• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Just watched Linus's performance results of the 5000 series

  • Thread starter Thread starter rn2
  • Start date Start date
So basically all the left over crap goes into 5600x.
AMD would have to have quite poor Yield to have enough chiplets with dead cores to make enough 6 core chiplets.
In reality, TSMC Yield is very good, they should have around 90% good dies and from the remaining 10%, most could be salvaged as 6 core dies.
So lets say they get 520 good dies and 30 salvaged (6 core) dies.
What's happening si AMD is binning every single die... a good portion of those good 520 dies getting some weaker cores fused off to get enough 6 core chiplets.
 
Sure, I don't generally see reviewers use faster than 3600. Speeds of 4000+ is mainly used by overclockers trying to set records and require high voltages that people don't want to use on a daily machine - it's also very expensive to buy.

No he did not set it up to favour anyone, he used the same ram speed on both systems.

atop looking for excuses, intel lost just like trump lost, just accept it
He set it up like that to gain views, just like it was posted here and probably plastered all over reddit.

If he had reviewed it and the gap was much smaller like it was on most other reviews then people wouldn't be so keen to talk about his content, Linus is very smart he knows how to get the views.

Stock vs stock is a fair comparison. Overclock vs overclock brings factors such as mobo quality and capabilities as well as CPU silicon quality into question. The kind of performance is then not guaranteed to us mortals who are watching these videos
You pay extra for a K CPU which has been specifically binned to hit those all core turbo frequencies.

I am not aware that HB use Intel systems with MCE turned on. If they did then that’s fairly naughty of them. Although with the way ryzen works - PBO is effectively an AMD overclock.

MCE is a 1 click feature just like turning on XMP or PBO so really its no different than either of those. All are overclocking features and should all be left on or off and not pick and choose

memory turning is fair enough. 3600C14 you can get standard XMP kit off the shelf. If they were using 4000 with 1.45V for testing and running c17 then that would be extremely dubious as that is fairly specialist ram and not even the best ram can guarantee that sort of the performance as you got to be pretty lucky
3600/14 ram is over double the price of what most users of ryzen buy and is a highly binned enthusiast kit which will easily do 4000/16. Also fine tuning memory is well beyond the average user especially compared to simply turning on MCE.
 
Sorry but I don't follow your logic. AIUI there are simply two 6-core CCDs. Not one high quality and one low.

I'm only going on what AMD did with the 3900X, not seen anything 5900X specific. With the 3900X there was always one CCD that would hit the advertised boost clocks and usually the second would not. It may be yields are very high with 7nm now so this is less obvious but we will know once they start getting tested in detail.
 
I'm only going on what AMD did with the 3900X, not seen anything 5900X specific. With the 3900X there was always one CCD that would hit the advertised boost clocks and usually the second would not.

In the videos I've watched all cores hit the boost clock.
 
In the videos I've watched all cores hit the boost clock.

https://www.kitguru.net/components/cpu/luke-hill/amd-ryzen-9-5900x-zen-3-cpu-review/all/1/

Mentions clocks dropping under all core load. With the higher core count chips power limits start to limit peak clocks under heavy multithreaded workloads.

"We see the 5900X running at around 4.3-4.4GHz under an all-core load situation in the likes of Cinebench R20 and Blender. This an impressive boost clock figure for the 12-core chip directly out of the box as it is around 300MHz faster than even the 3900XT 12-core processor."
 
Gamers Nexus have just put out a video which goes into more detail.


Apparently the DDR4-3600 usage was relevant but not the prime factor. It's a question of ranks and memory bandwidth. There's a comment to the video which goes into detail:

