@@visitante-pc5zc Because the memory controller is integrated into the same 4nm die as the cores. In the regular Ryzen 7000 lineup, memory controller sits in the 6nm cIOD. That's also why FCLK (Infinity fabric controller) reach 2400Mhz easily while it usually goes up to 2000-2100Mhz on Ryzen 7000x.
@@visitante-pc5zc Because DDR5 shows actual performance gains when using RAM with a higher frequency/higher data rate. You can gain upwards of 5% going from base speeds to 6000MHz, and being able to run 8000MHz or more is even more performance.
Skatterbencher overclocked both the CPU and IGPU, and he confirmed it loves faster ram. Perhaps some Patriot Viper Xtreme 5 PC5-65600 (8200MT/s) . Skatterbencher doubled his tomb raider score from stock when he oc'd the CPU and IGPU, and he only overclocked his ram to EXPO 7800 but said he wasnt sure what timings to adjust to to get better stability. So now we have more thermal headroom for even greater overclocking and tinker with ram timings on Hynix A die kits. Maybe we can get closer to matching a 6500 xt???
@@lucasrem Hah. That tool probably costs as much as a CNC mill. Which is to say, new cars are cheaper. On the other hand, a 3D printer that could actually use that level of accuracy would probably cost the same.
@@tr9036 Is the package power limit the same for cpu only vs gpu only vs mixed workloads? The gpu cores are also located on a different part of the die (obviously) so the heat density of the workloads should presumably be different. It isn't a given at all that the result would be the same.
@@Your_Paramour So, you know most of the things pretty detailly. I suggest you to not expect anything major.If cpu goes up to 93, gpu will go max 70-80 since it is just a simple gpu.
Or the frequency changes. When I load the iGPU on the 7600 when it's overclocked, even by 100Mhz, it locks it to ~2700 despite having a little temperature headroom. (Also ASUS's UEFI setting for GFX Core Offset doesn't seem to work, I have to use the AMD overclocking settings for voltage adjustment. CPU CO settings work fine.)
Now clock the living crap out of the gpu and put some 7200 ram and compare stock with a somewhat cpu bound game to the overclocked version. I firmly believe many good games will run well over 60 fps on that thing. Typical games i can see this for is PUBG, Wow, Tomb Raider, and even Cyberpunk, but it would be amazing to see how it does with Tarkov.
Those are some FANTASTIC RESULTS!!! What a HUGE IMPROVEMENT!!!! WOW!!! And I just LOVE your new potical toy!!! I hope you show more of it when Zen 5 comes out and you design the de-lidder and everything... I LOVE seeing the process!!!
I would like to see an overclock on the integrated graphics using liquid metal and the AIO directly on the DIE to know which dedicated GPU it can replace in extreme situations. If there is a channel that can do this, believe it is yours!
16:30 I got mine recently and Nuctua instructions daid for amd 5 you need to munt the cooler offset due obvious design of the chip,but for the 8700g is not needed I think. So 1 thing we learn,even other manufacturers could benefit. So thanks!
So I finally got round to delidding my 8700g in A09 case with Thermalright AXP120-67 cooler. Prior to delidding it was running at 95C under aida extreme stress test and 42.2C at idle (fan speed 1300rpm) Post liquid metal it's running at 78.4C with CPU fan at full speed and 83C with fan at 1500rpm. 37C idling. I used conductonaut extreme for cpu/IHS and kryonaut extreme for IHS/Cooler and delidded with ryzen 7000 delid-die-mate from Thermal Grizzly. Decided to go with kapton tape method to protect rest of processor. Extremely happy with result 😊 Thanks Roman
I actually just bought this processor for a compact build. I would love it if you made an improved heat spreader that is compatible with the 8700g. From that i can tell all it would take is a bump in the middle like on the ihs. Im really glad i saw this though because i was going to buy your original improved heat spreader Mondy.
You don't need to buy a delidder or anything. Use a piece of dental floss to cut the glue. Do each leg individually and it only takes 10 min. Completely safe. Then put some nail polish on the SMDs and apply liquid metal. You'll get the same results as shown here.
After a few years use, my 2400G was becoming horrible to cool. I used your previous delidding video for that CPU and it completely changed the CPU. Better than new temperatures, even moving it to a smaller case/smaller cooler afterwards. It’s disappointing that AMD don’t solder these, as I never want to NEED to delid again. I thought that went away with Skylake. :(
Great video, really enjoyed! Also Thanks Roman for never failing to have something entertaining to watch, when im bored on sundays. Have a nice Weekend!
Awesome video & test, thank you! Currently using a 5700G for a DAW, so the 8700G obviously has my interest. I definately would be interested in a liquid metal version with the performance gains shown!
16:32 "why are we even doing this?" honestly, it's the best performance increase that somebody on an AMD platform would see... unless I'm not informed and AMD has switched back to using paste on their CPU's also, it's great content. I love seeing when liquid metal makes a difference on pasted CPU's.
* on an actual AMD platform Because pretty sure a vintage build with delidding, watercooling and LM can easily achieve more perf increase compared to stock.
@@PainterVierax in the "old" days when an amd 450mhz was the top of the line cpu, it could be clocked stable to 900+mhz with watercooling. Try that with any new cpu today
Hardly understand any of what you’re talking about, but I love how approachable you make everything. Very very cool channel. Thanks for your input into the English community!
