EXPO or XMP for AMD config ?

[kapone32]

He said EXPO boots longer than XMP, which is wrong according to @ir_cow who reviews motherboards for TPU

There is a video from Level 1 where Wendell discusses EXPO and I was able to confirm it. The first time you boot your AM5 PC the Memory is trained and could take up to 2 to 5 minutes to boot to Windows. If however you pick the 1st Expo profile it will only do that on that occasion. If you choose the 2nd Expo profile the board will train the Memory on every boot cycle. On my Asus X670E board I can fully confirm that it takes 35-45 seconds on Expo 1 and 2-4 minutes on Expo 2.

 

[TokyoQuaSaR]

EXPO = AMD / XMP = Intel

Also have a look into the online manual of the mainboard you're going to buy. Look under "Supported Memory List" and only buy one of them. They're certified to run smooth with the board.
You can also Google what kit is faster or just order 2 different kits and test it out yourself which one is faster. And get only 2 sticks not 4. 2 are more OC stable.

Yes I get that EXPO is better for AMD, my question was more about is EXPO better than a slightly lower CL, since EXPO is from what I know supposed to be a fine tune for AMD compared to XMP. But I guess it has more chance to have been tested for AMD motherboards so in terms of stability EXPO is better, so I guess I should take EXPO for this one reason rather than performance.

 

[Klemc]

Can it depend of which AMD CPU ?.. i have 7800X3D, XMP 6000-C32 Fury Renegade, it accepts the Asus tweaked DOCP profile.

 

[TokyoQuaSaR]

I own this specific G.Skill kit F5-6000J3636F16GX2-FX5 which is AMD Expo ready with CL36-36-36-96 at 6000MT/s it's rock stable first boot ever on my Asus Prime X670-P WiFi I was amazed how fast it actually booted with 32GB of memory after hearing several reviewers saying that boot can take some time.

Link: https://www.gskill.com/product/165/396/1662622664/F5-6000J3636F16GX2-FX5

I am overall happy with the performance vs the price right now if I pay the kit today I would only pay about £128,49 and it's without any rgb.

I know G.Skill recently announced their DDR5 5600MT/s CL28-34-34-89.

Their Neo kit F5-5600J2834F16GX2-TZ5NR costs £173,73 which is still some money but we only in the start of the DDR5 generation so it should improve a lot over time.

G.SKILL Announces New Extreme Low-Latency DDR5-5600 CL28 Memory Kit - G.SKILL International Enterprise Co., Ltd.

4 May 2022 – G.SKILL International Enterprise Co., Ltd., the world’s leading manufacturer of extreme performance memory and gaming peripherals, is excited to announce an extremely low-latency DDR5 memory kit specification at DDR5-5600 CL28 in 64GB 32GBx2 and 32GB 16GBx2 kit capacities, under the...

www.gskill.com

Well speaking of being specific, these aren't excactly the modules I am mentionning since yours are CL36. Here are the ones I found in Japan, both for about 18500 yen :

F5-6000J3040F16GX2-RS5W F5-6000J3040F16GA2-RS5W - Overview - G.SKILL International Enterprise Co., Ltd.

Ripjaws S5 is a performance DDR5 DRAM memory series made with hand-screened memory ICs that passed G.SKILL rigorous validation tests. Each Ripjaws S5 memory kit strikes an ideal balance between performance, compatibility, and stability and is available in matte black or matte white aluminum...

www.gskill.com

F5-6000J3238F16GX2-FX5 - Overview - G.SKILL International Enterprise Co., Ltd.

Designed for DDR5-enabled AMD AM5 platforms, Flare X5 series DDR5 memory is built for high performance in a low-profile design. Featuring AMD EXPO overclocking profiles for easy memory overclocking, users can easily enable overclocked performance by simply enabling EXPO in the BIOS on compatible...

www.gskill.com

I don't need RGB neither since my huge Noctua D15 is probably gonna cover them up entirely

Also DDR5 5600MT/s CL28 = 10ns = DDR5 6000MT/s CL30, 10ns is my target for CAS latency, as it has always been for SDRAM since year 2000 or so. With the same CL (in ns, not cycle), I would definitely go for the higher frequency, so the one I suggest is definitely better, and it's also much cheaper in Japan (£111).

Usually for DRAM, JEDEC first samples start somewhere around a CL of 17ns or so, then when it's getting more widely available, it gets closer to 14ns, and at the same time overclocked modules reach 10ns. Then when the process is fully mature, the best overclocking modules can have a CL down to 7.5ns (for instance DDR4 3600 CL14 = 7.78ns). At least that's how it has been for every single generation of DRAM so I don't expect that to change with DDR5 (it's all about capacitance charging speed so while it improves during one generation development cycle, it doesn't change over generations).

