AMD's Infinity Fabric

nlel1975 · 1 Oct 2018 at 22:29

robin66 said:
Not stupid I'd think. You just didn't do enough research I would guess?

Yeah all I do is browse the web and do the odd 3D Mark benchmark run, get bored with BF4 after 10 minutes but it seems to run ok but it is quite old now. I was hoping the extra cores of the Ryzen would help in BF5 which I will no doubt buy and not get value for money from.

Cold_And_Miserable · 2 Oct 2018 at 09:45

SiDeards73 said:
OP has zero credibility when running 32GB of ram for "gaming" and complaining he bought "Slow ram" OP needs to learn how to spec PC components correctly to achieve the best performance per £ spent as lets be honest, buying 32GB of slow ram for gaming is...well...ludicrous and screams you have no idea what you are doing.

Someone wrote that "slow ram" dual-rank modules were faster for some operations but there is no solid information.

Zeed · 2 Oct 2018 at 13:43

With Zen+ sweet spot is 3466cl14 With 3600cl14 it mostly ends up Unstable :/

I could run You same timings 3000 vs 3533 im running in 3dmark if ya want.

Zeed · 2 Oct 2018 at 15:25

Ran a forza 4k locked max settings. 3000cl14 vs 3466cl14

As you see jump is not big but exists. As long as one gets 3000-3200 CL14 its all good

Cl 16 is wayyyy behind

nlel1975 · 3 Oct 2018 at 12:39

Just seen this review of the Ryzen 5 2600

https://www.forbes.com/sites/antony...view-the-best-ever-pc-processor/#1785633756be

At stock speeds they scored 1235 in Cinebench with 3000 memory, I get 1213 in Cinebench with 2133 memory. A whole 22 points less, they do say that with 3200 and above you'll get "much" better results but how much better are we talking? You'd get better results from overclocking the CPU or indeed buying a better CPU to begin with. By the way not trolling.

Freddie1980 · 3 Oct 2018 at 13:41

nlel1975 said:
Just seen this review of the Ryzen 5 2600

https://www.forbes.com/sites/antony...view-the-best-ever-pc-processor/#1785633756be

At stock speeds they scored 1235 in Cinebench with 3000 memory, I get 1213 in Cinebench with 2133 memory. A whole 22 points less, they do say that with 3200 and above you'll get "much" better results but how much better are we talking? You'd get better results from overclocking the CPU or indeed buying a better CPU to begin with. By the way not trolling.

Cinebench doesn't need fast memory to speed up the infinity fabric as the cores are being kept busy by computational work so there is little delay. AIDA64 shows a lot of scaling between memory kits as well as some games (again it depends on what the software is asking your CPU to do).

Also it's not just the base speed but latency as well needs to be taken into consideration. When I first got into computers and self builds getting memory with lower latency was seen as a big deal but after a while the importance of memory speeds dropped off as prices dropped and we could all afford high end kits. These days it's gone back to how it was back in the early 2000's.

nlel1975 · 3 Oct 2018 at 13:50

From a bang for the buck perspective, I still don't see the point of buying super expensive RAM when that money is better spent on a better CPU or GPU. I got a really good deal on the 32GB 2133 stuff I have so am slightly biased. I paid £ 161.05 ex VAT for it with free delivery. I am in Jersey so no VAT either.

Here's my AIDA64 trial version memory benchmark...

This will no doubt highlight the differences between memory speeds.

blaineoliver · 3 Oct 2018 at 15:16

Interestingly you would get much closer to 47GB/s with faster memory (and thats on a 1700 which will be slower than your 2nd gen.)
http://www.legitreviews.com/ddr4-me...tform-best-memory-kit-amd-ryzen-cpus_192259/2

Cinebench puts each job into a core and lets it run, its not swapping data between cores (the thing that is sped up by the infinity fabric being faster). So Cinebench will show little to no difference.

Final page of the review shows where the performance/price sweetspot is, and it also accounts for worse memory prices.

http://www.legitreviews.com/ddr4-me...tform-best-memory-kit-amd-ryzen-cpus_192259/6

nlel1975 · 3 Oct 2018 at 15:38

In very simple terms I guess RAM speed makes most difference to system responsiveness. A GPU to frame rates, and CPU to number crunching. Depends what you value most I guess. Benchmarks are great and I love them but I guess it would be pretty hard for a human to detect the differences in responsiveness between different RAM speeds as we are talking nanoseconds.