Man uses stupidly expensive measuring equipment to verify his company's products are compatible with a part... More seriously, it was fun and I wish I had those cool toys, I mean tools. I'll give some basics in case you or anyone else isn't aware. With the caveat that my degree focused on the electronics and low level programming side. Thermal transfer is more of a Mechanical Engineer's domain. The goal is to move heat away from the chip as efficiently as possible. Three things affect that, chip temperate, temperature of the thing the heat is moving to, how well the stuff between those two conducts heat. Which is further affected by two things, how much stuff there is and it's "thermal conductivity". More or less, the chip's temperature is relatively fixed. A perfectly efficient normal cooler would mean the other side is room temperature. Since he's using the same cooler for both tests that's also fixed here. Which means playing with that 3rd variable. Stupid example, if you put a square of fabric cut from an oven mitt, the heat wouldn't move from the chip to the cooler that well. Both because it would be thick and because fabric does not conduct heat well. Same thing goes with air. Any air gap means you have an insulator. The thermal paste that came on the chip had a thickness and liquid metal is thinner, so he was measuring to make sure there would not be an air gap. The reasons those two products are better than stock is because the Cryosheet has a higher thermal conductivity. Meanwhile, liquid metal is both thinner* and has a higher thermal conductivity. To be honest, the Cryosheet result was the craziest part to me. Mostly because that's the type of thing that could be easily done by AMD at the factory for < $10 per chip to improve performance by a meaningful margin. * Manufacturing tolerances and material selection do mean AMD isn't deliberately making things thick just to decrease performance.
I would rather concentrate on OC'ing the iGPU part of the Ryzen 8000. That is the sole purpose of its existence, and of course, monolithic AMD die always comes with a higher FCLK and memory clock limit.
And you can totally push it to ~5.3Ghz with a tweaked base clock and good cooling, although it's not got the best silicon quality so cores can use a lot of power. Impressed with what can be done with a 2 CU RDNA2 iGPU as well, although it's unfortunate that OCing it can seemingly limit the CPU core frequency.
fitztech Most people do 65 Watt CPU now, the 14900 H/HX in all of these hardcore for gaming systems is the best. Need a nerdy tweak system, for for these 14900 K CPU's, make it yourself at home ?
Great info and fun to watch, Roman! :) With liquid metal and your skills, I'm very interested to see both CPU and iGPU max. performance under simultaneous load, i.e. mimicking the heaviest CPU and graphics demanding games.
Please try to OC the igpu and ram like Skatterbench do too. I am highly interested in what coolingcould bring to the table for gfx oc. Maybe 3.2-3.3ghz?
It's an APU, thus the 8000. The APUs are what set the full cadence, although the AMx socketable "mid-gens" (relative to Ryzen desktop _CPUs)_ don't always release to retail market. The _CPUs_ only follow every other gen in that cadence. (Except 2000-series, because that was basically a refresh).
I run an 8700g in an Asrock Deskmini x600, just the ten degree drop from the kryosheet makes doing this mod worth it. The 25 degree drop from Liquid Metal is massive for the Deskmini. Thinking I’m going to split the difference and delid mine and add some PTM7950. Thanks for putting in the work Roman!!
From the CES videos, I remember seeing AMD showcasing I think baulders gate 3 and starfield running at fairly low settings... But both were running at the same time, with no perceivable lag (at least from the video I saw). Quite exciting seeing an APU manage that. :)
The KryoSheet looks (and must be, as it compresses) porous. If so, I wonder if Liquid Metal (or other thermal paste) couldd be 'rubbed-in' to the sheet to further reduce any tiny air-spaces in the graphene structure?
@@PlayingItWrong do you know that pretty much all liquids have surface tension, I would go as far as to say that LM has higher surface tension than water.
Hey Roman have you tried cooling the glue of the IHS down with ice spray or something to a brittle state before delidding? It might make it much easier with elasticity of the glue mostly gone.
8:02 It seems like the IHS is not contacting the die properly? How could you fix that? And yes please design a direct die mount for these CPUs, it would be amazing!
7940HS at 80W on the Zephyrus G14 can crack 18k on CBR23 Multi-core and with undervolting it can crack 19k. I'm surprised the 8700G is scoring under 17k at 85W?? The 7840HS at 80W can do 17k as well.
I just compared the benchmarks of my son's 7940HS inside of a Mini PC with the ones posted online for 8700G. The latest has a maximum 10% increase in performance, mostly due to higher TDP and the BIOS being more flexible for overclocking. But my Mini PC, equipped with 32GB RAM and 1TB SSD costed a little bit more than the actual price of an 8700G. I found this video very useful for the ITX enthusiasts, but I will stick a while with my overclocked and undervolted 5700G and my son's 7940HS.
Binning, my dude. Just because two chips are made to the same spec, doesn't mean they're "the same". With the way semiconductor fabrication works, you end up with a bunch of chips of which only some are of the "perfect" quality, others might not be able to clock high enough at a given powerlimit, or have some of their cores just not work at all, that's why SKUs exist, even though they all use "the same" die. The only reason these desktop APUs even exist is probably that AMD found themselves with a bunch of dies that were fully functional, but not at the kind of powerlimit suitable for a laptop.
If the APU pricing made sense (it doesn't) I'd be really tempted to delid one of these, use a kryosheet and some really high speed memory to build the highest performance tiny box possible.
That's kind of sad. 2000-series APU (Raven Ridge) used paste, but 3000 (Picasso), 4000 (Renoir), and 5000 (Cezanne) APUs used solder. And now 8000-series is back to paste....