 

[kapone32]

Can it depend of which AMD CPU ?.. i have 7800X3D, XMP 6000-C32 Fury Renegade, it accepts the Asus tweaked DOCP profile.

As long as no WHEA errors you are fine. The difference mainly is that Expo is generally only certified (tested) on AM5 platforms as an exclusive. That takes away Intel's thank you very much and helps the vendor and the consumer. Anyone with their finger on the pulse has seen that DDR5 32GB kits are not very different in pricing vs DDR4 and don't include the Budget RAM that does not have Temp (and other variables) monitoring in HWINFO64. You have to have the more expensive kits on DDR4 to get that function. I mean like GSkill Royal class of RAM.

 

[TokyoQuaSaR]

He said EXPO boots longer than XMP, which is wrong according to @ir_cow who reviews motherboards for TPU

I see, well the one who mentionned that EXPO can possibly take more training time is apparently also from TPU staff, although maybe not testing mb I don't know. Anyway, it's indeed better to know that as a fact rather than guessing. I come from a pretty old PC, I currently use a 4770K (2013). It's funny how in the past most of the time used for booting was from Windows loading, and nowadays it's more and more from the motherboard, thanks to SSDs and also slow motherboards BIOS... I didn't know it was that much longer though.

 

[kapone32]

Well speaking of being specific, these aren't excactly the modules I am mentionning since yours are CL36. Here are the ones I found in Japan, both for about 18500 yen :

F5-6000J3040F16GX2-RS5W F5-6000J3040F16GA2-RS5W - Overview - G.SKILL International Enterprise Co., Ltd.

Ripjaws S5 is a performance DDR5 DRAM memory series made with hand-screened memory ICs that passed G.SKILL rigorous validation tests. Each Ripjaws S5 memory kit strikes an ideal balance between performance, compatibility, and stability and is available in matte black or matte white aluminum...

www.gskill.com

F5-6000J3238F16GX2-FX5 - Overview - G.SKILL International Enterprise Co., Ltd.

Designed for DDR5-enabled AMD AM5 platforms, Flare X5 series DDR5 memory is built for high performance in a low-profile design. Featuring AMD EXPO overclocking profiles for easy memory overclocking, users can easily enable overclocked performance by simply enabling EXPO in the BIOS on compatible...

www.gskill.com

I don't need RGB neither since my huge Noctua D15 is probably gonna cover them up entirely

Also DDR5 5600MT/s CL28 = 10ns = DDR5 6000MT/s CL30, 10ns is my target for CAS latency, as it has always been for SDRAM since year 2000 or so. With the same CL (in ns, not cycle), I would definitely go for the higher frequency, so the one I suggest is definitely better, and it's also much cheaper in Japan (£111).

Usually for DRAM, JEDEC first samples start somewhere around a CL of 17ns or so, then when it's getting more widely available, it gets closer to 14ns, and at the same time overclocked speed reach 10ns. Then when the process is fully mature, the best overclocking modules can have a CL down to 7.5ns (for instance DDR4 3600 CL14 = 7.78ns). At least that's how it has been for every single generation of DRAM so I don't expect that to change with DDR5 (it's all about capacitance charging speed so while it improves during one generation development cycle, it doesn't change over generations).

I got 5200 speed with 30 timings. I love it.

 

[TokyoQuaSaR]

I haven't measured the exact time in a while, but all 6000 32GB kits (expo and XMP) boot at roughly the same time. 6000 64GB kits takes like a extra minute the first time. 128GB, 2-3 m in minutes and 6200 32GB slightly longer vs 6000. Not enough to be noticable. Once it's "trained" either you can never unplug the power or set context restore. Boot times will be like 30 seconds after the initial enabling of XMP/expo.

I have 64GB 5600 in my work computer. Takes like 30 seconds per boot with context restore on. About a minute otherwise. But seriously by the time I press the power button and sit down, it's already booting into Windows

I see, thank you very much for this interesting insight !

 

[kapone32]

I see, well the one who mentionned that EXPO can possibly take more training time is apparently also from TPU staff, although maybe not testing mb I don't know. Anyway, it's indeed better to know that as a fact rather than guessing. I come from a pretty old PC, I currently use a 4770K (2013). It's funny how in the past most of the time used for booting was from Windows loading, and nowadays it's more and more from the motherboard, thanks to SSDs and also slow motherboards BIOS... I didn't know it was that much longer though.

