• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

DDR5 Memory Performance Scaling with AMD Zen 5

Joined
Aug 20, 2007
Messages
21,499 (3.40/day)
System Name Pioneer
Processor Ryzen R9 9950X
Motherboard GIGABYTE Aorus Elite X670 AX
Cooling Noctua NH-D15 + A whole lotta Sunon and Corsair Maglev blower fans...
Memory 64GB (4x 16GB) G.Skill Flare X5 @ DDR5-6000 CL30
Video Card(s) XFX RX 7900 XTX Speedster Merc 310
Storage Intel 905p Optane 960GB boot, +2x Crucial P5 Plus 2TB PCIe 4.0 NVMe SSDs
Display(s) 55" LG 55" B9 OLED 4K Display
Case Thermaltake Core X31
Audio Device(s) TOSLINK->Schiit Modi MB->Asgard 2 DAC Amp->AKG Pro K712 Headphones or HDMI->B9 OLED
Power Supply FSP Hydro Ti Pro 850W
Mouse Logitech G305 Lightspeed Wireless
Keyboard WASD Code v3 with Cherry Green keyswitches + PBT DS keycaps
Software Gentoo Linux x64 / Windows 11 Enterprise IoT 2024
I would've thrown in some FCLK syncable rates like DDR5-4200, etc. Always been curious about those and what 1:1:1 could possibly do to boost those.

Other than that, great review.
 
Joined
Apr 29, 2014
Messages
4,293 (1.11/day)
Location
Texas
System Name SnowFire / The Reinforcer
Processor i7 10700K 5.1ghz (24/7) / 2x Xeon E52650v2
Motherboard Asus Strix Z490 / Dell Dual Socket (R720)
Cooling RX 360mm + 140mm Custom Loop / Dell Stock
Memory Corsair RGB 16gb DDR4 3000 CL 16 / DDR3 128gb 16 x 8gb
Video Card(s) GTX Titan XP (2025mhz) / Asus GTX 950 (No Power Connector)
Storage Samsung 970 1tb NVME and 2tb HDD x4 RAID 5 / 300gb x8 RAID 5
Display(s) Acer XG270HU, Samsung G7 Odyssey (1440p 240hz)
Case Thermaltake Cube / Dell Poweredge R720 Rack Mount Case
Audio Device(s) Realtec ALC1150 (On board)
Power Supply Rosewill Lightning 1300Watt / Dell Stock 750 / Brick
Mouse Logitech G5
Keyboard Logitech G19S
Software Windows 11 Pro / Windows Server 2016
Very interesting, I like seeing those tweaked results to get the DDR5 6400 to perform that well. Makes me want to pick up a kit and switch to a 9XXX X3D chip in the future and play with it.

Great review!
 
Joined
Mar 4, 2005
Messages
3,628 (0.50/day)
System Name TheReactor / HTPC
Processor AMD 7800x3d 5050Mhz / Intel 10700kf (5.1ghz All Core)
Motherboard ASrock x670e Taichi / ROG Strix z490-e gaming
Cooling HeatKiller VI CPU/GPU Block -2xBlackIce GTX 360 Radiators - Swiftech MCP655 Pump
Memory 32GB G.Skill 6000Mhz DDR5 / 32GB G.Skill 3400Mhz DDR4
Video Card(s) Nvidia 3090ti / Nvidia 2080ti
Storage Crucial T700 2TB Gen 5 / Samsung Evo 2Tb
Display(s) Acer Predator xb271hu - 2560x1440 @144hz
Case Corsiar 550
Audio Device(s) on board
Power Supply Antec Quattro 1000W
Mouse Logitech G502
Keyboard Corsair Gaming k70
Software Windows 10 Pro 64bit
Nice review. Good to know my DDR5 6000 is only about 1-2% slower than much more expensive, slightly faster ram. Not a single reason to even consider an upgrade. Not only that but if building a new system the cost of anything above 6000 is a very poor, nearly worthless ROI.
 

