***OFFICIAL*** Ryzen 5000 / Zen 3 Launch Thread REVIEWS BEGIN PAGE 39

[thesmokingman]

Anyone willing to guess how Mindfactory.de CPU sales percentages will look like in Q1 2021 ?
Ryzen 3000 pushed it to upper 80%, Ryzen 5000 will be complete brutality.

AMD Ryzen 5000 smashes processor sales records

Lots of people buying them, apparently

global.techradar.com

"The best CPU launch EVER EVER EVER"Mindfactory reports thousands of CPUs sold and inventory exhausted within 2 hours.#AMDRyzen5000 pic.twitter.com/M7sdnwXGj6November 6, 2020

 

[alexruiz]

Hey guys.

If quoting benchmark numbers from techpowerup, remember that those guys are "very unlucky" and always get the lowest binned Ryzens and Radeons...
Yes, their numbers for Ryzen and Radeon are always slower than everyone else (translation: they cannot even configure the AMD systems properly. I would not be surprised if they deploy all systems from a master image that already has patches for meltdown, so apply those also on AMD )

I personally ONLY use their numbers when comparing between product lines (between Radeons, or between Ryzens) but never to compare one product line to another.
So, other than a boatload of benchmarks and cool graphs, their data is unreliable

 

[Mopetar]

For some reason that video won't play for me right now, but Intel CPUs benefit from higher rank RAM as well. Did they test against how much Intel gains also?

If you're talking about the GN video, then yes the did test the Intel CPUs with the same four sticks of memory and specifically mentioned that it improves the performance on Intel CPUs as well. Most of the charts don't show the full Intel results (since the focus is on a lot of different memory timings for the 5600X to show which is best), but they do include some at the end that have 2x vs. 4x Intel comparison specifically.

 

[Hitman928]

If you're talking about the GN video, then yes the did test the Intel CPUs with the same four sticks of memory and specifically mentioned that it improves the performance on Intel CPUs as well. Most of the charts don't show the full Intel results (since the focus is on a lot of different memory timings for the 5600X to show which is best), but they do include some at the end that have 2x vs. 4x Intel comparison specifically.

So then the rank of memory would not explain TPU's weird results, right? If they switched to 4 sticks, Intel would increase in performance as well and you'd be back to where you began, unless Zen3 has a much higher benefit from additional ranks, the gap is too big for that to explain the difference.

 

[Timorous]

For some reason that video won't play for me right now, but Intel CPUs benefit from higher rank RAM as well. Did they test against how much Intel gains also?

HUB did briefly and found similar gains for Intel and AMD so while it might explain performance differences between reviews it does not explain why some places have the 5xxx series a decent margin ahead of 10th gen on average and others have them neck and neck. The most likely explanation I can think of is that GN used medium settings @ 1080p to shift more of the bottlenecking to the CPU where as TPU used maximum in game settings at all resolutions and HUB were somewhere in the middle.

 

[lightmanek]

so is someone going to run geekbench 5.3 on Zen3 now that it is release?

edit: and then update the title to say geekbench single core -> 1800

I'm on Windows with lots of services running in the background and only using 5900XT, so can't do 1800 on my first run, but here is my score anyway. Stock CPU, 1900IF 3800DDR4 CL18 18 18

https://browser.geekbench.com/v5/cpu/4658414

 

[DisEnchantment]

Zen3 support in LLVM is yet to come.

LLVM Lands Very Basic Support For AMD Zen 3 CPUs - Phoronix

www.phoronix.com

LLVM 10 adds some support for Zen2 and LLVM12 will add support for Zen3.
But LLVM10 is a wild ride, in some cases the performance bump is large in some cases it regressed for Zen2 wrt LLVM9. Unlike Intel which is fairly stable due to being largely unchanged.
LLVM11 should be more stable for Zen2 and it also supports the offloading of compute to GPU in systems like Renoir. So if not being explicit about compiler settings, performance can swing drastically.
Anyway, LLVM11 is not even stable yet, and LLVM12 is still many months away.