Mirsad Redzovic on Youtube said:
Steve, this is long, but worth the read.
Apparently, it's not the "4x8GB" that helps the Zen3.
It's that using full "4-Ranks" (256-Bits of memory bandwitdh) will be optimal because of the memory cycles. It's something about allowing more cycle to start while more are finishing.
Ram sticks have ranks.
Most if not all 8GB sticks are Single Rank.
Which means if you remove the heat spreader, you will have all of the IC's on one side of the PCB.
1 stick of 8 chips on one side will requires 64-Bits of bandwidth. (single rank)
Using 2 sticks of Single Ranked RAM sticks will use 128-Bits of memory bandwidth.
When you use 4 sticks of SIngle Ranked RAM sticks, you'll fully utilize the 256-Bit bandwidth. More than that will overload/strain most CPU's memory controllers and you'll reach diminishing returns on CPU performance for trying to double your RAM. You may also limit your OC capabilities.
The "Sweet Spot" is 4 ranks of memory.
This CAN BE HAD with 2 sticks of RAM, as long as both sticks are Dual-Ranked (meaning there are chips on both sides of the PCB).
This is typically found in 16GB sticks. This is why people going for 32GB system memory may be in luck and reap these benefits with 2 sticks of 16GB Dual Ranked sticks. That will be a total of 4-Ranks (256-Bits).
Upon searching what kind of config would be best for a Zen3 system seeking 64gb of memory, I have stumbled upon the fact that the Crucial Ballistix MAX 16GB sticks are SINGLE RANKED. (Micron chips).
Not only that, but they come in 4000mhz CL18 @ 1.35v. Which would be a GREAT candidate for people trying to reach the 2Ghz infinity fabric.
The beauty of this ram is that you can buy 4 sticks of 16GB dimms, and be running only 4 Ranks and reap all the performance benefits while not overloading the CPU memory controller. This way you can achieve 64GB of ram for those that require it.
I was watching some reviews on this memory and left some useful feedback for some.
I have pasted it below for people seeking this type of information:
Regarding the Crucial Ballistix MAX 4000Mhz 32GB (2x16GB kit)
If anyone is wondering why this ram is so expensive compared to other 32gb kits...
Each 16gb module is single ranked. Totaling 2 ranks.
That means that the memory IC chips are all on one side. 8 chips. Each chip size is 2gb. Each rank requires 64-bits of bandwidth.
These are Micron Chips. Micron is one of the only companies that does this in the performance market.
Samsung B-Die chips are limited to 1gb IC's which means if you were to get a 16gb module with Samsung B-Die or other competitors, they would have 8 chips on "each" side to equal the 16gb ram capacity, meaning they are dual ranked, using 128-bits per module.
AMD platforms do NOT like more than 4 ranks of memory (256-bits).
So lets say you want 64gb of ram. Getting 4 of these sticks would equal 4 ranks of memory (256-bits), which in Daisy Chain memory topology, can actually perform better than just two sticks of single rank memory (using only 128-bits).
If you wanted 64gb of ram in 4 ram slots and you went with Samsung B-Die IC's or non-Micron, you'd get 16gb modules with memory IC's on both sides of all 4 dimms which equals 8 ranks since each dimm is dual ranked (512-bits). This could overload your CPU's Memory Controller, limiting your OC to your CPU AND your Memory.
I suggest really doing research on finding the best ram for your AMD platform if that's what you have. Intel isn't AS sensitive to this, but it still is to a point.
There is a fine balance with AMD's Zen2 currently. It's somewhere around 16 cas latency and 3800Mhz Ram with a 1900Mhz infinity fabric.
These 4000Mhz dimms are great because if Zen3 can overclock their infinity fabric to 2ghz, you can run a 1:1 with this ram. The advantage to this is: A higher infinity fabric will equal great performance boosts across the board on AMD (gaming, computing etc...). When you run 3200mhz ram, your infinity fabric is 1600mhz (same frequency as the ram before DDR). If you run 3600mhz Ram, your infinity fabric gets cut in half since it will not go to 1800mhz unless you force it and possibly adjust voltage for stability. So now you see why if Zen3 allows a 2000mhz infinity fabric, this 4000mhz ram would be quite the sweet spot as you could run a 1:1 divider.
Not to mention, when overclocking your ram & cpu on Ryzen... the SOC voltage has to be increased, and often times once that's pushed passed 1.1v, you can be victim of having your PCI-E 4.0 down-spec itself to PCI-E 3.0. This can affect users which are trying to benefit from the max GPU performance from newer graphics cards, and most importantly... any users using new generation PCI-E 4.0 NVME drives such as Corsair's or Gigabyte's current offerings.
Hope this answers your question as to why this ram is expensive compared to other 32gb (2x16gb) kits. These sticks are single ranked and most if not all are dual ranked. Because of this, it allows you to not complicate all of the above mentioned since it can operate a 64GB setup with only 4-ranks @ 256bits that won't need a high SOC voltage, and allow a possible Infinity fabric overclock of 2GHz on the new Zen3 platform which means you can run a 1:1 divider between the memory and the fabric. (2000mhz D.D.R. "Double Data Rate" = 4000mhz ram).
Finding 4000mhz 16gb modules that are single ranked @ CL18 and 1.35v is not an easy task. Micron chips are charging for that novelty. I don't blame them for the price.
Hope this helps someone in their searches for the right ram sticks for their Ryzen Zen3 & Zen2 systems.
 
Gamers Nexus have just put out a video which goes into more detail.


Apparently the DDR4-3600 usage was relevant but not the prime factor. It's a question of ranks and memory bandwidth. There's a comment to the video which goes into detail:

That comment is so good, basically explains clearly how it works in a way that is easy to understand, surprised gamersnexus hadn't figured that out and actually I'd go as far as to say made them look a little silly.
 
That comment is so good, basically explains clearly how it works in a way that is easy to understand, surprised gamersnexus hadn't figured that out and actually I'd go as far as to say made them look a little silly.

I find it hard to watch and listen to Gamer Nexus. His delivery is so dry and technical its hard to pay attention and I end up skipping to the graphs :)
 
Back
Top Bottom