Panos · 3 Oct 2018 at 16:51

nlel1975 said:
Just seen this review of the Ryzen 5 2600

https://www.forbes.com/sites/antony...view-the-best-ever-pc-processor/#1785633756be

At stock speeds they scored 1235 in Cinebench with 3000 memory, I get 1213 in Cinebench with 2133 memory. A whole 22 points less, they do say that with 3200 and above you'll get "much" better results but how much better are we talking? You'd get better results from overclocking the CPU or indeed buying a better CPU to begin with. By the way not trolling.

Dual Rank & Single Rank benchmarks with 8700K and Ryzen

https://www.computerbase.de/2018-04/amd-ryzen-2000-test/5/#abschnitt_singlerank_vs_dualrankram

Their 2700X works at stock speeds, no PBO etc. Look at the gaming benchmarks of a CPU that is boosting couple of cores at 4-4.1Ghz but has good memory.
It already starts with 15% lower boost clocks than the 8700K but the difference is tiny.....
Also look at the whooping boost between 2933 and 3466+timings.

kitfit1 · 3 Oct 2018 at 21:36

nlel1975 said:
From a bang for the buck perspective, I still don't see the point of buying super expensive RAM when that money is better spent on a better CPU or GPU. I got a really good deal on the 32GB 2133 stuff I have so am slightly biased. I paid £ 161.05 ex VAT for it with free delivery. I am in Jersey so no VAT either.

Here's my AIDA64 trial version memory benchmark...

This will no doubt highlight the differences between memory speeds.

The ram you buy dosn't need to be super expensive, you are correct on that view. But if you want to post a very poor result, then just carry on using cheap ram but don't expect anyone to gravitate to your own view.

To be honest, your Aida result is something that any normal person would be embarrassed to put on a public forum and i'm not sure i can understand why you did. Maybe you are trying to show how bad your ram is ? Or are you trying to prove that ram speed makes no difference ? In either case, you chose your own ram and as my AIDA64 result shows, Choice of ram and abilty to clock it makes a massive difference.

Hagar · 3 Oct 2018 at 22:19

nlel1975 said:
From a bang for the buck perspective, I still don't see the point of buying super expensive RAM when that money is better spent on a better CPU or GPU. I got a really good deal on the 32GB 2133 stuff I have so am slightly biased. I paid £ 161.05 ex VAT for it with free delivery. I am in Jersey so no VAT either.

Here's my AIDA64 trial version memory benchmark...

This will no doubt highlight the differences between memory speeds.

And this is one from my 1700 @3.6 with 16GB of 3333C14 memory (£149 incl. VAT back in the day).

Vince · 3 Oct 2018 at 22:55

An example just to show what difference speed has on latency, not at all comparable as it's running 8 sticks in numa so I can't push the speed that hard, I can probably do a run at 3466 but it wouldn't be stable.

The latency on the second run is almost as low as the ops running uma memory when latency should be much quicker in uma.

Frequency improves latency = faster system.

nlel1975 · 4 Oct 2018 at 06:28

If my RAM reads at 32GB/s which isn't slow then in two seconds it can read 64GB. I am prepared to wait an extra second.

blaineoliver · 4 Oct 2018 at 06:48

I love how the OP cannot admit that he didn't buy the best choice ram and doesn't this entire article trying to get others to agree so they didn't feel bad.

You bought 32GB of ram because it would make the system 'faster' than your last 16GB system, and then got ram with really slow speeds and has half the bandwidth of the faster 16GB kits others use. So it's not really faster. Granted you can fit more stuff in 32GB of ram which has its use cases but from the sounds of it you're not in any shape a power user.

You have a case of cognitive dissonance around this, or in other terms it's sour grapes. "used to refer to an attitude in which someone adopts a negative attitude to something because they cannot have it themselves"
https://en.m.wikipedia.org/wiki/Cognitive_dissonance

nlel1975 · 4 Oct 2018 at 06:54

Can't you just accept that I am happy with the RAM I bought? My old RAM was 1333 so it would be slower in AIDA64. Anyone who thinks 32GB/s read is slow is insane. Yes benchmarks make it look slow but am I bothered if a RAM intensive tasks take twice as many seconds to complete.

Vince · 4 Oct 2018 at 18:05

nlel1975 said:
Can't you just accept that I am happy with the RAM I bought? My old RAM was 1333 so it would be slower in AIDA64. Anyone who thinks 32GB/s read is slow is insane. Yes benchmarks make it look slow but am I bothered if a RAM intensive tasks take twice as many seconds to complete.

Surely that depends on use case? will your 32gb of slow ram serve you well for some games and browsing and even some ram heavy applications? Sure it will. Would it be fine for what I would wan't to do with the system? Absolutely not. I have some database workloads that destroy my system memory so going from 70GB/s read down to like half of that would increase my work time and when time = money that just wouldn't do.