I have toyed with having one of these in a smaller PC, while I would probably just stream games to it via steam I do like the idea of it being able to run some of the older games I might play and it makes more sense than a console at least in the use case I'm thinking of. You kind of covered the reasoning though :P
I'd be interested in seeing what this mod does for the IGPU, since that is the big selling point on desktop for this chip. Also, I'm not sure the Amd naming scheme is Deceptive, their following a pattern. the 4000 series was oem/integrated chip and based on the 3000's series architecture, the 6000 series was oem/integrated chip and based on the 5000 series architecture, so the 8000 series is a.... well, you see my point.
Being more generous about the liquid metal maintenance factor: My personal experience showed inevitable efficiency degradation in six months to the level of efficiency closer to MX-4 thermal paste and a significant dip below that in the span of a year. I suspect the pumping effect to be the cause, because the rate of degradation roughly corresponds to the temperature gradient you achieve. This effect gets so dramatic with liquid metal because over a thermal cycle it's going through a phase transition to some extent: the chip and the heat exchanger heat up, the gap between them closes in, the compound gets more liquid and gets pished out. Then, during the cooling phase, the gap between the chip and the heat exchanger opens back up, but the compound also cools down, becoms less runny, and the capillary effect can't get all of it back in. This way, bit by bit, tiny fractions of your liquid metal escape and are replaced by the atmosphere air (wich is quite a good insulator). To put it simply: the hotter your CPU/GPU runs under load - the sooner this thing will escape due to the heat cycles expanding and shrinking the gap it's applied to. As a consequence, the more this thing degrades (and the hotter the load cycles get) - the faster it's degradation rate gets. Sure it's good when served fresh, but I personally would never return to that three months wonder. Especially factoring in it's cost and dubiousity of application, that gets amplified 10fold when you have to do the cleanup process from the previous batch.
I wonder how iGPU overclocking will behave with this thermal headroom 🤔. JayzTwoCents managed 3.045 Ghz with the 7900XTX. Considering that the 8700G has only 12 CUs (72 less than the 7900XTX) under an 178mm die, maybe 3.3 Ghz ?
DDR5 is so insanely fast that nobody feels the need to decouple IF clock from RAM clock anymore. Also, AMD's process has been refined to the point that they can comfortably make monolithic CPU dies without much trouble with yield. Their chiplet tech is now re-oriented towards adding some other features like integrated graphics or AI accelerators or other add-on silicon that adds different functionality. These days you won't see CPU cores in chiplets anymore unless you go Ryzen 9 or Threadripper/Epyc.
Most ryzen chips can do 6200 in 1:1 mode with the latest agesa, some cpus can do 6400 without going crazy on vsoc. The best speed is the max speed your chip can do in 1:1 mode. Fclk you just push as high as you can before you start losing performance.
I have had a Ryzen 7 3700X for about three years, and I plan to wait for Zen 6; I figure that a similar position in the model line would double my core count with at least an additional gigahertz of base clock speed and a meaningful IPC increase. If performance becomes a problem in the meantime, then I will give my BeQuiet Dark Rock Pro 4 a third fan (likely replacing the front fan at the same time) and enable PBO.
Optical measurement from keyence is super cool. I was sceptical of liquid metal because of high ambient temperature of 30-40C with humidity >90% in the sub-continent. Do you think laptops and handheld device with LM might leak or slip out even at rest? The sheet are quite easy to apply and not require frequent maintenance.
I'm running a 5700g in an Asrock DeskMini system, and might be interested in this chip if Asrock decides to make an AM5 version. I do typically overclock as much as I can, but push the GPU side the hardest. I would like to see that kind of testing done, but understand why it's not popular. I would be interested in seeing a lapped version tested, maybe in 3dMark to push the iGPU side.
0:33 I disagree with you on that, the AMD nomenclature is completely transparent if you stop trying to "assume" and actually go and check their naming scheme. I don't like it either, but they had to make it like that eventually because the sheer amount of products they can have at the same time in the market, usually spanning several families, lithographies and variants.
Derbauer, question, the 3D analyzer said x:y 1078 and 1011 and 1389 and 1231 (mils) question divide by 100 correct? Approximately 170 mm2 of die area. Question what is the square mm area of the 8700 die? mb
Very impressed to see 20k points from the 8700g in R23. That's how my i7 13700 performed with the stock intel cooler. (On a good cooler it can score 29000 points)
17:40 oh crap I’ve remounted mine twice already! Once when switching from a 5800x3d to a 7800x3d and a second time when switching hung from soft to hardline. My temps are still really good though!
Very intreasting Video It would be interesting to see just how far you can push the chip. I think these are for small portable systems, say small pc for car motorhome coach or plane where space and power available may be limited.
Unless the game is old enough to run the whole vram budget in the 3d cache, pretty good. Anything that uses over 100mb of vram irrelavent it will still hit the slow ddr5.
Contact frames or IHS definetly don't make commercial sense given that this CPU is already more of a niche use case, so it would be trying to sell to the niche of the niche. That being said it would be fun to just grab an old prototype of contact frame and adapt it to see how far you can push this thing.
It'd be cool to to see the igpu overclock as well and some balancing of it. Like i assume high overclock of cpu wouldn't be possible with high gpu overclock, but fiddling around which could be somewhat a sweetspot would be interesting to see. Maybe only the igpu overclock would be the best solution? idk
On the 7600 overclocking the iGPU at all seemingly leads to mandatory ~2.7Ghz clock-limiting of the CPU all-core workload under simultaneous iGPU load (even if headroom exists). If temperature exceeds 95C then the CPU speed is sacrificed further in preference to GPU performance.
