen 3, running an unsynced fabric (meaning anything other than 1:1), you encountered a raw 10ns latency hit that could not be overcome with adjusting primary/tertiary timings, it simply always existed. I've been trying to test if that rule holds true here, but for the life of me I cannot get my memory to post at 4400 C22 with a 2200 fabric despite trying several different kits and processors. 4000 C22 at 2000 FCLK is also unstable. Something appears to be broken with board training so I suspect we will have to find a full set of timings to dial in to make this stable, or I'll need a board with a more robust training algorithm. 

 

Long memory rant aside, I see your point about running two test platforms using identical hardware, I personally do this for my work as it's my personal preferred testing procedure. That said, if one side has a more robust IMC (frankly I think both AMD and Intel are in trouble with their DDR5 memory controllers at the moment) they should be able to show that benefit off in benchmarks as a boon for choosing their platform. This is no different from when Intel dominated AVX benchmarks prior to AMDs adoption when comparing to traditional SSE results. If you have better instruction sets and feature supports and it matters for the given application under test, you have to be able to utilize that advantage as long as you disclose the methodology.

 

Also, since NDA is lifted and these are live, I'll share the rest of my R23 CPU results:

 

 

AMD Platform:

•  Motherboard: ASrock X670E Steel Legend

•  Cooler: Cooler Master ML240L V2 AIO (pump 100%, fans auto)

•  Chassis: Cooler Master MasterFrame 700 Open Bench

•  Memory: G Skill Ripjaws F5-5600J3636C16GX2-RS5W (2x16GB 5600 C36) (EXPO loaded)

•  Ambient Room Temp: 74F

 

Intel Platform:

•  Motherboard: ASUS Z690 ROG Strix Gaming F

•  Cooler: Cooler Master ML240L V2 AIO (pump 100%, fans auto)

•  Chassis: Cooler Master MasterFrame 700 Open Bench

•  Memory: G Skill Ripjaws F5-5600J3636C16GX2-RS5W (2x16GB 5600 C36) (EXPO loaded)

•  Ambient Room Temp: 74F

 

Haven't really paid attention to any R23 reviews, but I'd be curious if my results match others in the field.