 

[therealmongo]

I'm on Windows with lots of services running in the background and only using 5900XT

 

[DrMrLordX]

Well, that's just a way to see it

Well, I did happen to say that it obviates the need for other gaming CPUs that might cost more. So you're only supporting my caveat. It's just not a "budget" CPU per se. It's a top-end gaming CPU at a midrange CPU price.

299 dollars to 249, it's 20 percent increase not 25.

Math is hard!

 

[uzzi38]

It finally happened.

1800 ST

But it's on Windows

5950X PBO ("MB limits") + Aorus B550i + 16GB 3800CL16LL + Custom Water 360mm. - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. B550I AORUS PRO AX with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

 

[Antey]

what did it cost?

"id": 69,
"name": "Processor Maximum Frequency",
"value": "5.17 GHz",
"ivalue": 5165,
"fvalue": 5165.0

 

[uzzi38]

what did it cost?

"id": 69,
"name": "Processor Maximum Frequency",
"value": "5.17 GHz",
"ivalue": 5165,
"fvalue": 5165.0

Yup. Though it's definitely not a stable 5.15GHz

 

[DisEnchantment]

It finally happened.

1800 ST

But it's on Windows

5950X PBO ("MB limits") + Aorus B550i + 16GB 3800CL16LL + Custom Water 360mm. - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. B550I AORUS PRO AX with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

System manufacturer System Product Name - Geekbench

Benchmark results for a System manufacturer System Product Name with an AMD Ryzen 7 5800X processor.

browser.geekbench.com

It is much easier on Linux

I noticed that the benchmark is easier on 8 cores. There is a lot of thread hopping and my 5950X cannot boost properly.

 

[Makaveli]

TPU redid their testing folks.

The results look very different when they use an ampere GPU and fast memory.

How is Intel Beating AMD Zen 3 Ryzen in Gaming?

Our Ryzen 5000 Zen 3 launch day reviews saw unexpected gaming FPS results many questioned. In this article, we will investigate these results in more detail, and do more testing to figure out what is going on. The results are surprising and set things right in the battle of AMD vs. Intel.

www.techpowerup.com

 

[moinmoin]

https://browser.geekbench.com/v5/cpu/4665766
It is much easier on Linux

I noticed that the benchmark is easier on 8 cores. There is a lot of thread hopping and my 5950X cannot boost properly.

In that test the frequency appears to be pretty stable at 4848MHz, so Windows needs 177-290MHz more to achieve a score similar to Linux.

GB5 is supposedly normalized to the score 1000 an i3 8100 achieves. I wonder if that target is already introducing abnormalities? In the ideal case Windows, Linux and x86 MacOS should achieve exactly the same 1000 on that i3 8100 setup but I have the feeling they wouldn't.

 

[Ajay]

"The best CPU launch EVER EVER EVER"Mindfactory reports thousands of CPUs sold and inventory exhausted within 2 hours.#AMDRyzen5000 pic.twitter.com/M7sdnwXGj6November 6, 2020

Sadly, I've read that the majority were snatched up by scalper bots

 

[Mopetar]

So then the rank of memory would not explain TPU's weird results, right? If they switched to 4 sticks, Intel would increase in performance as well and you'd be back to where you began, unless Zen3 has a much higher benefit from additional ranks, the gap is too big for that to explain the difference.

The GN Video didn't show off the full results for Intel, so I can't make a definitive claim as to how much difference the delta would change, but the GN results seemed to show the results of increasing to a 4x configuration on Intel made less of a difference than they do for AMD, but that may just be an improper takeaway on my part from incomplete data. That said, you should probably just watch the video for yourself instead of making bad assumptions about its contents in order to point to conclusion that you've already arrived at before viewing any of the evidence.

Yes, it's possible that there are other factors at play which explain the differences between the results that TPU has and those of other reviewers, but the most glaring is that Zen3 seems to suffer a bit more than Intel when only using a 8 GB x2 memory setup. It seems doubtful that TPU would have intentionally done that as they seem to test all of their CPUs with that particular setup. They would need to retest all of the Intel CPUs as well in order to make sure that the results are fair and representative, but it still does seem like the difference in their results can be largely explained by testing with two sticks of RAM instead of four.