I'm not seeing PBO2 tuning where you give it a negative offset. Wouldn't that also be able to get the same results, if you're lucky with the ASIC quality.
The issue I have with forcing all core OCing is that they stay at that frequency all the time. or at least it used to be like that when I tried on my 5600x. Seems like a huge power waste to run that all the time for the odd moments you need max CPU power. Too bad they don't let you adjust the PBO settings to boost higher with a low voltage.
Please do a test with Igpu performance. I might jump to an 8700g from my 5600+A2000 chip just for fun. De-lid and see how much performance I can get in a 5L case and just IGPU :)
I can see the 8700g to be useful for Linux users, especially those that use virtual machines often. Having a powerful APU opens up possibilities for better high res recording as well as keeping the host system on that GPU if you plan on GPU forwarding. Definitely a cool processor in my opinion.
I would love to see you redo this with a Ryzen AI 9 370 HX. I'm not sure what motherboards will support this, but it performs well without overclocking.
I would like to see a direct die frame for the 8700G because even though this is a year old now, the GPU performance, and specifically, the ability to allocate up to 64GB to the GPU should make this an excellent choice for running LLMs
You ought to give the iGPU a shot at potential overclocking. Maybe find some ram that can clock relatively high? See where the 780M iGPU lands in the dedicated GPU space.
@DeBauer I'd like to see what this modified APU can do, I 'd like to see Actually Hardcore Overclocking experiment further. perhaps Ancient Gameplays. Perhaps people that optomised (fclock?) with the previous generation (5700G, 5600G) to maximize performance. I would really like to see what this can do without an discreet GPU, both gaming and more professional use cases, such as video rendering, transcoding, or CAD/CAM use cases. Thank you for reading, sir.
Ok, that optical height measurement tool is sick.
AMD is also going to intel's path of shiiiity TIM
"Oh, Dear can I get an optical height measurement tool for my computer?"
@@DIYTech21 AMD is dead, they failed.
Intel is still producing CPU's !
German engineering is dead too !
@@lucasremhuh
@@lucasrem bro is mad that AMD is better than intel.
I love how your cat @6:05 looks exhausted from all the testing and overcloking he had to make.
Now put ddr5-8000 and OC the iGPU. The memory controller on the ryzen APUs tend to OC much higher.
100% this
Why
Some people hit 10000-11000
@@visitante-pc5zc Because the memory controller is integrated into the same 4nm die as the cores.
In the regular Ryzen 7000 lineup, memory controller sits in the 6nm cIOD.
That's also why FCLK (Infinity fabric controller) reach 2400Mhz easily while it usually goes up to 2000-2100Mhz on Ryzen 7000x.
@@visitante-pc5zc Because DDR5 shows actual performance gains when using RAM with a higher frequency/higher data rate. You can gain upwards of 5% going from base speeds to 6000MHz, and being able to run 8000MHz or more is even more performance.
Would of been good to see the graphics improvements from that aswell.
"Would *have..."
@@tomppeli.woulda….
@@tomppeli. waddle
I was waiting for such a video from a Buildzoid, but not sure if he's actually going to do something like that
Skatterbencher overclocked both the CPU and IGPU, and he confirmed it loves faster ram. Perhaps some Patriot Viper Xtreme 5 PC5-65600 (8200MT/s) . Skatterbencher doubled his tomb raider score from stock when he oc'd the CPU and IGPU, and he only overclocked his ram to EXPO 7800 but said he wasnt sure what timings to adjust to to get better stability. So now we have more thermal headroom for even greater overclocking and tinker with ram timings on Hynix A die kits. Maybe we can get closer to matching a 6500 xt???
That 3D profiler is so cool.
You need a measuring tool for the 3D printer ?
Germans are cool, need that on social ? Klates menschen
@@lucasrem Hah. That tool probably costs as much as a CNC mill. Which is to say, new cars are cheaper. On the other hand, a 3D printer that could actually use that level of accuracy would probably cost the same.
Would be interesting to see the temp changes when using the igpu as well.
Everything is in same die. What do you think it will be?
@@tr9036 Is the package power limit the same for cpu only vs gpu only vs mixed workloads? The gpu cores are also located on a different part of the die (obviously) so the heat density of the workloads should presumably be different. It isn't a given at all that the result would be the same.
@@Your_Paramour So, you know most of the things pretty detailly. I suggest you to not expect anything major.If cpu goes up to 93, gpu will go max 70-80 since it is just a simple gpu.
Or the frequency changes. When I load the iGPU on the 7600 when it's overclocked, even by 100Mhz, it locks it to ~2700 despite having a little temperature headroom. (Also ASUS's UEFI setting for GFX Core Offset doesn't seem to work, I have to use the AMD overclocking settings for voltage adjustment. CPU CO settings work fine.)
i was wondering why he didnt test it, as an improvement in the igpu could mean this thing is viable to run game even better
Now clock the living crap out of the gpu and put some 7200 ram and compare stock with a somewhat cpu bound game to the overclocked version. I firmly believe many good games will run well over 60 fps on that thing. Typical games i can see this for is PUBG, Wow, Tomb Raider, and even Cyberpunk, but it would be amazing to see how it does with Tarkov.
Maybe can scope in on streets without losing frames XD
I am more interested in improving 1% low fps in games like CS2 to avoid those laggspikes when playing on competitive levels.