bug

Joined
May 22, 2015
Messages
13,810 (3.96/day)
Processor Intel i5-12600k
Motherboard Asus H670 TUF
Cooling Arctic Freezer 34
Memory 2x16GB DDR4 3600 G.Skill Ripjaws V
Video Card(s) EVGA GTX 1060 SC
Storage 500GB Samsung 970 EVO, 500GB Samsung 850 EVO, 1TB Crucial MX300 and 2TB Crucial MX500
Display(s) Dell U3219Q + HP ZR24w
Case Raijintek Thetis
Audio Device(s) Audioquest Dragonfly Red :D
Power Supply Seasonic 620W M12
Mouse Logitech G502 Proteus Core
Keyboard G.Skill KM780R
Software Arch Linux + Win10
Nice review. Good to know my DDR5 6000 is only about 1-2% slower than much more expensive, slightly faster ram. Not a single reason to even consider an upgrade. Not only that but if building a new system the cost of anything above 6000 is a very poor, nearly worthless ROI.
I mean, RAM is the second fastest component in a PC, behind the CPU. You want better performance, you better look for real bottlenecks.
I know, there are some workloads that are sensitive to RAM bandwidth. If you frequently need that, by all means, upgrade the RAM.
 
Joined
Mar 4, 2005
Messages
3,628 (0.50/day)
System Name TheReactor / HTPC
Processor AMD 7800x3d 5050Mhz / Intel 10700kf (5.1ghz All Core)
Motherboard ASrock x670e Taichi / ROG Strix z490-e gaming
Cooling HeatKiller VI CPU/GPU Block -2xBlackIce GTX 360 Radiators - Swiftech MCP655 Pump
Memory 32GB G.Skill 6000Mhz DDR5 / 32GB G.Skill 3400Mhz DDR4
Video Card(s) Nvidia 3090ti / Nvidia 2080ti
Storage Crucial T700 2TB Gen 5 / Samsung Evo 2Tb
Display(s) Acer Predator xb271hu - 2560x1440 @144hz
Case Corsiar 550
Audio Device(s) on board
Power Supply Antec Quattro 1000W
Mouse Logitech G502
Keyboard Corsair Gaming k70
Software Windows 10 Pro 64bit
1-2%? Sure, technically it is "better performance". Will I notice that extra performance...unlikely, especially in my use case as a gaming system. 1-2FPS isn't going to move the needle.
 
  • Like
Reactions: bug
Joined
Jul 24, 2024
Messages
258 (1.87/day)
System Name AM4_TimeKiller
Processor AMD Ryzen 5 5600X @ all-core 4.7 GHz
Motherboard ASUS ROG Strix B550-E Gaming
Cooling Arctic Freezer II 420 rev.7 (push-pull)
Memory G.Skill TridentZ RGB, 2x16 GB DDR4, B-Die, 3800 MHz @ CL14-15-14-29-43 1T, 53.2 ns
Video Card(s) ASRock Radeon RX 7800 XT Phantom Gaming
Storage Samsung 990 PRO 1 TB, Kingston KC3000 1 TB, Kingston KC3000 2 TB
Case Corsair 7000D Airflow
Audio Device(s) Creative Sound Blaster X-Fi Titanium
Power Supply Seasonic Prime TX-850
Mouse Logitech wireless mouse
Keyboard Logitech wireless keyboard
Pardon me, I can't find a page relating to bandwidth and latency results (AIDA64). W1zzard, is it possible to provide those results, please? Many thanks!
 
Joined
Sep 11, 2023
Messages
25 (0.05/day)
You're confusing CLOCK frequency with TRANSFER frequency, which was the point I made really clear, I thought.

"Frequency" in DRAM is not a term uniquely used for the clocking operations. Yes. The clock pulse frequency or "clockspeed" of DDR is half that of the actual operating frequency, but this clock pulse by itself is merely a control signal used to sync the pacing and rate of the actual transfer frequency, which is the real number of times a second your DDR is sending or receiving data.

