*Official Benchmark Thread* - Post it here or it didn't happen :D

[electrosoft]

@win32asmguy

 

So i finally got a chance to delid that binned 12900k and you can see now why it ran so hot in the face of superior V/F curve at 4.2 even against my cool hand luke which was still pulling a touch more but running much cooler on the desktop and laptop capped equally. Having tested ~6 12900k's and one 12900KS chips, yours just flew in the face of, "why is it pulling so low yet running so warm even capped at 140w and/or UV -.100 on the desktop even with a full 360mm AIO??"

 

I think it is a case of a great low range V/F chip that had a terrible stim job.

 

In this pic the STIM is grossly evenly distributed and I've never seen whole areas missing before especially with those odd patches looking this bad compared to other delid's I've done. First glance at the STIM pattern (especially around the edges of the die) left me going, "ahhhh there we go..."

 

We'll see how it does after a proper clean up and relid.

 

 

 

[Raiderman]

@electrosoftI have not bought any new memory yet. I guess if it says AMD Expo, it is Hynix memory? I cant seem to find where they list the name of the ram on the egg.

https://www.newegg.com/g-skill-32gb/p/N82E16820374427?Item=N82E16820374427

[Papusan]

On 7/4/2023 at 11:53 PM, Papusan said:

Not white but if you can find the new Corsair HX1500i ATX 3.0 PSU at a good price then you have a good unit. You may also have chance to find the older 2021 version of HX1500i (my son have one) but without the ATX 3.0 specs. The good thing... You'll also avoid get the small flimsy 12VHPWR connector on the PSU side even if the PSU is speced with ATX 3.0 specs. See my older post above. 

 

I needed dual PSU setup to fulfill my needs. Imagine, if I bought the new modern who usually offer minimum of 8 pins connectors than the older PSUs... I would be damn angry. New is newer. Not better. 

From Aris on Hardware Buster. He is spot on. Exactly what I have said the whole time. Stupid have the flimsy 12VHPWR connector on the PSU panel (ATX 3.0 models). Not only due the change of new revised connector but having this small fragile Power connector on the PSU you also have the same bendgate mess as for the GPU connection. + they have a very bad locking mechanism. (He evnen adviced the PSU manufacturer to use old fashion 8-pins on the PSU side for the 12VHPWR cable). See and learn from 1:29

 

Very pleasant to see him come to same conclusion as me 🙂

 

 

 

New has to be better bro @Rage Set😉

 

This looks fun. But will probably be watered down to nothing. Same also against Nvidia. For the records... I have posted this before. CableMod don't use the terminals they bragged about in Igorslab review of their angled adapters. They have changed and use cheapo home-cooked junk. From what I have read, CM have swapped to 2 dimple design. And 3 dimple design wasn't good enough for Intel. They wanted 4 spring design. Go figure what CM have done.

Class action time. Please message me if you had a cablemod melted adapter that broke your GPU.

[Mr. Fox]

4 hours ago, Raiderman said:

@electrosoftI have not bought any new memory yet. I guess if it says AMD Expo, it is Hynix memory? I cant seem to find where they list the name of the ram on the egg.

https://www.newegg.com/g-skill-32gb/p/N82E16820374427?Item=N82E16820374427

I am not 100% positive, but I do not believe that EXPO has anything to do with the brand of the IC. I believe it is an AMD standard like XMP is for Intel. If I am not mistaken, EXPO means it has an "AMD XMP profile" available in the SPD that is optimized for AMD. I think you might find memory modules with an EXPO profile using Micron, Samsung or SK Hynix ICs on them. Samsung B-die DDR5 performs well and is stable up to around 6000-6200, maybe even 6400 with golden sample memory chips. SK Hynix is preferred due to the much higher memory overclocking capacity, like 7200 and higher, that it offers.