7200 is child's play. These chips are setting records for DDR5. Already 10600 has been done at ambient.
DDR5 10 000 MT/S+++ and you've got GTX 1070 performance :-D
@@alexskywalker888on intel
Those are some FANTASTIC RESULTS!!! What a HUGE IMPROVEMENT!!!! WOW!!!
And I just LOVE your new potical toy!!! I hope you show more of it when Zen 5 comes out and you design the de-lidder and everything... I LOVE seeing the process!!!
I would like to see an overclock on the integrated graphics using liquid metal and the AIO directly on the DIE to know which dedicated GPU it can replace in extreme situations. If there is a channel that can do this, believe it is yours!
The analytical analysis' this channel does is freaking amazing. I love the testing this channel does. Keep it up!
16:30 I got mine recently and Nuctua instructions daid for amd 5 you need to munt the cooler offset due obvious design of the chip,but for the 8700g is not needed I think. So 1 thing we learn,even other manufacturers could benefit. So thanks!
@8:30 the 8700G top left reminds me of Duron/Athlon days
like the black 'pads'?
@@Enozenim_LJOholy shit yea, looks like s462 cpus
@@00zero557Aikr super cool coincidence
and the colour! @@Enozenim_LJO
The black pads on the Durons. Lol, yep, I feel that bro.
So I finally got round to delidding my 8700g in A09 case with Thermalright AXP120-67 cooler.
Prior to delidding it was running at 95C under aida extreme stress test and 42.2C at idle (fan speed 1300rpm)
Post liquid metal it's running at 78.4C with CPU fan at full speed and 83C with fan at 1500rpm. 37C idling.
I used conductonaut extreme for cpu/IHS and kryonaut extreme for IHS/Cooler and delidded with ryzen 7000 delid-die-mate from Thermal Grizzly.
Decided to go with kapton tape method to protect rest of processor.
Extremely happy with result 😊
Thanks Roman
Haha, sustained 5.3GHz on all cores and sub 80*C under load is insane! What a beast.
Makes me excited for ryzen 9000 and what it’s capable of
Why
Check out intel.
Yes _but_ only 16MB L3 cache and PCIe 4.x so you're limited in other ways, will have to push RAM and IF speed for it.
just 110W is as well as impressive.
I actually just bought this processor for a compact build. I would love it if you made an improved heat spreader that is compatible with the 8700g. From that i can tell all it would take is a bump in the middle like on the ihs. Im really glad i saw this though because i was going to buy your original improved heat spreader Mondy.
You don't need to buy a delidder or anything. Use a piece of dental floss to cut the glue. Do each leg individually and it only takes 10 min. Completely safe. Then put some nail polish on the SMDs and apply liquid metal. You'll get the same results as shown here.
Thank you so much for the strobe warning. My epileptic ass appreciates it.
After a few years use, my 2400G was becoming horrible to cool. I used your previous delidding video for that CPU and it completely changed the CPU. Better than new temperatures, even moving it to a smaller case/smaller cooler afterwards.
It’s disappointing that AMD don’t solder these, as I never want to NEED to delid again. I thought that went away with Skylake. :(
Great video, really enjoyed!
Also Thanks Roman for never failing to have something entertaining to watch, when im bored on sundays.
Have a nice Weekend!
Thanks for testing this cpu, wana to upgrade 3 systems with 5800x , a bit work soon , chance cpu / MB / memory...
what a showcase, so informative and enjoyable😄
I love these types of videos as you do it just for fun and to see what happens. Keep doing more of these.
Man this was so exciting to watch thanks for the great vid!
Awesome video & test, thank you! Currently using a 5700G for a DAW, so the 8700G obviously has my interest. I definately would be interested in a liquid metal version with the performance gains shown!
Can recommend upgrading. I also had a 5700G, now a 8700G, vast improvement!
Man, you brought dealing to a whole other level, kudos to you! Keep going!
16:32 "why are we even doing this?" honestly, it's the best performance increase that somebody on an AMD platform would see... unless I'm not informed and AMD has switched back to using paste on their CPU's
also, it's great content. I love seeing when liquid metal makes a difference on pasted CPU's.
* on an actual AMD platform
Because pretty sure a vintage build with delidding, watercooling and LM can easily achieve more perf increase compared to stock.
@@PainterVierax in the "old" days when an amd 450mhz was the top of the line cpu, it could be clocked stable to 900+mhz with watercooling. Try that with any new cpu today
@@scaniav8530able I know, that's why I'm saying! It's crazy how nowadays people spend a lot more extra money for a lot less perf gains.
@@PainterVierax I know. That's why I rather run mostly stock and paired to only what games I play now
Hardly understand any of what you’re talking about, but I love how approachable you make everything. Very very cool channel. Thanks for your input into the English community!
Man uses stupidly expensive measuring equipment to verify his company's products are compatible with a part... More seriously, it was fun and I wish I had those cool toys, I mean tools.
I'll give some basics in case you or anyone else isn't aware. With the caveat that my degree focused on the electronics and low level programming side. Thermal transfer is more of a Mechanical Engineer's domain.
The goal is to move heat away from the chip as efficiently as possible. Three things affect that, chip temperate, temperature of the thing the heat is moving to, how well the stuff between those two conducts heat. Which is further affected by two things, how much stuff there is and it's "thermal conductivity". More or less, the chip's temperature is relatively fixed. A perfectly efficient normal cooler would mean the other side is room temperature. Since he's using the same cooler for both tests that's also fixed here. Which means playing with that 3rd variable.