Think of it like humans walking or running, for example. All of the terminology is in steps/cadence/paces/strides and they are the most obvious, default way to talk about something that has an underlying frequency of HALF the steps/paces/strides per minute, The "clockspeed" of your legs is rate at which you swing a leg through one complete cycle - say from the point your heel strikes the ground until the point at which the same heel strikes the ground again - but for every single one of those, you take TWO paces. Nobody calls them half-paces or half-steps, yet that's what they are if you are incorrectly laser focused on the actual cycle rate of each leg.

What would be the transfer frequency?? There's either frequency or transfer rate not a mix of both.
A frequency refers to a cycle, an event that is repeated.
Mistaking frequency for transfert is like mistaking a one-way trip for a round trip. Only one of them can be repeated.

Using your example, with someone running at a cadence of 180 steps per minute or 3 steps per seconds.
We can either say "3 steps per second" or 1.5Hz if we want a unit that is harder to use and less relevant.
We cannot say 3Hz as that would mean that a full event is a single step. With the same foot each time so we can repeat exactly the same step the next cycle.
If you want to use the frequency you have to count the full cycle, with two steps, one for each foot.
 
Joined
Feb 20, 2019
Messages
8,320 (3.93/day)
System Name Bragging Rights
Processor Atom Z3735F 1.33GHz
Motherboard It has no markings but it's green
Cooling No, it's a 2.2W processor
Memory 2GB DDR3L-1333
Video Card(s) Gen7 Intel HD (4EU @ 311MHz)
Storage 32GB eMMC and 128GB Sandisk Extreme U3
Display(s) 10" IPS 1280x800 60Hz
Case Veddha T2
Audio Device(s) Apparently, yes
Power Supply Samsung 18W 5V fast-charger
Mouse MX Anywhere 2
Keyboard Logitech MX Keys (not Cherry MX at all)
VR HMD Samsung Oddyssey, not that I'd plug it into this though....
Software W10 21H1, barely
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
What would be the transfer frequency?? There's either frequency or transfer rate not a mix of both.
A frequency refers to a cycle, an event that is repeated.
Mistaking frequency for transfert is like mistaking a one-way trip for a round trip. Only one of them can be repeated.

Transfer rate is an amount transferred over time, ie Gigabytes per second.
Transfer frequency (in Hz, that's the SI unit for frequency) is the number of times a transfer happens per second.
For DDR5-4800, the transfer frequency is 4800MHz.

Clock frequency is irrelevant to this discussion, and it's been mostly irrelevant to the stated speed of RAM for the last 26 years of DDR's market history.
The only time you'll encounter the clock frequency of RAM these days is when discussing the 1:1 ratio of FCLK and MCLK in the BIOS. Note the terminology, "MCLK" refers to clock, ie the timing frequency, not the operating frequency of the memory. The clocks haven't matched the operating frequency for 26 years, I'm simply staggered that people haven't worked this out yet. DDR transfer frequency is 2x the clock frequency, GDDR5 and GDDR6 are 4x the clock frequency, and Rambus now have RAM thats 32x the clock frequency. Can you see that clock frequency is irrelevant now? That's why we talk in terms of transfer frequency and transfer rate.

In case it's not crystal clear:
Clock frequency and transfer frequency are both frequencies, but they are not the same frequencies and haven't been since SDR RAM fell out of use a quarter of a century ago. Since the ratio of clock frequency to transfer frequency varies depending on which type of memory you're talking about, there's no point even using clock frequency in a discussion unless you're comparing different memory types that use different clock-frequency-to-transfer-frequency ratios.

Mistaking frequency for transfer is like mistaking a one-way trip for a round trip.
Mistaking WHICH frequency?!
See? It really really matters. Omitting which frequency you're talking about makes your statement worthless because it's incomplete. You also have to define "transfer" better too. I know you mean transfer frequency from context, but you've already confused transfer frequency and transfer rate above.

What you really meant to say was "Mistaking clock frequency for transfer frequency is like mistaking a one-way trip for a round trip". That would be a correct sentence.

My entire point is that people don't understand the terminology properly and make nonsensical, incomplete statements like your quotes that just add to the confusion. Don't do that; please learn and spread education, rather than sowing confusion and ambiguity.
 