But listen lets go back to the original question, in the first line you ask if infinity fabric is much faster with faster memory, people in here have answered that question backing up their findings with relevant benchmarks, read them and you can see for yourself that yes ryzen is faster with faster memory, whether or not that matters to you or not was never being questioned.

What I think you are struggling to grasp is that this forum is a forum full of enthusiasts, people that want the absolute best for their money given their specific workload. You asked a targeted question and got the right answer. Yes it does matter! The difference is probably an average of around 15% but can be anything up to what 40 or 50%, just looking at the memory benchmarks I posted above going from your exact memory speed / timings to 3000 mhz gained me what almost 40% in memory bandwidth and 20% in latency, in the right workload those numbers matter. If you are happy with your ram then fine be happy but accept that the answer is what it is. Personally I have fast 3466 rated memory (trident z) which I spent probably as much on as some spend on their whole system (£800 odd £) and even though for the most part I run at 3000 across 8 dims for me it was worth it. Horses for courses but whatever you do don't try and convince a forum full of enthusiasts that it isn't worth it for them just because it isn't for you. Generally speaking that isn't going to go down well

Think of it as walking into a lambo meet and trying to convince everybody there that a nissan would be better.

humbug · 4 Oct 2018 at 18:37

My own testing, rig in sig

nlel1975 · 4 Oct 2018 at 18:40

Fair enough but we all have different budgets, expectations, and uses. I had a i5-2500K system with 16GB of DDR3 1333 and a Radeon 7950 3GB. In all honesty it was still perfectly fine for my basic needs and I hadn't loaded a game for years.

Despite that I had a bit of spare cash so looked at the 3D Mark Timespy results to find a CPU that would double my CPU score of 2400 and a GPU that would double my graphics score of 2000 but I didn't want to spend a fortune. So at first I bought the Radeon RX 580 8GB for £ 187.49 ex VAT. I don't pay VAT as I live in Jersey. That got my graphics score up to 4300.

Then found the Ryzen 5 2600 fitted the CPU bill for £ 118 ex VAT free delivery that would get me 5800 CPU score. Then I selected a B450 motherboard to go with it, I wanted a full sized ATX board and at first chose the ASRock B450 Pro then changed my mind for the MSI B450 Tomahawk because of the VRM.

After that it was the memory decided to future proof it a bit by buying 32GB not because 32GB is faster that all depends on RAM speed but because I had 16GB before and didn't see the point of buying another 16GB. So I scoured my favourite retailer and found 16GB DDR4 2133 for £ 80.75 ex VAT each stick.

I knew it wasn't the fastest but the 16GB of DDR3 1333 had served me well before and I'm not a power user. In hindsight with the price of ocUK faster 3000 stuff for about £ 38 more I should have gone with that but I did save a bit of money and am unlikely to notice the speed difference as I'm not a power user and 32GB/s read speed is perfectly adaquate.

So I paid about £ 550 ex VAT for the components and sold my old CPU, Motherboard, RAM and graphics card for £ 110. So it cost me £ 440 net. This system doubles my 3D Mark Timespy CPU and GPU scores and I have double the RAM and it's faster too. Once again reiterate that 32GB/s read speed is adaquate for my needs although better to buy faster stuff for a little more but I wouldn't buy 3200 stuff.

At the end of the day I have no regrets and am happy with the new system and that's the main thing not what anyone else thinks.

Vince · 4 Oct 2018 at 19:13

humbug said:
My own testing, rig in sig

1.11v soc and I thought I was cheeky at 1v. Doesn't really matter mind as i'm pretty sure no board is going to do 3466 or even close fully populated, both boards I have literally state in the manual that 3200+ on 4 dimms but only 2133+ on 8. The asus x399-a prime I have won't even allow me 2933 with the same settings as the taichi, the tiachi I can boot all 8 at 3200 and 3466 but I just can't get it fully stable and given I often use it for stuff that I need to be 100% rock solid stable shooting for the stars hardly seemed worth it. I'll probably wait until 7nm TR and shove one of them into my board, who knows might be able to push the ram further then. Mind you given I have a pretty decent 1950x that just seems to like to be at 3.7 with a boost of 4.08/4.09 (what it seems to be every time I look) I think I might just stick until something out there just really sucks me in, I have been really tempted by 2950x but am really not sure I can justify the cash to change.

I'm thinking that chip to have will be the rumored 5/9 chiplet TR cpu I keep seeing pictures of. Surely that can't be that far away now, what do we recon 2nd H 2019 for 7nm TR?