Stupid example, if you put a square of fabric cut from an oven mitt, the heat wouldn't move from the chip to the cooler that well. Both because it would be thick and because fabric does not conduct heat well. Same thing goes with air. Any air gap means you have an insulator. The thermal paste that came on the chip had a thickness and liquid metal is thinner, so he was measuring to make sure there would not be an air gap.
The reasons those two products are better than stock is because the Cryosheet has a higher thermal conductivity. Meanwhile, liquid metal is both thinner* and has a higher thermal conductivity. To be honest, the Cryosheet result was the craziest part to me. Mostly because that's the type of thing that could be easily done by AMD at the factory for < $10 per chip to improve performance by a meaningful margin.
* Manufacturing tolerances and material selection do mean AMD isn't deliberately making things thick just to decrease performance.
It be interesting to see what kind of improvements it could do in gaming vs stock. Cool video!
I would rather concentrate on OC'ing the iGPU part of the Ryzen 8000. That is the sole purpose of its existence, and of course, monolithic AMD die always comes with a higher FCLK and memory clock limit.
Roman you are the best! I haven't seen YT feed me your channel in a few months so glad they fixed that
Looks sweet. I’m still loving my 7600 non - x. The low power draw and low temps while still providing performance continues to impress me.
The 7600 and 7600X are great intro AM5 products.
And you can totally push it to ~5.3Ghz with a tweaked base clock and good cooling, although it's not got the best silicon quality so cores can use a lot of power. Impressed with what can be done with a 2 CU RDNA2 iGPU as well, although it's unfortunate that OCing it can seemingly limit the CPU core frequency.
fitztech
Most people do 65 Watt CPU now, the 14900 H/HX in all of these hardcore for gaming systems is the best.
Need a nerdy tweak system, for for these 14900 K CPU's, make it yourself at home ?
@@lucasrem what is bro doing?
Great info and fun to watch, Roman! :) With liquid metal and your skills, I'm very interested to see both CPU and iGPU max. performance under simultaneous load, i.e. mimicking the heaviest CPU and graphics demanding games.
Saw the -25° in the thumbnail and was looking forward to seeing some LN2 action 🤦🤣
i bought from you an i7 8700k delidded plus liquid metal, still going strong after 5 years. Never needed to reapply the LM, using it daily.
would also be interesting to see igpu overclocking. also ram speeds is an interesting topic with those APUs) a lot of possible content actually
Please try to OC the igpu and ram like Skatterbench do too. I am highly interested in what coolingcould bring to the table for gfx oc. Maybe 3.2-3.3ghz?
You made this very interesting. Thank you.
Hands down best hardware channel on tube! Great stuff!
It's an APU, thus the 8000. The APUs are what set the full cadence, although the AMx socketable "mid-gens" (relative to Ryzen desktop _CPUs)_ don't always release to retail market. The _CPUs_ only follow every other gen in that cadence. (Except 2000-series, because that was basically a refresh).
Awesome..... I want to see more stuff to deep dive in hardwares technology and new testing mechins.... 🔥
I run an 8700g in an Asrock Deskmini x600, just the ten degree drop from the kryosheet makes doing this mod worth it. The 25 degree drop from Liquid Metal is massive for the Deskmini. Thinking I’m going to split the difference and delid mine and add some PTM7950. Thanks for putting in the work Roman!!
Definitely going to help some PC user down the road. I see those questions on forums asking for help to run some software on their older PC.
Great video, thanks a lot!
great video sir, love watching your reviews, keep up great work
I love these kinds of videos, thanks Roman!
From the CES videos, I remember seeing AMD showcasing I think baulders gate 3 and starfield running at fairly low settings... But both were running at the same time, with no perceivable lag (at least from the video I saw). Quite exciting seeing an APU manage that. :)
That was an interesting video. That scanner is mind-blowing!
The KryoSheet looks (and must be, as it compresses) porous. If so, I wonder if Liquid Metal (or other thermal paste) couldd be 'rubbed-in' to the sheet to further reduce any tiny air-spaces in the graphene structure?
What about surface tension.😊
@@panjak323Half Life
@@panjak323it's not water, but some materials pair with liquid metals well because of the ions involved. Easiest way to find out is to try 😅
pretty sure LM would just destroy the structure of the sheet. kryosheets if you haven't got one, are extremely fragile.
@@PlayingItWrong do you know that pretty much all liquids have surface tension, I would go as far as to say that LM has higher surface tension than water.
At 14:11 there is a stream of smoke/vapor. What was that from @der8auer-en?
Hey Roman have you tried cooling the glue of the IHS down with ice spray or something to a brittle state before delidding? It might make it much easier with elasticity of the glue mostly gone.
WOW!!! the tools for the job... 3d profiler is amazing !!!
8:02 It seems like the IHS is not contacting the die properly? How could you fix that? And yes please design a direct die mount for these CPUs, it would be amazing!
That optical scanner is so cool. Could you scan gaming mice with it to create a public library of 3D scans for gamers to decide on their next mouse?
Love how the cat just chills there! 😺😺
Great video thanks Der8auer
7940HS at 80W on the Zephyrus G14 can crack 18k on CBR23 Multi-core and with undervolting it can crack 19k. I'm surprised the 8700G is scoring under 17k at 85W?? The 7840HS at 80W can do 17k as well.
