Jump to content

Does RAM Speed REALLY Matter?? (DDR5 Edition)

Plouffe

How much does RAM speed matter? It's a question we've covered before, but with new platforms from AMD and Intel, and of course the new DDR5 spec, it's time to go down the rabbit hole once more...

 

G Skill Trident Z DDR4 3600 CL14: https://geni.us/3hU2Gaq

G.Skill Trident Z5 5600 CL40: https://geni.us/41OzVMc

G.Skill Trident Z5 NEO DDR5 6000 CL30: https://geni.us/4OZOQG

G.Skill Trident Z5 DDR5 6800 CL34: https://geni.us/WIuJe

G.Skill Trident Z5 DDR5 7200 CL34: https://geni.us/QrzKq

Crucial DDR5 4800 CL40: https://geni.us/ZmTm

Crucial DDR5 5200 CL42: https://geni.us/ywIZkl

 

Purchases made through some store links may provide some compensation to Linus Media Group.

 

 

Link to comment
Share on other sites

Link to post
Share on other sites

Yes, it matters.

How much it matters depends on you.

Speed costs, how fast do you want to spend?

NOTE: I no longer frequent this site. If you really need help, PM/DM me and my e.mail will alert me. 

Link to comment
Share on other sites

Link to post
Share on other sites

That's an informative video - thanks for making it!

 

That being said, I really think you should start including ML/AI benchmarks. A good chunk of top-end gaming system are using thing as a co-justification of purchase (even if for learning), and that probably represents more of your viewers who would be using the system for that "productivity" purpose than 3D rendering. 

 

Besides, that's the current driver in GPU tech development, without which neither the prices nor the capabilities make sense. RTX 4090 is priced through the roof to a point it makes no sense for gaming, but being the only card supporting FP8 besides H100 and more than doubling FP FLOPs compared to RTX 3090Ti, it justifies an immediate buy, even at MSRPx2 for anyone doing ML. 2x RTX 4090 and you can fine-tune 6B parameters LLMs, such as InstructGPT, which is huge. But at that point other components start becoming bottlenecks, because you need to be keeping the cards fed with data you are streaming from storage, and there is not any guidelines for those components. CPU, RAM, SSD, Motherboards - any could impact performance, but it's not clear in which way.

Link to comment
Share on other sites

Link to post
Share on other sites

Only one DDR4 kit when comparing to DDR5?!, Really?

Not everyone is running their DDR4 RAM at 3600MHz CL14...

A PC Enthusiast since 2011
AMD Ryzen 7 5700X@4.65GHz | GIGABYTE GTX 1660 GAMING OC @ Core 2085MHz Memory 5000MHz
Cinebench R23: 15669cb | Unigine Superposition 1080p Extreme: 3566
Link to comment
Share on other sites

Link to post
Share on other sites

23 minutes ago, Andrei Chiffa said:

That being said, I really think you should start including ML/AI benchmarks. A good chunk of top-end gaming system are using thing as a co-justification of purchase (even if for learning), and that probably represents more of your viewers who would be using the system for that "productivity" purpose than 3D rendering. 

If you want to do AI wouldn't some kind of AI accelerator card make more sense. Especially if you have such a large budget. An accelerator card like the Asus one that comes with 8 of Coral AI M.2 TPU modules. Or get some of those Coral AI dual edge TPU m.2 for 40$ if you don't want to spend 1200$ on an accelerator. Wouldn't purpose made hardware for AI/ML make more sense/ provide better performance?

Link to comment
Share on other sites

Link to post
Share on other sites

Er ehere is the DDR5 5600 CL28 kit mentioned in the video like.... or the info about the 5600 kit that was tweaked to this setting..? Linus mentioned it multiple times like.

Ryzen 9 5900X | ALFII 280 | X570 MEG ACE | 32GB Patriot 3733-CL16-20-20-38 | Msi Tri-X 4080 | S-Blaster Z | Sabrent Rocket4 plus-g, Crucial P1, WD Green | Fractal ION 850W 80+ Gold | Define R6 | LG 34GN850 | L-tech K120 & Razer D-adder Mini |

Link to comment
Share on other sites

Link to post
Share on other sites

50 minutes ago, Vishera said:

Only one DDR4 kit when comparing to DDR5?!, Really?

Not everyone is running their DDR4 RAM at 3600MHz CL14...

Indeed, I am running 32GB of DDR4-3733@CL16-20-20--38

Ryzen 9 5900X | ALFII 280 | X570 MEG ACE | 32GB Patriot 3733-CL16-20-20-38 | Msi Tri-X 4080 | S-Blaster Z | Sabrent Rocket4 plus-g, Crucial P1, WD Green | Fractal ION 850W 80+ Gold | Define R6 | LG 34GN850 | L-tech K120 & Razer D-adder Mini |

Link to comment
Share on other sites

Link to post
Share on other sites

54 minutes ago, AndreiArgeanu said:

If you want to do AI wouldn't some kind of AI accelerator card make more sense. Especially if you have such a large budget. An accelerator card like the Asus one that comes with 8 of Coral AI M.2 TPU modules. Or get some of those Coral AI dual edge TPU m.2 for 40$ if you don't want to spend 1200$ on an accelerator. Wouldn't purpose made hardware for AI/ML make more sense/ provide better performance?