Last edited:
Joined
Sep 11, 2023
Messages
25 (0.05/day)
Transfer rate is an amount transferred over time, ie Gigabytes per second.
That's data rate, not transfert rate.
"Transfer" and "rate" are rather simple words and used together mean a number of transfers relative to another value, here time was implicit and you understood this part as "per second" yourself.
Words have meaning and you cannot just say you are right to use the wrong ones because you don't accept common definition and replace them by your own.

Transfer frequency (in Hz, that's the SI unit for frequency) is the number of times a transfer happens per second.
This new definition of yours's, this "transfer frequency", what should it's unit be? It will need to be "something per second" or "something in hertz" to be easily understood.
And we cannot really use the hertz with no clarification because it's already used as is for the carrier signal.
We need to add another word for clarity, like transfert. And we could write it as a "T" and to express the number of time it happens in a second add a "per second" somewhere.

For DDR5-4800, the transfer frequency is 4800MHz.
Yeah, or instead of writing "a transfer frequency of 4800MHz" to try and be smart because "what is a clockspeed these days anyway?" just write 4800MT/s like everybody.
That means exactly the same thing in a clearer way.
And this way there is no confusion with the rest of the things that you believe are "stuck in the past" like the actual carrier signal frequency. Computer still relies on those to transmit data... It's not just a "control signal used to sync the pacing and rate of the actual transfer frequency" it's the real signal.

Clock frequency is irrelevant to this discussion, and it's been mostly irrelevant to the stated speed of RAM for the last 26 years of DDR's market history.
The only time you'll encounter the clock frequency of RAM these days is when discussing the 1:1 ratio of FCLK and MCLK in the BIOS. Note the terminology, "MCLK" refers to clock, ie the timing frequency, not the operating frequency of the memory. The clocks haven't matched the operating frequency for 26 years, I'm simply staggered that people haven't worked this out yet. DDR transfer frequency is 2x the clock frequency, GDDR5 and GDDR6 are 4x the clock frequency, and Rambus now have RAM thats 32x the clock frequency. Can you see that clock frequency is irrelevant now? That's why we talk in terms of transfer frequency and transfer rate.
Yeah, so maybe stop talking about it if that's not relevant to the discussion. Because I definitely wasn't talking about this.
My only point is that we have two words "MHz" and "MT/s" to describe two different things. There is no need to try and stretch the definition of a hertz to avoid using MT/s.

In case it's not crystal clear:
Clock frequency and transfer frequency are both frequencies,
That's were your stretching the definitions. The hertz is prevalent for use in cyclic events, using it to amalgamate two physically distincts events is incorrect in my book.
The two transfers per cycle are not the same, they don't happen in the same conditions! A single transfer doesn't physically repeat itself in a periodic manner like the carrier signal does.

but they are not the same frequencies and haven't been since SDR RAM fell out of use a quarter of a century ago. Since the ratio of clock frequency to transfer frequency varies depending on which type of memory you're talking about, there's no point even using clock frequency in a discussion unless you're comparing different memory types that use different clock-frequency-to-transfer-frequency ratios.
Since there is no point in using the frequency to describe the memory (I agree with you) why not use the right unit, the MT/s?

Mistaking WHICH frequency?!
See? It really really matters. Omitting which frequency you're talking about makes your statement worthless because it's incomplete. You also have to define "transfer" better too. I know you mean transfer frequency from context, but you've already confused transfer frequency and transfer rate above.

What you really meant to say was "Mistaking clock frequency for transfer frequency is like mistaking a one-way trip for a round trip". That would be a correct sentence.

My entire point is that people don't understand the terminology properly and make nonsensical, incomplete statements like your quotes that just add to the confusion. Don't do that; please learn and spread education, rather than sowing confusion and ambiguity.
You're the only person believing that "frequency" might mean "transfer rate" instead of "frequency". The word I used have meaning, just stick to it don't infer your misunderstanding of what is a carrier signal.
Coincidently you're also the only one that doesn't want to use MT/s to avoid confusion.
 
