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

[Gideon]

Btw, yeshua, your previous nickname on AT forum was birdie?

Of course it was. In phoronix comment section he copy pasted the same negative first post he posted here under that name. Probably elsewhere too.

Im so tired of the pricing argument.

1. AMD is limited by TSMC's 7nm output
2. They will sell every CPU they can make until at least early next year.
3. Look at the retail prices now. If they'd lower the MSRP, the retailers would just pocket the difference.

It's not that hard. AMDs pricing is totally justified until supply normalizes


This stupid argument will only have any validity if AMD doesn't release 5600 and 5700X in Q1 (they will). They will also almost surely cut the price of 5800x (and maybe 5600x) some time before Rocket Lake.

 

[ondma]

I'd like to see a review where a 10700k is running JEDEC DDR4-2933 comparing to a 5800x running JEDEC DDR4-3200. Not everyone wants to assemble a computer and would rather buy a system from Dell, HP, etc and would most likely choose a system based on their choice of CPU primarily. These systems don't use XMP memory.

I cant imagine anyone buying a k processor and not getting a z motherboard. Hence they would be able to run XMP profile.

 

[Ranulf]

The heat density of these 7nm dies is significantly higher than that of the older dies built on 32nm. There are at least two published reviews(Anandtech and Computerbase) stating that the 5800X runs significantly hotter than the other chips, and now user reports from reddit are also coming in. All this indicates that heat is going to be a problem with the 5800X.

Hmm, so another reason to skip it. Given the earlier post about a user's temps, I wonder if its not enough thermal paste? Might need to put enough to fully cover the whole IHS or at least the areas above the dies.

 

[mmaenpaa]

Is a Noctua NH-U12A adequate for cooling a Ryzen 5800x? I purchased this cooler about 2 weeks ago to be installed in my next rig.

According to Noctua, it should be fine. If it can handle 3950X with "best turbo/overclocking headroom " it should handle 5800X/5900X also.

NH-U12A

The NH-U12A is the latest, 5th generation of Noctua’s award-winning NH-U12 series of 120mm size premium-quality quiet CPU coolers. Conceived to push the envelope of the 120mm form factor, it combines two of Noctua’s flagship model NF-A12x25 PWM 120mm fans with a completely redesigned heatsink...

noctua.at

 

[Racan]

Zen 3 loves dual-rank memory. This might explain the discrepancy in reviews. So dual rank 3200 CL14/3600 CL16 kits are optimal imo.

 

[Veradun]

Zen 3 loves dual-rank memory. This might explain the discrepancy in reviews. So dual rank 3200 CL14/3600 CL16 kits are optimal imo.

Even if by the smallest margin more ranks have always meant more performance @ same spec

 

[yeshua]

1. AMD is limited by TSMC's 7nm output
2. They will sell every CPU they can make until at least early next year.
3. Look at the retail prices now. If they'd lower the MSRP, the retailers would just pocket the difference.

1. NVIDIA is limited by Samsung's 8nm output.
2. They will sell every GPU they can make until at least early next year.
3. Look at the retail prices now. If they'd lower the MSRP, the retailers would just pocket the difference.

Looks like the three released Ampere cards should have been priced at the very least 2-3 times higher according to your logic. Nice try, nice try.

Zen 3 loves dual-rank memory. This might explain the discrepancy in reviews. So dual rank 3200 CL14/3600 CL16 kits are optimal imo.

My 3700X started to run MT workloads faster after I added two more memory sticks to my existing two sticks, so Zen 3 looks to be following the same route however it looks like this time around a performance difference is even higher.

 

[chrisjames61]

To be frank duopolies are not much better than monopolies (I've got a very strong suspicion AMD and NVIDIA have been cooperating their GPU pricing strategies for the past at least four years) and I'm really sad we have just two x86 vendors and basically two GPU vendors - it's way too easy to collude without leaving any trail.

Birdie, why don't you take your trolling and conspiracy theories back to Phoronix site and stop ruining threads here?

(Edit). Now he is at TechPowerUp pasting verbatim his posts here and Phoronix.

 

[Racan]

Even if by the smallest margin more ranks have always meant more performance @ same spec

Of course, but Zen 3 seems to benefit more than Skylake or Zen 2.

 

[Veradun]

Of course, but Zen 3 seems to benefit more than Skylake or Zen 2.