W1zzard is going to be doing some additional testing over the weekend to see the effects of using 4 sticks of memory and look at a few other factors such as PCIe 3.0 vs. 4.0 to see if those make a difference. I think we'll have some definitive answers to the question of why TPU saw different results as well as a much greater understanding of how much some of the different configurations contribute to the differences we're seeing.

 

[amrnuke]

TPU redid their testing folks.

The results look very different when they use an ampere GPU and fast memory.

How is Intel Beating AMD Zen 3 Ryzen in Gaming?

Our Ryzen 5000 Zen 3 launch day reviews saw unexpected gaming FPS results many questioned. In this article, we will investigate these results in more detail, and do more testing to figure out what is going on. The results are surprising and set things right in the battle of AMD vs. Intel.

www.techpowerup.com

Pretty incredible.
Since ain't nobody gaming at 1080p (I'd hope) with a 10900K/5900X and 3090, I just averaged out the % difference between the 10900K and 5900X with 3800CL16 and 3090 at 1440p and 4K:

1440p avg: 5900X +3.24% over 10900K
(Note: 5900X only loses 1 game (Far Cry 5) by 10.3%, if you exclude that one outlier, its lead at 1440p is +4.7%)

4K avg: 5900X +2.56% over 10900K, wins in all games

It should be noted that the lead generated by them using better RAM and better GPU at 1440p and 4K is massive (10, 20, 30% at times). If you already have those two items, both the 10900K and 5900X are going to be close.

 

[Hitman928]

The GN Video didn't show off the full results for Intel, so I can't make a definitive claim as to how much difference the delta would change, but the GN results seemed to show the results of increasing to a 4x configuration on Intel made less of a difference than they do for AMD, but that may just be an improper takeaway on my part from incomplete data. That said, you should probably just watch the video for yourself instead of making bad assumptions about its contents in order to point to conclusion that you've already arrived at before viewing any of the evidence.

I don't appreciate the accusation of forming a conclusion based on an assumption as if I was refusing to watch the video. I tried to watch the video multiple times, it wouldn't load so I was relying on those who had watched it, hence why I was asking questions to try and get more information. Apparently some Youtube servers were down for several hours and it made many videos unwatchable. I was saying based on what I had seen from other reviews and what I was being told about that video, it wouldn't explain the difference but was asking for confirmation.

With that said, I can watch the video this morning and don't even see where they tested Intel at all in the video. Granted, I skipped through it due to time, but I didn't see any charts where they tested Intel comparatively. TPU has new results now so it's probably best just to read through their new write-up now and see what they found.

 

[moinmoin]

I tried to watch the video multiple times, it wouldn't load so I was relying on those who had watched it, hence why I was asking questions to try and get more information. Apparently some Youtube servers were down for several hours and it made many videos unwatchable.

Yeah, the video didn't load for me either at the time.

 

[Hail The Brain Slug]

I don't appreciate the accusation of forming a conclusion based on an assumption as if I was refusing to watch the video. I tried to watch the video multiple times, it wouldn't load so I was relying on those who had watched it, hence why I was asking questions to try and get more information. Apparently some Youtube servers were down for several hours and it made many videos unwatchable. I was saying based on what I had seen from other reviews and what I was being told about that video, it wouldn't explain the difference but was asking for confirmation.

With that said, I can watch the video this morning and don't even see where they tested Intel at all in the video. Granted, I skipped through it due to time, but I didn't see any charts where they tested Intel comparatively. TPU has new results now so it's probably best just to read through their new write-up now and see what they found.

At the end they show two games (Shadow of the tomb raider and F1 2019) with dual vs. quad config on the 10600k.

The 5600X gains 8.7% and 4.7% average fps while the 10600k gains 3.1% and loses 0.3%

They didn't really say why they only showed those two, but I assume it was a best case (tomb raider at 3.1%) and worst case (minor loss in F1) for Intel.

Zen 3 definitely appears to gain more going to a quad rank (dual rank per channel) configuration.