Joined
Jul 31, 2024
Messages
406 (3.10/day)
For DDR5-4800, the transfer frequency is 4800MHz.

No.
The 4800Mhz can not be measured so therefore it does not exists with DDR5-4800. Summary: That was my initial comment to check with an oscilloscope (or theorectically) if you can measure 2400Mhz with DDR5-4800. These days is kinda easy for humanity as they get the frequency shown on scopes. In my days it was all analogue measurements.

You may look up SI units, what an oscilloscope does and the definition of a repeating waveform. Some logic uses the rising and the falling flank of a clock signal.
A definition can not be stretched Hz - is well defined. It can be measured.
 
Joined
Oct 24, 2022
Messages
221 (0.28/day)
You mean with 24H2 Beta? No, others tested it already, I focused on this article instead. Once 24H2 final is out we'll test it of course, and I'm sure there will be even more performance improvements. Next AGESA will have more improvements, too, so stay tuned

Can you, please, do some tests showing the performance difference with a Ryzen 9700X with DDR5 6000 memory with different timings?

For example:
28-36-36-96
30-36-36-76
32-38-38-76
36-36-36-76
36-38-38-76
38-46-46-76
40-40-40-77
 

MAG

Joined
Jan 12, 2005
Messages
8 (0.00/day)
Location
Germany
Which Y-Cruncher version has been used during the review?

1728062164485.png


1728062187136.png


Version 0.8.5 is even faster

1728067697142.png
 
Last edited:

MrSleezebag1964

New Member
Joined
Jan 14, 2023
Messages
8 (0.01/day)
All - Longtime Zen 3 owner planning Zen 5 build sometime this year. I recall back in the days of selecting RAM for 5800X, it was often reported that 64GB couldn't run at the same speed as reliably as 32GB kits. Specifically, 3600/CL 16 could be run in practically every system when using a 32 GB (2 X 16 kit). But for 64GB kits, 3200/CL 16 was generally suggested (This was my understanding as of ~January 2021 - I realize some things may have changed by now!).

Does the same hold true with Zen 5? I'd like to simply aim for the sweet spot of performance withOUT trying to ultra-optimize/tune/test for hours on end. Sounds like 6000/CL<lowest possible> is ideal - but do I need to back off speed or latency a bit if planning to run 48GB or higher?

Will likely aim for the 9800X3D - and I understand historically 3D cache has not been as sensitive to RAM speeds.

Originally planned to go for Zen4 and then decided to wait it out at last minute. I had saved these two kits to PC Parts Picker as potential options. (In case I needed low profile/no RGB for ITX build - I'm guessing they are the same product under the heat shield).
  • G.SKILL Trident Z5 Neo RGB: F5-6000J3040G32GX2-TZ5NR (30-40-40-96)

  • G.SKILL Flare X Series 64GB: F5-6000J3040G32GX2-FX5 (30-40-40-96)

 
Joined
Feb 28, 2015
Messages
95 (0.03/day)
Hm... FLCK 2000mhz only
Even on my crapy ddr5 5600 CL36 when I changed FLCK from 2000 to 2133 I get few% improvements
 

bug

Joined
May 22, 2015
Messages
13,810 (3.96/day)
Processor Intel i5-12600k
Motherboard Asus H670 TUF
Cooling Arctic Freezer 34
Memory 2x16GB DDR4 3600 G.Skill Ripjaws V
Video Card(s) EVGA GTX 1060 SC
Storage 500GB Samsung 970 EVO, 500GB Samsung 850 EVO, 1TB Crucial MX300 and 2TB Crucial MX500
Display(s) Dell U3219Q + HP ZR24w
Case Raijintek Thetis
Audio Device(s) Audioquest Dragonfly Red :D
Power Supply Seasonic 620W M12
Mouse Logitech G502 Proteus Core
Keyboard G.Skill KM780R
Software Arch Linux + Win10
All - Longtime Zen 3 owner planning Zen 5 build sometime this year. I recall back in the days of selecting RAM for 5800X, it was often reported that 64GB couldn't run at the same speed as reliably as 32GB kits. Specifically, 3600/CL 16 could be run in practically every system when using a 32 GB (2 X 16 kit). But for 64GB kits, 3200/CL 16 was generally suggested (This was my understanding as of ~January 2021 - I realize some things may have changed by now!).