That's due to the fact that 3200 has more headroom for improvement on Zen3 than on Zen2, so going dual rank make it benefit more (like it would do going from 3200 to 3466)

 

[coercitiv]

Of course, but Zen 3 seems to benefit more than Skylake or Zen 2.

We barely got through the claim that Skylake benefits more than Zen 3 from fast memory, and now we're getting ready to do it all over again... in reverse.

 

[Det0x]

Zen 3 loves dual-rank memory. This might explain the discrepancy in reviews. So dual rank 3200 CL14/3600 CL16 kits are optimal imo.

Some screenshots from the video:


Intel also don't benefit in the same way in all benchmarks as can be seen here here: (same score for 2x8 and 4x8

 

[Racan]

We barely got through the claim that Skylake benefits more than Zen 3 from fast memory, and now we're getting ready to do it all over again... in reverse.

GN in these two cases do show Zen 3 gaining more but maybe their results are wrong somehow or these games are an outlier.

20:29 - Intel 2 vs. 4 Sticks Memory (Tomb Raider)
20:56 - Intel F1 2020 2 vs. 4 Sticks

 

[Racan]

Techpowerup were testing with single rank

 

[JoeRambo]

Of course, but Zen 3 seems to benefit more than Skylake or Zen 2.

Zen3 benefits more cause individual cores have massively improved available memory parallelism. For example page table walkers were expanded - guess what, when TLB miss happens, it usually means that L3 cache will miss as well, so ZEN3 will have more parallel request to memory happening sooner.
Total of 192 of L3 misses can be outstanding compared to 96 on ZEN2. + there are cascading events, like taking less L3 misses means that memory misses that were separated by more time happen with less separation now.

More items on average in memory controller queue => ZEN3 is more sensitive to memory latency and parallelism. Having more ranks ( either from DR or from more 1R DIMMs on the channel ) increases parallelism.

 

[yeshua]

RetiredEngineer® @chiakokhua Nov 7
Official figures from AMD:

Zen2 CCD: 3.9B xtors, 74 mm², TSMC N7

Zen3 CCD: 4.15B xtors, 80.7 mm², TSMC N7

Common cIOD: 2.09B xtors, 125 mm², GF 12nm

Physical dies a bit bigger perhaps due to remains of 'dicing streets', and/or distortion to photo after perspective correction.

 

[amrnuke]

1. NVIDIA is limited by Samsung's 8nm output.
2. They will sell every GPU they can make until at least early next year.
3. Look at the retail prices now. If they'd lower the MSRP, the retailers would just pocket the difference.

Looks like the three released Ampere cards should have been priced at the very least 2-3 times higher according to your logic. Nice try, nice try.

That's what-about-ism and irrelevant.

You are making a claim that the 5600X (and it seems the entire Zen3 line) is unfairly priced. You have yet to provide any logic behind that argument. You keep running with this price increase thing, and citing the 5600X (lowest-end release day product, $299) vs 3600 (cheapest prior-gen chip with same core count, $199). The Intel analog is 10600K ($265, 6 cores) vs 9400F ($182, 6 cores). So, tell me, why should AMD, a for-profit company with a superior product, not price their product higher than the nearest competing analog (10600K, $265)? AMD has a 12% CPU task and 4% gaming 720p performance lead over it, so a 13% price difference, giving nonlinear scaling of cost and performance, seems reasonable. As evidenced by the product being out of stock almost everywhere.

 

[RickITA]

Hi, this is my first post. For me it's amazing that @4450MHz a single 5600X core requires 10.2W, while the 5800X core 14.6W on average. Should not be the same core? Why such a large difference?

Point is, AMD is not selling (yet?) the CPU I wanted, the 5700X (8 cores, 65W) and I don't know what to do. Should I go for a 5600X? Do you think a 5700X will be available soon? I use Stata (statistic / data science software, single core) and Mathematica (here about half of my applications are also single core).

I have small mITX case (about 13.4 liters), a Gigabyte B550i Aorus Pro AX and 32GB RAM.

 

[amrnuke]

Hi, this is my first post. For me it's amazing that @4450MHz a single 5600X core requires 10.2W, while the 5800X core 14.6W on average. Should not be the same core? Why such a large difference?

Point is, AMD is not selling (yet?) the CPU I wanted, the 5700X (8 cores, 65W) and I don't know what to do. Should I go for a 5600X? Do you think a 5700X will be available soon? I use Stata (statistic / data science software, single core) and Mathematica (here about half of my applications are also single core).

I have small mITX case (about 13.4 liters), a Gigabyte B550i Aorus Pro AX and 32GB RAM.