And? They're the same silicon
laptops get better silicon
I just compared the benchmarks of my son's 7940HS inside of a Mini PC with the ones posted online for 8700G. The latest has a maximum 10% increase in performance, mostly due to higher TDP and the BIOS being more flexible for overclocking. But my Mini PC, equipped with 32GB RAM and 1TB SSD costed a little bit more than the actual price of an 8700G.
I found this video very useful for the ITX enthusiasts, but I will stick a while with my overclocked and undervolted 5700G and my son's 7940HS.
Better bins for laptops because of the power and heat envelope is so tight.
Binning, my dude. Just because two chips are made to the same spec, doesn't mean they're "the same". With the way semiconductor fabrication works, you end up with a bunch of chips of which only some are of the "perfect" quality, others might not be able to clock high enough at a given powerlimit, or have some of their cores just not work at all, that's why SKUs exist, even though they all use "the same" die. The only reason these desktop APUs even exist is probably that AMD found themselves with a bunch of dies that were fully functional, but not at the kind of powerlimit suitable for a laptop.
If the APU pricing made sense (it doesn't) I'd be really tempted to delid one of these, use a kryosheet and some really high speed memory to build the highest performance tiny box possible.
Great improvement really, nice video.
That's kind of sad.
2000-series APU (Raven Ridge) used paste, but 3000 (Picasso), 4000 (Renoir), and 5000 (Cezanne) APUs used solder.
And now 8000-series is back to paste....
Especially disappointing since it's hardly the cheapest of chips.
That was pretty awesome. Scored about the same as my old 5900X
"Why are we even doing this on a 8700g? I dont know, because I like it."
Hell yeah, it's called passion
I have toyed with having one of these in a smaller PC, while I would probably just stream games to it via steam I do like the idea of it being able to run some of the older games I might play and it makes more sense than a console at least in the use case I'm thinking of. You kind of covered the reasoning though :P
I'd be interested in seeing what this mod does for the IGPU, since that is the big selling point on desktop for this chip. Also, I'm not sure the Amd naming scheme is Deceptive, their following a pattern. the 4000 series was oem/integrated chip and based on the 3000's series architecture, the 6000 series was oem/integrated chip and based on the 5000 series architecture, so the 8000 series is a.... well, you see my point.
Love your tools / toys.
Loved the video. It was worth it. It's a laptop product from AMD, so it's always interesting to see how it behaves unlimited on desktop.
Super cool video as always.
I really want to see how much you can push the iGPU on this thing with deliding and liquid metal !
Being more generous about the liquid metal maintenance factor:
My personal experience showed inevitable efficiency degradation in six months to the level of efficiency closer to MX-4 thermal paste and a significant dip below that in the span of a year.
I suspect the pumping effect to be the cause, because the rate of degradation roughly corresponds to the temperature gradient you achieve.
This effect gets so dramatic with liquid metal because over a thermal cycle it's going through a phase transition to some extent: the chip and the heat exchanger heat up, the gap between them closes in, the compound gets more liquid and gets pished out.
Then, during the cooling phase, the gap between the chip and the heat exchanger opens back up, but the compound also cools down, becoms less runny, and the capillary effect can't get all of it back in.
This way, bit by bit, tiny fractions of your liquid metal escape and are replaced by the atmosphere air (wich is quite a good insulator).
To put it simply: the hotter your CPU/GPU runs under load - the sooner this thing will escape due to the heat cycles expanding and shrinking the gap it's applied to. As a consequence, the more this thing degrades (and the hotter the load cycles get) - the faster it's degradation rate gets.
Sure it's good when served fresh, but I personally would never return to that three months wonder. Especially factoring in it's cost and dubiousity of application, that gets amplified 10fold when you have to do the cleanup process from the previous batch.
Hi der8! Great Vid 🙂
awesome results! makes me wonder how much better gpu performance you would get from this.
I wonder how iGPU overclocking will behave with this thermal headroom 🤔.
JayzTwoCents managed 3.045 Ghz with the 7900XTX.
Considering that the 8700G has only 12 CUs (72 less than the 7900XTX) under an 178mm die, maybe 3.3 Ghz ?
Pieter/SkatterBencher got 3.3 GHz on his 8700G.
@@Wasmachineman wow, didn't know that... Then maybe 3.5Ghz is possible
That moment when you realize you used to game on a 300Mhz PC.
Our first PC experience was the MSX Sakhr 3,58 MHz
Very fun video. Cool what you can coax out of a mobile chip with a good heat solution.
why is nobody talking about infinity fabric and expo and ram overclocking anymore? :D is the 6000mhz still the best to get for am5 cpus?
The next Ryzen 8000 desktop cpu's will support up to 6400mhz i think
DDR5 is so insanely fast that nobody feels the need to decouple IF clock from RAM clock anymore. Also, AMD's process has been refined to the point that they can comfortably make monolithic CPU dies without much trouble with yield.
Their chiplet tech is now re-oriented towards adding some other features like integrated graphics or AI accelerators or other add-on silicon that adds different functionality.
These days you won't see CPU cores in chiplets anymore unless you go Ryzen 9 or Threadripper/Epyc.
Most ryzen chips can do 6200 in 1:1 mode with the latest agesa, some cpus can do 6400 without going crazy on vsoc. The best speed is the max speed your chip can do in 1:1 mode. Fclk you just push as high as you can before you start losing performance.