Does the same hold true with Zen 5? I'd like to simply aim for the sweet spot of performance withOUT trying to ultra-optimize/tune/test for hours on end. Sounds like 6000/CL<lowest possible> is ideal - but do I need to back off speed or latency a bit if planning to run 48GB or higher?

Will likely aim for the 9800X3D - and I understand historically 3D cache has not been as sensitive to RAM speeds.

Originally planned to go for Zen4 and then decided to wait it out at last minute. I had saved these two kits to PC Parts Picker as potential options. (In case I needed low profile/no RGB for ITX build - I'm guessing they are the same product under the heat shield).
  • G.SKILL Trident Z5 Neo RGB: F5-6000J3040G32GX2-TZ5NR (30-40-40-96)

  • G.SKILL Flare X Series 64GB: F5-6000J3040G32GX2-FX5 (30-40-40-96)

This has always been the case: larger memory sizes mean higher latency, lower clocks to keep it stable.
That is why getting RAM/VRAM "just in case" is usually not the best option. If you find yourself running out of 32GB RAM and swapping then yes, going 64GB, even at lower frequencies, means your system will be faster overall. Otherwise, stick with 32GB.
 

MrSleezebag1964

New Member
Joined
Jan 14, 2023
Messages
8 (0.01/day)
This has always been the case: larger memory sizes mean higher latency, lower clocks to keep it stable.
That is why getting RAM/VRAM "just in case" is usually not the best option. If you find yourself running out of 32GB RAM and swapping then yes, going 64GB, even at lower frequencies, means your system will be faster overall. Otherwise, stick with 32GB.
Ya makes sense - Honestly didn't think I'd need more than 32GB for a long time, but Flux AI Image gen model really puts system through the ringer! It will swap VRAM to RAM and 32GB just ain't enough - which then spills to virtual memory makes entire system chug along painfully. I was thinking 48GB would likely be sufficient for capacity while perhaps offering best compromise between performance/capacity. A little cheaper than a 64GB kit too.

So, in theory if one were to run one of the aforementioned 64GB 6000\CL30 kits - would it be advised to apply EXPO profile, and then INCREASE timings a bit? Or decrease the mt/s speed?
 

bug

Joined
May 22, 2015
Messages
13,810 (3.96/day)
Processor Intel i5-12600k
Motherboard Asus H670 TUF
Cooling Arctic Freezer 34
Memory 2x16GB DDR4 3600 G.Skill Ripjaws V
Video Card(s) EVGA GTX 1060 SC
Storage 500GB Samsung 970 EVO, 500GB Samsung 850 EVO, 1TB Crucial MX300 and 2TB Crucial MX500
Display(s) Dell U3219Q + HP ZR24w
Case Raijintek Thetis
Audio Device(s) Audioquest Dragonfly Red :D
Power Supply Seasonic 620W M12
Mouse Logitech G502 Proteus Core
Keyboard G.Skill KM780R
Software Arch Linux + Win10
Ya makes sense - Honestly didn't think I'd need more than 32GB for a long time, but Flux AI Image gen model really puts system through the ringer! It will swap VRAM to RAM and 32GB just ain't enough - which then spills to virtual memory makes entire system chug along painfully. I was thinking 48GB would likely be sufficient for capacity while perhaps offering best compromise between performance/capacity. A little cheaper than a 64GB kit too.

So, in theory if one were to run one of the aforementioned 64GB 6000\CL30 kits - would it be advised to apply EXPO profile, and then INCREASE timings a bit? Or decrease the mt/s speed?
That last part depends on what you need most. Some apps are more sensitive to latency than bandwidth, some are the other way around. If you're seeing frequent overspills, chance are your workload love bandwidth more. But the only surefire way to tell is to test.
 