Remember when you could actually get something from the fridge or sweeten your hot beverage while waiting for your PC to boot? Much less getting online which was an exercise that meant you could ride to the Gas station and buy smokes and back before getting online.

 

[TokyoQuaSaR]

EXPO is made specifically for AM5 get that. You should have no issues. I bought and XMP kit for mine and setting XMP gave me WHEA errors.

Another interesting insight, thank you very much. Looks like EXPO is a much better choice then.

 

[ir_cow]

I'm just going off my observation here. Back in Dec someone wanted to know the boot times. Me and someone else (can't remember) time it and basically 6000 (2x16GB) takes about 30-1 minute depending on the AGEAS and motherboard for the first time enabling XMP/EXPO. After which, if you don't disabled the profile or unplug the motherboard power, booting will be slightly slightly lower.

To use a made up example. Say it takes 50 seconds to train the first time. Every boot after will be like 40 seconds. Enable context restore and it does down all the way to 17 seconds for me.

So lots of confusion going on and not just on TPU. Not only does the amount of DIMMs you have impact the boot times, so does the amount of ranks and speed. For me booting 6400 takes minutes (if it boots at all) and it never gets faster. On top of the varies of configuration people are measuring the time differently.

Here is a 3 of ways to measure
Power to BIOS menu
first time enabling profile, restart to BIOS.
Power to widows (worst one to use)

I think it's safe to say after where from 17 seconds to a few minutes it will take AFTER XMP/Expo is already enabled.

There is a video from Level 1 where Wendell discusses EXPO and I was able to confirm it. The first time you boot your AM5 PC the Memory is trained and could take up to 2 to 5 minutes to boot to Windows. If however you pick the 1st Expo profile it will only do that on that occasion. If you choose the 2nd Expo profile the board will train the Memory on every boot cycle. On my Asus X670E board I can fully confirm that it takes 35-45 seconds on Expo 1 and 2-4 minutes on Expo 2.

I respect Wendell. He knows stuff. ASUS is the odd one out. It has memory Profile 1, Profile 2 and Profile 1+ Tweak. Profile 1 is NOT the SPD data. Its ASUS taking the EXPO/XMP profile and adjusting it. Think of it like a mini OC. Profile 2 is just the defaults straight from the SPD data and Tweak is like a even heavier sub-timings adjustments on top of Profile 1. 1 and Tweak are not the actual memory XMP/EXPO that is sold to you.

So on top of the ASUS Profile 1,2 and Tweak is the actual binned profile from the memory (this is a different menu). If you are using ASUS profile 1, it will take a LOT longer to boot.

Thanks ASUS for confusing people.

 

[TokyoQuaSaR]

Remember when you could actually get something from the fridge or sweeten your hot beverage while waiting for your PC to boot? Much less getting online which was an exercise that meant you could ride to the Gas station and buy smokes and back before getting online.

Well it's pretty far away, I mean since 2010 and my first SSD I think it has usually taken about 1min for booting, before that, well, yeah HDD could take 2 min just to boot Windows, maybe more sometimes, the BIOS part was pretty short though. With the PC I built in 2014 I think the boot time increased a bit because of UEFI BIOS, but now with these AM5 mb this seems to take even longer.

PS: I just measured the boot time from power off (but power supply plugged in) to my wallpaper displayed in Windows and it took 41s. I have 3 SDD, RAID enabled on 2 HDD, 1 optical drive, quite a lot of USB devices, mb is my old Asus Z87 Pro with 4770K. I remember this mb having quite a slow boot compared to the one I bought in 2008 actually (of course the SSD makes the Windows 10 boot way faster compared to the Win 7 boot on HDD back then, although I bought an SSD soon after in 2010).

I'm just going off my observation here. Back in Dec someone wanted to know the boot times. Me and someone else (can't remember) time it and basically 6000 (2x16GB) takes about 30-1 minute depending on the AGEAS and motherboard for the first time enabling XMP/EXPO. After which, if you don't disabled the profile or unplug the motherboard power, booting will be slightly slightly lower.

To use a made up example. Say it takes 50 seconds to train the first time. Every boot after will be like 40 seconds. Enable context restore and it does down all the way to 17 seconds for me.

So lots of confusion going on and not just on TPU. Not only does the amount of DIMMs you have impact the boot times, so does the amount of ranks and speed. For me booting 6400 takes minutes (if it boots at all) and it never gets faster. On top of the varies of configuration people are measuring the time differently.