@@Goodmanperson55 For the CPU side it's best to keep the Infinity fabric tied, but for the iGPU the faster the ram the better it is
infinity fabric on 8000G can go up to 2400 and ram 6400 is the max sweet spot i guess
I have had a Ryzen 7 3700X for about three years, and I plan to wait for Zen 6; I figure that a similar position in the model line would double my core count with at least an additional gigahertz of base clock speed and a meaningful IPC increase. If performance becomes a problem in the meantime, then I will give my BeQuiet Dark Rock Pro 4 a third fan (likely replacing the front fan at the same time) and enable PBO.
Optical measurement from keyence is super cool. I was sceptical of liquid metal because of high ambient temperature of 30-40C with humidity >90% in the sub-continent. Do you think laptops and handheld device with LM might leak or slip out even at rest? The sheet are quite easy to apply and not require frequent maintenance.
I'm running a 5700g in an Asrock DeskMini system, and might be interested in this chip if Asrock decides to make an AM5 version. I do typically overclock as much as I can, but push the GPU side the hardest. I would like to see that kind of testing done, but understand why it's not popular. I would be interested in seeing a lapped version tested, maybe in 3dMark to push the iGPU side.
0:33 I disagree with you on that, the AMD nomenclature is completely transparent if you stop trying to "assume" and actually go and check their naming scheme. I don't like it either, but they had to make it like that eventually because the sheer amount of products they can have at the same time in the market, usually spanning several families, lithographies and variants.
Derbauer, question, the 3D analyzer said x:y 1078 and 1011 and 1389 and 1231 (mils) question divide by 100 correct? Approximately 170 mm2 of die area. Question what is the square mm area of the 8700 die? mb
Just mind boggling how Amd didn't ship the cpu with solder/liquid metal to begin with
great tuning result!
Very impressed to see 20k points from the 8700g in R23. That's how my i7 13700 performed with the stock intel cooler. (On a good cooler it can score 29000 points)
17:40 oh crap I’ve remounted mine twice already! Once when switching from a 5800x3d to a 7800x3d and a second time when switching hung from soft to hardline. My temps are still really good though!
He’s got a share in Thermal Grizzly, so might be the reason why he recommends replacing it after using it once 🫢
Wow, I was not expecting the first type of delid, I didnt know such method exist.
9:40 Oh no, the mad man has lightning in his workshop ;)
Very intreasting Video It would be interesting to see just how far you can push the chip.
I think these are for small portable systems, say small pc for car motorhome coach or plane where space and power available may be limited.
Kinda makes me want to see what absolute peak we could get in performance from one of these with full CPU-iGP-RAM OC direct die custom loop.
How well would the integrated GPU perform with 3DV-cache added?
Unless the game is old enough to run the whole vram budget in the 3d cache, pretty good. Anything that uses over 100mb of vram irrelavent it will still hit the slow ddr5.
There's no 3d v cache for G-series chips
It would have to access the cache through the infinity fabric.
@@BusAlexey I know. I was speaking in a what if scenario.
Contact frames or IHS definetly don't make commercial sense given that this CPU is already more of a niche use case, so it would be trying to sell to the niche of the niche.
That being said it would be fun to just grab an old prototype of contact frame and adapt it to see how far you can push this thing.
It'd be cool to to see the igpu overclock as well and some balancing of it. Like i assume high overclock of cpu wouldn't be possible with high gpu overclock, but fiddling around which could be somewhat a sweetspot would be interesting to see. Maybe only the igpu overclock would be the best solution? idk
On the 7600 overclocking the iGPU at all seemingly leads to mandatory ~2.7Ghz clock-limiting of the CPU all-core workload under simultaneous iGPU load (even if headroom exists). If temperature exceeds 95C then the CPU speed is sacrificed further in preference to GPU performance.
I'm not seeing PBO2 tuning where you give it a negative offset. Wouldn't that also be able to get the same results, if you're lucky with the ASIC quality.
The issue I have with forcing all core OCing is that they stay at that frequency all the time. or at least it used to be like that when I tried on my 5600x. Seems like a huge power waste to run that all the time for the odd moments you need max CPU power. Too bad they don't let you adjust the PBO settings to boost higher with a low voltage.
Please do a test with Igpu performance. I might jump to an 8700g from my 5600+A2000 chip just for fun. De-lid and see how much performance I can get in a 5L case and just IGPU :)
I can see the 8700g to be useful for Linux users, especially those that use virtual machines often. Having a powerful APU opens up possibilities for better high res recording as well as keeping the host system on that GPU if you plan on GPU forwarding. Definitely a cool processor in my opinion.
I would love to see you redo this with a Ryzen AI 9 370 HX. I'm not sure what motherboards will support this, but it performs well without overclocking.
I would like to see a direct die frame for the 8700G because even though this is a year old now, the GPU performance, and specifically, the ability to allocate up to 64GB to the GPU should make this an excellent choice for running LLMs
Thanks for the entertaining video.
You ought to give the iGPU a shot at potential overclocking. Maybe find some ram that can clock relatively high? See where the 780M iGPU lands in the dedicated GPU space.
Wait, dont you have to account for the tape when you were measuring height? You placed the IHS on the tape...
Direct die cooling would be interesting to see just how far you could push the igpu.
@DeBauer I'd like to see what this modified APU can do, I 'd like to see Actually Hardcore Overclocking experiment further. perhaps Ancient Gameplays. Perhaps people that optomised (fclock?) with the previous generation (5700G, 5600G) to maximize performance.
I would really like to see what this can do without an discreet GPU, both gaming and more professional use cases, such as video rendering, transcoding, or CAD/CAM use cases.
Thank you for reading, sir.