Surprisingly, nope. Even if we go around the whole topic of writing and running code on custom hardware, the modern models are gigantic and require a lot of vRAM in inference alone; training is even worse than that. The current recommendation for ML are RTX4090s in local inference and fast loop development; H100s for on-premises and Azure otherwise. 

 

Besides, for price-performance ratio, GPUs have been all the rage since at least 2019. In 2020 Nvidia claimed to have trained at 540B parameters model, whereas Google had to write a new scheduler and dataflow layer to do the same on their TPUs in 2022.

Link to comment
Share on other sites

Link to post
Share on other sites

At 3:19, Linus was talking about how to use AIDA64's memory tests to identify system bottlenecks. Does anyone have a guide or insight on how to start the tests and interpret the data?

Follow the topics you create using the "Follow" button in the top right corner!

One day I will have my GTX 970. One day. PC specs are at my profile.

Not sure how to check what part works with what? Check out my compatibility guide!

Link to comment
Share on other sites

Link to post
Share on other sites

Wait is it a bad thing that my RAM is 2400 MHz? I didn't know RAM even had speed

Link to comment
Share on other sites

Link to post
Share on other sites

34 minutes ago, PokestarFan said:

Wait is it a bad thing that my RAM is 2400 MHz? I didn't know RAM even had speed

 

RAM speed can make a difference depending on application, although for gaming it's usually a marginal impact with other hardware components being much more determinant of performance. 

 

Through the era of complete Intel dominance (roughly 2011-2017) people got in a mentality of "RAM is RAM", but when AMD Ryzen chips came out they were much more sensitive to RAM speeds. 

Corps aren't your friends. "Bottleneck calculators" are BS. Only suckers buy based on brand. It's your PC, do what makes you happy.  If your build meets your needs, you don't need anyone else to "rate" it for you. And talking about being part of a "master race" is cringe. Watch this space for further truths people need to hear.

 

Ryzen 7 5800X3D | ASRock X570 PG Velocita | PowerColor Red Devil RX 6900 XT | 4x8GB Crucial Ballistix 3600mt/s CL16

Link to comment
Share on other sites

Link to post
Share on other sites

2 hours ago, Vishera said:

Only one DDR4 kit when comparing to DDR5?!, Really?

Not everyone is running their DDR4 RAM at 3600MHz CL14...

Because you are in the  0.01% that actually do anything beyond XMP. You want that specialized test? Do it yourself and post the results 🙂 3600mhz Cl14-cl18 is gonna be the MOST people on ryzen, and 3200mhz is the vast majority on Intel systems before the 12th gen series. 

 

You cant honestly expect them to test every single thing that has been shown to be a complete waste of time on DDR4. The diminishing returns beyond 3600mhz are pretty stark, and for most people even getting their systems to boot beyond it can be quite a challenge. Plus its LTT, they arent known for indepth testing anyway, they just do quick tests that can be done quickly because thats their audience, most dont understand deep technical things anyway and it would cost them a ton more money in time. Could they do it? Sure, is there any ROI on it? Most def not since most of their core audience does not care. If you care about that stuff, you either do it yourself or you go to a much more indepth tech channel. They want simple "Does it just work or not" answers, is it simply worth it? Anything beyond that you would already know that since clearly the work has been put in to understand it.

Link to comment
Share on other sites

Link to post
Share on other sites

Thanks for informative video, unfortunately it did not go far enough for me.

  • I am really missing the a compile benchmark (e.g. chromium compile, maybe Linux kernel compile), ideally on a recent Linux distribution (latest Fedora linux ?) but even Windows one is better than none. For people using PC mainly for programming, it is a very useful benchmark.
  • Users who have been burned by silent data corruption usually have the following requirement "PC must have ECC memory" (i.e. Registered ECC or Unbuffered ECC, not the "DDR5-ecc-on-dimm-only" where Linux/Windows-kernel is not notified when a (un)correctable error is detected). Especially if it is a mixed-use machine (i.e. not only gaming).
  • Different types of ECC memory have different effect on latency (Ryzen + ECC UDIMM vs Epyc + ECC Registered - are there any desktop/workstation tasks where ECC Registered memory's higher latency really tanks the performance ?)
  • Higher number of memory channels and effect on benchmarks (e.g. that chromium compile) - mainly because Zen2-derived Epyc CPUs are appearing on eBay (e.g. Epyc 7302p+), with performance probably close to 50% of Ryzen 7950X (there are not that many benchmarks) and at close to half the price for MB+CPU vs MB+CPU, but with Epyc having more PCIE lanes and higher memory limit (and probably somewhat higher idle power consumption).
Link to comment
Share on other sites

Link to post
Share on other sites

5 hours ago, ook512 said:

Thanks for informative video, unfortunately it did not go far enough for me.