Here is a 3 of ways to measure
Power to BIOS menu
first time enabling profile, restart to BIOS.
Power to widows (worst one to use)

I think it's safe to say after where from 17 seconds to a few minutes it will take AFTER XMP/Expo is already enabled.


I respect Wendell. He knows stuff. ASUS is the odd one out. It has memory Profile 1, Profile 2 and Profile 1+ Tweak. Profile 1 is NOT the SPD data. Its ASUS taking the EXPO/XMP profile and adjusting it. Think of it like a mini OC. Profile 2 is just the defaults straight from the SPD data and Tweak is like a even heavier sub-timings adjustments on top of Profile 1. 1 and Tweak are not the actual memory XMP/EXPO that is sold to you.

So on top of the ASUS Profile 1,2 and Tweak is the actual binned profile from the memory (this is a different menu). If you are using ASUS profile 1, it will take a LOT longer to boot.

Thanks ASUS for confusing people.

Do you imply avoiding Asus, or is it "just" more confusing ? Someone mentionned Gigabyte being pretty fast at booting also, I don't say it's my primary concern when choosing a mb but it's still good to know, if you have some insight about that too.

 

[ir_cow]

Do you imply avoiding Asus, or is it "just" more confusing ? Someone mentionned Gigabyte being pretty fast at booting also, I don't say it's my primary concern when choosing a mb but it's still good to know, if you have some insight about that too.

No, not saying avoid ASUS. Honestly the slowest one I tested was BIOSTAR. Everyone else is within reason. Gigabyte, MSI and ASRock all have the same settings, just in different places. Gigabyte calls it "Low Latency Support / EXPO High Speed Support". Which isn't really a support function lol. More likely to cause stability issues. ASUS calls this Profile 1 and Profile+1 Tweak. ASRock has this enabled already by default as "Aggressive timings", somewhere in some sub-menu. Pick your poison. Gigabyte and MSI are the ones you won't have to mess with the BIOS if you just want normal use. Hope this makes sense.

All these extra functions adds to stability issues and boot times. Now throw in context restore and you are asking for trouble.

 

[MarsM4N]

Do you imply avoiding Asus, or is it "just" more confusing ? Someone mentionned Gigabyte being pretty fast at booting also, I don't say it's my primary concern when choosing a mb but it's still good to know, if you have some insight about that too.

I heard AMD is leaving it up to the motherboard manufacturers to "brew" their own boot training solutions. Some do better, some worse. Still haven't seen a roundup review, so it would require some research to see who's doing it better or worse.

The fastest RAM for gaming FPS right now, from what I've seen, would be DDR5-6000 CL30. AM5 is profitting more from lower CL than higher frequency. I haven't seen any reviews of DDR-5200 CL28 yet, could be faster than DDR5-6000 CL30, but I am highly sceptical. Also AMD themselfs thinks DDR5-6000 is the "sweet spot":

AMD Confirms DDR5-6000 RAM Is The Sweet Spot For Ryzen 7000 CPUs

Directly from the horse's mouth

www.tomshardware.com

 

[ir_cow]

Here is the screenshot of the BIOS. Only one that isn't AMD is ASUS because I am lazy and don't want to pull out the X670E HERO. It will looks basically the same though. Understand ASUS XMP 1/2/Tweak isn't the memory XMP profile. This is ASUS "AI Overclock Tuner", which people get confused with the actual XMP/EXPO below it. Its stupid, but you have to pick one or you cannot set a XMP/EXPO profile with "Auto". It just sticks to DDR5-5200 otherwise. XMP II in this example is actually the default without anything else being done to the timings besides loading whatever XMP/EXPO you pick.

 

[Radxge]

Using XMP 6000 CL32 on an AsRock X670E Pro RS (AM5) and booting has been lightning fast since installed latest bios and system is rock stable.

 

[Shou Miko]

Well speaking of being specific, these aren't excactly the modules I am mentionning since yours are CL36. Here are the ones I found in Japan, both for about 18500 yen :

F5-6000J3040F16GX2-RS5W F5-6000J3040F16GA2-RS5W - Overview - G.SKILL International Enterprise Co., Ltd.

Ripjaws S5 is a performance DDR5 DRAM memory series made with hand-screened memory ICs that passed G.SKILL rigorous validation tests. Each Ripjaws S5 memory kit strikes an ideal balance between performance, compatibility, and stability and is available in matte black or matte white aluminum...

www.gskill.com

F5-6000J3238F16GX2-FX5 - Overview - G.SKILL International Enterprise Co., Ltd.