[electrosoft]

6 hours ago, Raiderman said:

@electrosoftI have not bought any new memory yet. I guess if it says AMD Expo, it is Hynix memory? I cant seem to find where they list the name of the ram on the egg.

https://www.newegg.com/g-skill-32gb/p/N82E16820374427?Item=N82E16820374427

 

I've been looking but I haven't found anything that associates Hynix memory for AMD with AMD Expo qualifications. It is just their version of XMP.

 

It really is a crap shoot which brand will be under the hood unless you track down a known, good used set of sticks in regards to maker. Lucky with AMD 7000 series, M-die is as good as A-die for us frequency wise as AMD gives up the ghost even on a top bin IMC around 6400-6600 and usually settles in around 6000-6200 on SR sticks. DR sticks can sometimes be problematic >5600.

 

At least Newegg and Amazon have good return policies if needed in case you end up with Samsung.

 

 

 

 

[Mr. Fox]

6 hours ago, electrosoft said:

 

I've been looking but I haven't found anything that associates Hynix memory for AMD with AMD Expo qualifications. It is just their version of XMP.

 

It really is a crap shoot which brand will be under the hood unless you track down a known, good used set of sticks in regards to maker. Lucky with AMD 7000 series, M-die is as good as A-die for us frequency wise as AMD gives up the ghost even on a top bin IMC around 6400-6600 and usually settles in around 6000-6200 on SR sticks. DR sticks can sometimes be problematic >5600.

 

At least Newegg and Amazon have good return policies if needed in case you end up with Samsung.

If maxing out the frequency is not the objective or the platform prevents it, M-die is actually better because you can run it with tighter timings than what is possible with A-die. Especially tRFC (in the low and mid 400's) and tCKE (in low single digits, like 5 or 6).

[seanwee]

New laptop new shunt mod

https://www.3dmark.com/3dm/96618692
https://www.3dmark.com/pr/2431316

[Raiderman]

6 hours ago, electrosoft said:

 

I've been looking but I haven't found anything that associates Hynix memory for AMD with AMD Expo qualifications. It is just their version of XMP.

 

It really is a crap shoot which brand will be under the hood unless you track down a known, good used set of sticks in regards to maker. Lucky with AMD 7000 series, M-die is as good as A-die for us frequency wise as AMD gives up the ghost even on a top bin IMC around 6400-6600 and usually settles in around 6000-6200 on SR sticks. DR sticks can sometimes be problematic >5600.

 

At least Newegg and Amazon have good return policies if needed in case you end up with Samsung.

 

 

 

 

So basically I have to purchase/RMA, purchase/ RMA until I receive the coveted hynix sticks I wish to have? 🤣

[cylix]

https://www.pcgamesn.com/intel/core-i9-14900k-specs-leak

[Mr. Fox]

1 hour ago, Raiderman said:

So basically I have to purchase/RMA, purchase/ RMA until I receive the coveted hynix sticks I wish to have? 🤣

You can usually tell by the part number on G.SKILL, but the speed is also a dead give-away in most cases. The rated XMP speed is a reflection of the capability of the memory IC. So, you won't see Micron or Samsung memory kits above 6000-6200 because that exceeds their normal functional capabilities. It is almost guaranteed that above 6200 is SK Hynix M-die or A-die.

Anything in the 6400-6800 range is SK Hynix, usually M-die up to 6400. Above 6400 usually Hynix A-die. Samsung can generally be speed rated up to about 6000. Below 6000 it could be anything and it really doesn't matter what it is because you're shopping for cheap parts and don't care about overclocking or performance tuning in the low-budget product category. There is that point of possible overlap in the 6000 realm. TeamGroup Delta is SK Hynix and they advertise that is the ony IC they utilize.

[Talon]

10 hours ago, seanwee said:

New laptop new shunt mod

https://www.3dmark.com/3dm/96618692
https://www.3dmark.com/pr/2431316

 

Nice! Anyway to get that CPU score up? Seems pretty low and holding back your overall hard. 

[tps3443]

18 minutes ago, Talon said:

 

Nice! Anyway to get that CPU score up? Seems pretty low and holding back your overall hard. 