  • I am really missing the a compile benchmark (e.g. chromium compile, maybe Linux kernel compile), ideally on a recent Linux distribution (latest Fedora linux ?) but even Windows one is better than none. For people using PC mainly for programming, it is a very useful benchmark.
  • Users who have been burned by silent data corruption usually have the following requirement "PC must have ECC memory" (i.e. Registered ECC or Unbuffered ECC, not the "DDR5-ecc-on-dimm-only" where Linux/Windows-kernel is not notified when a (un)correctable error is detected). Especially if it is a mixed-use machine (i.e. not only gaming).
  • Different types of ECC memory have different effect on latency (Ryzen + ECC UDIMM vs Epyc + ECC Registered - are there any desktop/workstation tasks where ECC Registered memory's higher latency really tanks the performance ?)
  • Higher number of memory channels and effect on benchmarks (e.g. that chromium compile) - mainly because Zen2-derived Epyc CPUs are appearing on eBay (e.g. Epyc 7302p+), with performance probably close to 50% of Ryzen 7950X (there are not that many benchmarks) and at close to half the price for MB+CPU vs MB+CPU, but with Epyc having more PCIE lanes and higher memory limit (and probably somewhat higher idle power consumption).

lvl1tech,servethehome.

 

MSI x399 sli plus  | AMD theardripper 2990wx all core 3ghz lock |Thermaltake flo ring 360 | EVGA 2080, Zotac 2080 |Gskill Ripjaws 128GB 3000 MHz | Corsair RM1200i |150tb | Asus tuff gaming mid tower| 10gb NIC

Link to comment
Share on other sites

Link to post
Share on other sites

Really good video that people was waiting for. But i also want to see comparison between 1 CCX and 2 CCX on AMD platform. Some Russian blogger shows that for 1 CCX is 6400 MT/s memory kit still shows more performance while for 2 CCX best performance at 6000 MT/s memory kit.
Link to video with timecode with gaming tests:
https://www.youtube.com/watch?v=SNbbW5vYiPY&t=738s

 

Link to comment
Share on other sites

Link to post
Share on other sites

Even though I like videos like this that eliminates bottlenecks and show the true difference between the ram, I still don't know if this would make a difference in a budgety system.

I am planning on buying a new computer but don't quiet have the budget for a 7950x and am looking at getting a 7600x paired with an RTX 3070 or an RX 6700 XT. The rig is going primarily used for gaming.

 

Can I transfer the result directly even though it is a much less powerful CPU and GPU?

Would it be better to look at timings instead of speed of the Ram when using AMD as stated in the video?

 

Which would be better.

CL28 5200mhz with response time of 10.76

CL32 6000mhz with response time of 10.66

 

The CL28 is cheaper and the response time is near the same on the two kits, does it even matter on a lower end cpu?

Link to comment
Share on other sites

Link to post
Share on other sites

On 2/15/2023 at 9:24 AM, DanininoDK said:

Even though I like videos like this that eliminates bottlenecks and show the true difference between the ram, I still don't know if this would make a difference in a budgety system.

I am planning on buying a new computer but don't quiet have the budget for a 7950x and am looking at getting a 7600x paired with an RTX 3070 or an RX 6700 XT. The rig is going primarily used for gaming.

 

Can I transfer the result directly even though it is a much less powerful CPU and GPU?

Would it be better to look at timings instead of speed of the Ram when using AMD as stated in the video?

 

Which would be better.

CL28 5200mhz with response time of 10.76

CL32 6000mhz with response time of 10.66

 

The CL28 is cheaper and the response time is near the same on the two kits, does it even matter on a lower end cpu?

I would think the one with a faster response time. Frankly, I find the video somewhat confusing at times. If there's no advantage between 5600-CL28 vs 6000-CL30, why even purchase the 6000 variant?

Link to comment
Share on other sites

Link to post
Share on other sites

On 2/16/2023 at 3:33 AM, shockwave1985 said:

If there's no advantage between 5600-CL28 vs 6000-CL30, why even purchase the 6000 variant?

BINGO.!! We have a winner !!!

 

That was indeed the point of the video, to educated people that "faster" and more expensive memory might indeed be a tremendous waste of Money. !!

 

"I'll buy this 6400 CL 36 because it is the fastest I can afford, and it must be the best."

 

Yeah, it might not be any faster than the 5600 CL 28 that costs $200 less.

 

YES. that is entirely the point of knowledge, to avoid over paying for "bling" that has ZERO return on investment. 

 

 

 

Link to comment
Share on other sites

Link to post
Share on other sites

  • 3 months later...

Would I be better off going 2X 32gig stick of Corsair Vengeance 6200Mhz CL32 or Corsair Vengeance (4x16GB) 6200MHz CL32 DDR5 memory? I wanted Corsair Dominator 6000 Mhz CL30 but I can't seem to find it in Australia.😮‍💨 I'm also open to any other suggestions or advice on lower ram speeds because at the end of the day, I want the best bang for the buck.

 

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×