Designed for DDR5-enabled AMD AM5 platforms, Flare X5 series DDR5 memory is built for high performance in a low-profile design. Featuring AMD EXPO overclocking profiles for easy memory overclocking, users can easily enable overclocked performance by simply enabling EXPO in the BIOS on compatible...

www.gskill.com

I don't need RGB neither since my huge Noctua D15 is probably gonna cover them up entirely

Also DDR5 5600MT/s CL28 = 10ns = DDR5 6000MT/s CL30, 10ns is my target for CAS latency, as it has always been for SDRAM since year 2000 or so. With the same CL (in ns, not cycle), I would definitely go for the higher frequency, so the one I suggest is definitely better, and it's also much cheaper in Japan (£111).

Usually for DRAM, JEDEC first samples start somewhere around a CL of 17ns or so, then when it's getting more widely available, it gets closer to 14ns, and at the same time overclocked modules reach 10ns. Then when the process is fully mature, the best overclocking modules can have a CL down to 7.5ns (for instance DDR4 3600 CL14 = 7.78ns). At least that's how it has been for every single generation of DRAM so I don't expect that to change with DDR5 (it's all about capacitance charging speed so while it improves during one generation development cycle, it doesn't change over generations).

I use a Noctua NH-D15 too just the chroma.black version so I went FlareX and my ram as I said was cheep so I took them and no issues at all.

Maybe when DDR5 matures even more and get into lower prices I will concider an upgrade for lower ns but for now 12ns is fine.

 

[TokyoQuaSaR]

I heard AMD is leaving it up to the motherboard manufacturers to "brew" their own boot training solutions. Some do better, some worse. Still haven't seen a roundup review, so it would require some research to see who's doing it better or worse.

The fastest RAM for gaming FPS right now, from what I've seen, would be DDR5-6000 CL30. AM5 is profitting more from lower CL than higher frequency. I haven't seen any reviews of DDR-5200 CL28 yet, could be faster than DDR5-6000 CL30, but I am highly sceptical. Also AMD themselfs thinks DDR5-6000 is the "sweet spot":

AMD Confirms DDR5-6000 RAM Is The Sweet Spot For Ryzen 7000 CPUs

Directly from the horse's mouth

www.tomshardware.com

Regarding the latency, if you say "profitting more from lower CL than higher frequency", it's CL in ns. DDR5-6000 CL30 = 10ns, 5200 CL28 = 10.77ns, a lower CL cycle number but higher in terms of ns will never be better no matter what. From what AMD said, DDR5-6000 is the sweet spot, and from what has always been SDRAM technology for 25 years, a CL of 10ns is also the sweet spot (for overclocked/non JEDEC voltage/timing modules). So DDR5-6000 CL30 being the sweet spot was also my guess, hence what I chose.

 

[ir_cow]

You can pretty much ignore the CAS value for DDR5 by itself. Lower is better, but the impact of CAS 28 vs 40 in real world applications is small. CAS 40 with tight sub-timings will beat out a CL28 XMP kit without any adjustments.

If your not into memory oc or tweaking, than lower is better. Just isn't a major importance.

 

[TokyoQuaSaR]

No, not saying avoid ASUS. Honestly the slowest one I tested was BIOSTAR. Everyone else is within reason. Gigabyte, MSI and ASRock all have the same settings, just in different places. Gigabyte calls it "Low Latency Support / EXPO High Speed Support". Which isn't really a support function lol. More likely to cause stability issues. ASUS calls this Profile 1 and Profile+1 Tweak. ASRock has this enabled already by default as "Aggressive timings", somewhere in some sub-menu. Pick your poison. Gigabyte and MSI are the ones you won't have to mess with the BIOS if you just want normal use. Hope this makes sense.

All these extra functions adds to stability issues and boot times. Now throw in context restore and you are asking for trouble.

I see thank you. I haven't read about context restore yet so I will have a look about that later.

You can pretty much ignore the CAS value for DDR5 by itself. Lower is better, but the impact of CAS 28 vs 40 in real world applications is small. CAS 40 with tight sub-timings will beat out a CL28 XMP kit without any adjustments.

If your not into memory oc or tweaking, than lower is better. Just isn't a major importance.

I was actually reading your article here : https://www.techpowerup.com/review/ddr5-memory-performance-scaling
4-5% is still something I mean. Where I live difference in price between DDR5-6000 CL30 and 40 is less than 5%, so I'd rather go for CL30...

 

[ir_cow]

@TokyoQuaSaR it is something. I guess now the cost is much lower too.

Not my article. That one is done by our leader