I was impressed as well. It’s actually a little faster than my 3090 desktop. And to be a laptop, that is just crazy. Only that laptop has benefits of DLSS3, frame insertion etc. Not bad at all. I think the GPU score is really good because the CPU uses less of the available max system power limit. So on average that 4090 will be a little quicker in GPU bound scenarios than the Intel/4090 laptops.

[Talon]

14 minutes ago, tps3443 said:

I was impressed as well. It’s actually a little faster than my 3090 desktop. And to be a laptop, that is just crazy. Only that laptop has benefits of DLSS3, frame insertion etc. Not bad at all. I think the GPU score is really good because the CPU uses less of the available max system power limit. So on average that 4090 will be a little quicker in GPU bound scenarios than the Intel/4090 laptops.

 

No that GPU score is so high because it's shunt modded. It has absolutely nothing to do with stock power at all. All 4090 laptop will happily sit at 175w TGP which is the max allowed through any vendor (except tongfang chassis that use a built in shunt). All 4090 laptops will happily sit at 175w in games/benchmarks. 

 

This is the BEST RTX 4090 Laptop (shunted of course) and i9 13900HX. It curb stomps that score. Legion 7i Pro Gen 8 beast mode.

 

https://www.3dmark.com/spy/38610285

 

 

[electrosoft]

11 hours ago, Mr. Fox said:

If maxing out the frequency is not the objective or the platform prevents it, M-die is actually better because you can run it with tighter timings that what is possible with A-die. Especially tRFC (in the low and mid 400's) and tCKE (in low single digits, like 5 or 6).

 

The 7000 series by design is optimal around 6000 but some golden IMCs can do 6400-6600. tRFC on M-die does tend to cap around 400-450 albeit my best set of M-die's do 384. tCKE can do close to zero on AMD.

 

 

 

Every kit of M-die topped out at CL30-36-36 1.4v @ 6000 SR while these A-die do 6000 28-35-35.

 

My current AMD timings on A-die (I'm going to go back in and re-test TRFC):

 

 

 

 

4 hours ago, Raiderman said:

So basically I have to purchase/RMA, purchase/ RMA until I receive the coveted hynix sticks I wish to have? 🤣

 

3 hours ago, Mr. Fox said:

You can usually tell by the part number on G.SKILL, but the speed is also a dead give-away in most cases. The rated XMP speed is a reflection of the capability of the memory IC. So, you won't see Micron or Samsung memory kits above 6000-6200 because that exceeds their normal functional capabilities. It is almost guaranteed that above 6200 is SK Hynix M-die or A-die.

Anything in the 6400-6800 range is SK Hynix, usually M-die up to 6400. Above 6400 usually Hynix A-die. Samsung can generally be speed rated up to about 6000. Below 6000 it could be anything and it really doesn't matter what it is because you're shopping for cheap parts and don't care about overclocking or performance tuning in the low-budget product category. There is that point of possible overlap in the 6000 realm. TeamGroup Delta is SK Hynix and they advertise that is the ony IC they utilize.

 

Unfortunately this isn't true anymore as I saw a kit of G.skill 7200 M-die's from G.skill for sale a few weeks back which completely upended my range game. Some may be able to as I know my G.Skill M-die 6400 kit can boot to 7200 but error out on TM5.

 

@Raiderman I would suggest following @Mr. Fox advice and maybe get a kit of TeamGroup so you can lock in the Hynix component of the equation.

 

My best set of M-die's ( 2x16GB pair I bought last year with third party aRGB heatsinks) are sitting in the drawer at the moment. I ran those for weeks on my 7800X3D both last set of BIOSes no problem before upgrading to 64GB A-die. They are dialed in and tested at 6000 with a sheet of custom timings. If for some reason the set you order still fail or still get that A6 hang let me know as these are rock solid with no issues on my X670E Carbon. If my set hangs then you will definitely know it is your motherboard or issue elsewhere not the memory.