Joined
Jul 24, 2024
Messages
258 (1.87/day)
System Name AM4_TimeKiller
Processor AMD Ryzen 5 5600X @ all-core 4.7 GHz
Motherboard ASUS ROG Strix B550-E Gaming
Cooling Arctic Freezer II 420 rev.7 (push-pull)
Memory G.Skill TridentZ RGB, 2x16 GB DDR4, B-Die, 3800 MHz @ CL14-15-14-29-43 1T, 53.2 ns
Video Card(s) ASRock Radeon RX 7800 XT Phantom Gaming
Storage Samsung 990 PRO 1 TB, Kingston KC3000 1 TB, Kingston KC3000 2 TB
Case Corsair 7000D Airflow
Audio Device(s) Creative Sound Blaster X-Fi Titanium
Power Supply Seasonic Prime TX-850
Mouse Logitech wireless mouse
Keyboard Logitech wireless keyboard
Choose RAM capacity per your requirements. Latency penalties are less important when there is not enough RAM. System will need to use a disk for swapping. Disk is much slower than RAM and has latency in miliseconds or hundreds of microseconds, while RAM has latency in 50-80 microseconds. Some less than 10 microsecond penalty is nothing compared to penalties you're experiencing with swapping to disk. Swapping also kills your SSD, slowly, but surely.

With Zen 4/5, get kit of 2 dimms (preferably dual-rank), 6000 MHz or 6400 MHz kit with the tightest timings possible.
 

MrSleezebag1964

New Member
Joined
Jan 14, 2023
Messages
8 (0.01/day)
Choose RAM capacity per your requirements. Latency penalties are less important when there is not enough RAM. System will need to use a disk for swapping. Disk is much slower than RAM and has latency in miliseconds or hundreds of microseconds, while RAM has latency in 50-80 microseconds. Some less than 10 microsecond penalty is nothing compared to penalties you're experiencing with swapping to disk. Swapping also kills your SSD, slowly, but surely.

With Zen 4/5, get kit of 2 dimms (preferably dual-rank), 6000 MHz or 6400 MHz kit with the tightest timings possible.
Appreciate it! Gaming is still primary driver behind best performance & hardware for planned build - I tinker with creativity/AI workloads, but just for fun/exploration.
 
Joined
Jun 19, 2023
Messages
112 (0.21/day)
System Name EnvyPC
Processor Ryzen 7 7800X3D
Motherboard Asus ROG Strix B650E-I Gaming WiFi
Cooling Thermalright Peerless Assassin 120 White / be quiet! Silent Wings Pro 4 120mm (x2), 140mm (x3)
Memory 64GB Klevv Cras V RGB White DDR5-6000 (Hynix A) 30-36-36-76 1.35v
Video Card(s) MSI GeForce RTX 4080 16GB Gaming X Trio White Edition
Storage SK Hynix Platinum P41 2TB NVME
Display(s) HP Omen 4K 144Hz
Case Lian Li Dan Case A3 White
Audio Device(s) Creative Sound BlasterX G6
Power Supply Corsair SF750 Platinum
Mouse Logitech G600 White
Software Windows 11 Pro
Lil late to the party here, but I noticed there are two entries for DDR5-6000 C28 on some of these charts. Is that just a typo?

1730363835752.png
1730364012108.png
 
Joined
Jun 10, 2021
Messages
121 (0.09/day)
Wish both 1:1 and 1:2 were tested at 6400 CL32. Potentially even showing overall latency difference from half clock
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
27,890 (3.71/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
Lil late to the party here, but I noticed there are two entries for DDR5-6000 C28 on some of these charts. Is that just a typo?
Oh, that's an older benchmark run, not sure how it managed to sneak in there. Just checked, I just deleted the data for this run because I am working on Arrow Lake scaling and needed to free up the name :/
 

BillWill

New Member
Joined
Nov 13, 2024
Messages
1 (0.04/day)
I think a better move would have been to synchronize FCLK and UCLK at 2000MHz (instead of 2100MHz for FCLK) when running DDR5-8000 in 2:1 mode. Reduces latency and may have lead to better results - would be interesting to see
 
Last edited:
Top