 

[Hitman928]

Just skipped through HWB video testing the rank increase where they did show 10900k results against 5900x results, at least in 2 games. In the two games tested, the 10900k actually gains more from increasing rank than the 5900x does, so there's some actual evidence that using 2 vs 4 sticks in the TPU did not make the difference.

Edit: @XabanakFanatik Thank you. As mentioned up above, HWB showed the 10900k getting more benefit. Either way the difference is small (a few percent at most) so again, it wouldn't explain the large gap between TPU results and other reviewers.

 

[Hail The Brain Slug]

Edit: @XabanakFanatik Thank you. As mentioned up above, HWB showed the 10900k getting more benefit. Either way the difference is small (a few percent at most) so again, it wouldn't explain the large gap between TPU results and other reviewers.

Yeah, GN only tested 10600k vs. 5600x with regards to ranks. I would think more cores would potentially benefit more. Although it's interesting to see the 5900x vs 10900k being inverse from the 5600x vs. 10600k.

I still want to see 2x1 vs. 4x1 vs 2x2. L1T has indicated it looks like Zen 3 benefits even more from a 2x2 config instead of a 4x1.

I hope L1T is working on a thorough comparison.

 

[Hitman928]

Reading quickly through the TPU update article, there is still a lot of weirdness going on and I don't think their conclusion bears out when looking at other sites. They basically conclude that AMD can win if you use fast 3800 MHz RAM so that the IF is clocked higher and latencies come down. However, other sites (see HWB) tested with 3200 MHz RAM and still had the AMD CPUs clearly on top, so this obviously isn't a requirement for Zen3 to have higher gaming performance. Their 2080 Ti performance versus GPU bottleneck graphs are strange (also, why do they go to 200% GPU bottleneck?) and I feel a bit contrived by only using a single frame rendered over and over again and then adding complexity to increase the GPU bottleneck. What constitutes added complexity is never explained either.

The switch to Ampere is interesting, perhaps some pcie4 benefit happening. It would be interesting to see a further investigation there. Guru3d also used a 2080 Ti and 2x8 RAM at sub 3800 MHz speed and also showed Zen3 as being faster in games at every resolution as well.

Edit: In the end, I do feel like Zen3 and Coffeelake are so close in gaming performance, it doesn't really make a difference unless there are certain games you play a lot that perform much better on one or the other. On average, though, I don't think you'll notice a difference going either way and it really comes down to price, power consumption, how much you value non-gaming performance, and supported features (e.g. pcie4). People will prioritize each of these differently and I think there are good products from both sides at certain tiers that you could make an argument for (and products that are now irrelevant as well with Zen3).

 

[Hail The Brain Slug]

Reading quickly through the TPU update article, there is still a lot of weirdness going on and I don't think their conclusion bears out when looking at other sites. They basically conclude that AMD can win if you use fast 3800 MHz RAM so that the IF is clocked higher and latencies come down. However, other sites (see HWB) tested with 3200 MHz RAM and still had the AMD CPUs clearly on top, so this obviously isn't a requirement for Zen3 to have higher gaming performance. Their 2080 Ti performance versus GPU bottleneck graphs are strange (also, why do they go to 200% GPU bottleneck?) and I feel a bit contrived by only using a single frame rendered over and over again and then adding complexity to increase the GPU bottleneck. What constitutes added complexity is never explained either.

The switch to Ampere is interesting, perhaps some pcie4 benefit happening. It would be interesting to see a further investigation there. Guru3d also used a 2080 Ti and 2x8 RAM at sub 3800 MHz speed and also showed Zen3 as being faster in games at every resolution as well.

Edit: In the end, I do feel like Zen3 and Coffeelake are so close in gaming performance, it doesn't really make a difference unless there are certain games you play a lot that perform much better on one or the other. On average, though, I don't think you'll notice a difference going either way and it really comes down to price, power consumption, how much you value non-gaming performance, and supported features (e.g. pcie4). People will prioritize each of these differently and I think there are good products from both sides at certain tiers that you could make an argument for (and products that are now irrelevant as well with Zen3).

It's because they only used 2x1 config, single rank per channel to test, even in their newly posted re-test.

And they completely glossed over the current news from several other techtube channels about how it makes a significant difference.