 

 

 

 

[Talon]

1 minute ago, electrosoft said:

 

The 7000 series by design is optimal around 6000 but some golden IMCs can do 6400-6600. tRFC on M-die does tend to cap around 400-450 albeit my best set of M-die's do 384. tCKE can do close to zero on AMD.

 

 

 

Every kit of M-die topped out at CL30-36-36 1.4v @ 6000 SR while these A-die do 6000 28-35-35.

 

My current AMD timings on A-die (I'm going to go back in and re-test TRFC):

 

 

 

 

 

 

Unfortunately this isn't true anymore as I saw a kit of G.skill 7200 M-die's from G.skill for sale a few weeks back which completely upended my range game. Some may be able to as I know my G.Skill M-die 6400 kit can boot to 7200 but error out on TM5.

 

@Raiderman I would suggest following @Mr. Fox advice and maybe get a kit of TeamGroup so you can lock in the Hynix component of the equation.

 

My best set of M-die's ( 2x16GB pair I bought last year with third party aRGB heatsinks) are sitting in the drawer at the moment. I ran those for weeks on my 7800X3D both last set of BIOSes no problem before upgrading to 64GB A-die. They are dialed in and tested at 6000 with a sheet of custom timings. If for some reason the set you order still fail or still get that A6 hang let me know as these are rock solid with no issues on my X670E Carbon. If my set hangs then you will definitely know it is your motherboard or issue elsewhere not the memory.

 

 

 

 

 

If you have a MC nearby, your best bet is to go into the store, grab a kit or ask the sales rep to see the kit your interested in and look at the model number on the box. GSkill identifies A-Die or M-Die or Samsung Dies right on the box. 

[electrosoft]

3 minutes ago, Talon said:

 

If you have a MC nearby, your best bet is to go into the store, grab a kit or ask the sales rep to see the kit your interested in and look at the model number on the box. GSkill identifies A-Die or M-Die or Samsung Dies right on the box. 

 

True, if @Raiderman has a local MC, he can cut out a lot of the legwork right then and there.

[Mr. Fox]

I do not believe there is a Microcenter in Boise, ID. So, Brother @Raiderman is in the same unfortunately situation I am in. No Microcenter in Phoenix, and none within even a slightly absurd drive distance. Closest locations to me are Tustin, CA and Dallas, TX.

 

Another good option (perhaps the BEST option) is to order naked green Hynix A-die sticks from AliExpress for less than half the cost of a retail kit, install aftermarket heat sinks (even a water block if you want to) and enjoy having some money left over, along with nearly guaranteed good overclocking sticks. The only potential downside is if you were hoping to take the lazy route and just set an XMP or EXPO profile and move on. That's not possible and performance tuning is manual.

 

The flip side of having no XMP or EXPO profiles is a proper Hynix A-die stick is already clocked to 5600 by default. You can add a little bit of voltage and push it to 6000 stable without touching any of the timings and leaving SPD default timings. So far that has worked for me consistently.  Adding an extra 400-600 MT/s to the memory clock has been effortless with A-die, with no need to unlock the PMIC for more voltage. From there all you need to do is make minor adjustments in small increments to timings, a few at a time, until you find where it becomes unstable, then back off a notch or two.

 

Side benefit here is you get memory that runs cooler without the trashy stock heating blankets, and no RGB software garbage to deal with if you're not content with looking at a never-ending rainbow puke light show. I don't know why it is so hard to find non-RGB memory kits. It seems like there are lots of people expressing their general displeasure with the rainbow puke, as it is a relatively common complaint among enthusiasts.

 

@Custom90gtdid you get your naked greens? How are they clocking for you, bro?

 

2 hours ago, electrosoft said:

The 7000 series by design is optimal around 6000 but some golden IMCs can do 6400-6600. tRFC on M-die does tend to cap around 400-450 albeit my best set of M-die's do 384. tCKE can do close to zero on AMD.

 

Every kit of M-die topped out at CL30-36-36 1.4v @ 6000 SR while these A-die do 6000 28-35-35.

 

My current AMD timings on A-die (I'm going to go back in and re-test TRFC)

If you are limiting memory clock to 6000 you can go a whole lot tighter on some of the timings, and run very low voltage, with Micron, Samsung or Hynix modules. I could run less than 300 tRFC with M-die clocked into the 6200-6400 range, but you need to be in the 400's with M-die and tRFC in the 600's and higher with A-die once you start pushing into the 7200 and higher speed realm.

[tps3443]

3 hours ago, Talon said:

 

No that GPU score is so high because it's shunt modded. It has absolutely nothing to do with stock power at all. All 4090 laptop will happily sit at 175w TGP which is the max allowed through any vendor (except tongfang chassis that use a built in shunt). All 4090 laptops will happily sit at 175w in games/benchmarks. 

 

This is the BEST RTX 4090 Laptop (shunted of course) and i9 13900HX. It curb stomps that score. Legion 7i Pro Gen 8 beast mode.

 

https://www.3dmark.com/spy/38610285

 

 

You know how all of these machines have a combined CPU and GPU max limit for both component loads? Some are 200 watts, others are 250 watts maybe. 
 

Well, I saw somewhere during testing with Battlefield 2042, and with the Ryzen 4090 machine it was straight whooping the Intel 4090 laptops. And this was because 4090 and Intel CPU was scaling back during gameplay because power consumption. The Intel chips were using much more power. While the Ryzen chip stayed clocked higher during gameplay, and it allowed the 4090 GPU to use more power too. 
 

I don’t think they can lock that 175 TGP watts during gameplay. Not even a GT77 can do that (Maybe games like Dying Lite 2 or low cpu power games). Because the CPU can hit just 60-70 watts in plenty of titles, and it starts throttling back the 4090 available power to the 130’s and 140’s range. It kind of sucks that they do this. But they just stay inside of that 230ish combined CPU/GPU power range. They have to power other components with the 330W brick like the panel, keyboard, trackpad, fans RGB lights, motherboard, storage, and other items. 

[Raiderman]

No MC here. The nearest is Portland Oregon, I believe. There is nothing that has AMD Expo that is above 6000mhz, that I see. So pick up some TG 6600mhz or something along those lines?

[Mr. Fox]

1 hour ago, Raiderman said:

No MC here. The nearest is Portland Oregon, I believe. There is nothing that has AMD Expo that is above 6000mhz, that I see. So pick up some TG 6600mhz or something along those lines?

That would work. But, you will need to research to see if they have EXPO profiles though. I have no idea if they do. If not, you would be looking at manual memory tuning if the Intel XMP profile won't work on AMD.

 

Edit: Looks like they do, but they are very specific about which ones support AMD EXPO. 

White: https://www.amazon.com/TEAMGROUP-T-Force-6000MHz-Optimized-FF8D532G6000HC38ADC01/dp/B0BRQ2NZTJ

Black: https://www.amazon.com/TEAMGROUP-T-Force-6000MHz-Optimized-FF8D532G6000HC38ADC01/dp/B0BRQ21XK1?th=1

NewEgg: https://www.newegg.com/team-32gb/p/N82E16820331989

 

The AMD EXPO product is called "Alpha" by TeamGroup.
https://www.teamgroupinc.com/en/product/delta-alpha-ddr5
On the AMD EXPO modules they don't specifically say they are SK Hynix. They say...

Quote

High-Quality ICs Selected for Stability & Reliability

TEAMGROUP has selected a high-quality IC that has undergone comprehensive compatibility and reliability testing to ensure a gaming memory that delivers both stability and compatibility.

They have Delta (RGB) and Vulcan (non-RGB) Alpha options.

 

https://www.teamgroupinc.com/en/products/compare/result/

 

Edited July 6, 2023 by Mr. Fox
added links

[Raiderman]

24 minutes ago, Mr. Fox said:

That would work. But, you will need to research to see if they have EXPO profiles though. I have no idea if they do. If not, you would be looking at manual memory tuning if the Intel XMP profile won't work on AMD.

 

Edit: Looks like they do, but they are very specific about which ones support AMD EXPO. 

White: https://www.amazon.com/TEAMGROUP-T-Force-6000MHz-Optimized-FF8D532G6000HC38ADC01/dp/B0BRQ2NZTJ

Black: https://www.amazon.com/TEAMGROUP-T-Force-6000MHz-Optimized-FF8D532G6000HC38ADC01/dp/B0BRQ21XK1?th=1

NewEgg: https://www.newegg.com/team-32gb/p/N82E16820331989

 

The AMD EXPO product is called "Alpha" by TeamGroup.
https://www.teamgroupinc.com/en/product/delta-alpha-ddr5
On the AMD EXPO modules they don't specifically say they are SK Hynix. They say...

They have Delta (RGB) and Vulcan (non-RGB) Alpha options.

 

https://www.teamgroupinc.com/en/products/compare/result/

 

Yeah, thats what I was saying. Anything that has AMD profiles/EXPO are not above 6000mhz. I bought the ram I have now without AMD EXPO, I think they are Intel XMP, as I didnt pay attention too much. So anything above 6000mhz will not have AMD EXPO, and like you said, I would be stuck with XMP, or manual tuning. So, I was wondering if I bought some ram that were AMD EXPO, they would function without the A6 hang.

[Mr. Fox]

19 minutes ago, Raiderman said:

Yeah, thats what I was saying. Anything that has AMD profiles/EXPO are not above 6000mhz. I bought the ram I have now without AMD EXPO, I think they are Intel XMP, as I didnt pay attention too much. So anything above 6000mhz will not have AMD EXPO, and like you said, I would be stuck with XMP, or manual tuning. So, I was wondering if I bought some ram that were AMD EXPO, they would function without the A6 hang.

Probably would work right. Looks like the price is sweet for $99 from Amazon. You could always grab them and if the are not great RMA them. They may be SK Hynix. But, if they run 6000 stable and do not have any issues with the A6 hang and perform well, it probably doesn't matter in the grand scheme of things.

 

Almost everything I have purchased in the TeamGroup brand has been good quality. I did have a memory kit that wasn't as good as some of the others (silicon lottery) from TeamGroup, but even that one was better than the G.SKUNK crap I returned. I have numerous SSDs, NVMe and USB flash drives from them and they have also been excellent.

[electrosoft]

@tps3443 did you see this post over on the OCN in regards to your hunt for another 13900ks?

 

 

 

[Mr. Fox]

Is it summer yet? Yup.

[Talon]

2 hours ago, tps3443 said:

You know how all of these machines have a combined CPU and GPU max limit for both component loads? Some are 200 watts, others are 250 watts maybe. 
 

Well, I saw somewhere during testing with Battlefield 2042, and with the Ryzen 4090 machine it was straight whooping the Intel 4090 laptops. And this was because 4090 and Intel CPU was scaling back during gameplay because power consumption. The Intel chips were using much more power. While the Ryzen chip stayed clocked higher during gameplay, and it allowed the 4090 GPU to use more power too. 
 

I don’t think they can lock that 175 TGP watts during gameplay. Not even a GT77 can do that (Maybe games like Dying Lite 2 or low cpu power games). Because the CPU can hit just 60-70 watts in plenty of titles, and it starts throttling back the 4090 available power to the 130’s and 140’s range. It kind of sucks that they do this. But they just stay inside of that 230ish combined CPU/GPU power range. They have to power other components with the 330W brick like the panel, keyboard, trackpad, fans RGB lights, motherboard, storage, and other items. 

Yes I’m aware of the combined/crossload power limits on these laptops. I have one. 
 

Got a link to that whooping you’re talking about? I’d be curious to see it. 1080p low settings to drive FPS in BF2042 I was seeing over 300fps on my 13900HX and 4080 laptop.
 

My 4090 likely does the same given it’s almost entirely CPU limited in that situation with the GPU pulling nowhere near max TGP. The crossload is under 240w CPU/GPU. Swap to 4K and you’ll be GPU TGP limited and nowhere near CPU limited.