If single-core speed is all you're after, there would seem to be little benefit to waiting for a 5700X, since the 5600X is within 6-7% of the 5950X in SPECfp and within 3% in SPECint for single-core speed, and within 5% of the 5950X in GB5 single-thread. However, if the half of your Mathematica work scales really well with cores, you're right, it may be worth waiting.

 

[Panino Manino]

Zen2 CCD: 3.9B xtors, 74 mm², TSMC N7

Zen3 CCD: 4.15B xtors, 80.7 mm², TSMC N7

Is this for real?
19% IPC with just 250 million more transistors?

 

[lobz]

Hi, this is my first post. For me it's amazing that @4450MHz a single 5600X core requires 10.2W, while the 5800X core 14.6W on average. Should not be the same core? Why such a large difference?

Point is, AMD is not selling (yet?) the CPU I wanted, the 5700X (8 cores, 65W) and I don't know what to do. Should I go for a 5600X? Do you think a 5700X will be available soon? I use Stata (statistic / data science software, single core) and Mathematica (here about half of my applications are also single core).

I have small mITX case (about 13.4 liters), a Gigabyte B550i Aorus Pro AX and 32GB RAM.

I'd 10000000% go for a 5900X, even if it means another month of saving up for it. AMD allows for limiting the TDP at almost whatever you'd like, that way MT performance will still be miles ahead a 8-core Zen 3, and it doesn't affect ST perf.

 

[yeshua]

Hi, this is my first post. For me it's amazing that @4450MHz a single 5600X core requires 10.2W, while the 5800X core 14.6W on average. Should not be the same core? Why such a large difference?

Point is, AMD is not selling (yet?) the CPU I wanted, the 5700X (8 cores, 65W) and I don't know what to do. Should I go for a 5600X? Do you think a 5700X will be available soon? I use Stata (statistic / data science software, single core) and Mathematica (here about half of my applications are also single core).

I have small mITX case (about 13.4 liters), a Gigabyte B550i Aorus Pro AX and 32GB RAM.

AMD motherboards allow to set the desired TDP for a CPU (called PPT in BIOS), so you could buy the Ryzen 5800X and limit it to 88W TDP (which is what the 3700X operates with). If you're not in a rush wait for the 5700X - I still hope it will be released but I guess that's not going to happen this year. However even with this limit enabled the 5800 will still be a lot hotter than all other Ryzen 5000 CPUs in low/light threaded workloads.

 

[Yosar]

Zen 3 just isn't all that great with anything lower than DDR4-3600. I'm a bit disappointed that none of these 'professional review' sites seem to have caught onto it. TPU has always used DDR4-3200 CL14 on both platforms, which normally favors Zen 2 a bit. However, it wasn't a flattering setup for Zen 3.

I seriously doubt. You base your opinion on TPU review and their 'explanation'. It's really not that hard to find review where results are totally different from TPU and using slower memory.
Here is a test of Witcher 3 with memory speed of 3200 on both platforms with not that great timings of CL 16-18-18-36.

Test procesora AMD Ryzen 9 5900X - Premiera architektury Zen 3 | PurePC.pl

Test procesora AMD Ryzen 9 5900X - Premiera architektury Zen 3 (strona 25) Test wydajności procesora AMD Ryzen 9 5900X (Zen 3 Vermeer) vs Intel Core i9-10900K (Comet Lake) w grach komputerowych i programach. Który procesor jest szybszy?

www.purepc.pl

As a bonus you know they tested in Novigrad so CPU bound scenario. In TPU review you actually don't know what they tested. They could have tested in some anonymous forest where GPU is bottleneck.
You see a difference? Ryzen 5000 wins without question. The same processors that lose in TPU review.
And seeing that in TPU review results between 1800X and 5800X are basically actually in the margin error (not a big difference) I suspect they are just incompetent and tested GPU bound scenario not CPU.
So their review is practically worthless.

Another one. This time with RAM 3200 MHz for Ryzen and 2933 MHz for intel (officially supported).

Premiera procesorów AMD Zen 3 - test Ryzen 9 5900X

Wiemy, że czekaliÅcie na test tego procesora! W nasze rÄce wpadÅ 12-rdzeniowy i 24-wÄtkowy model Ryzen 9 5900X â to prawie najwydajniejsza jednostka z nowej generacji Zen 3, która ma zainteresowaÄ graczy i entuzjastów.

www.benchmark.pl

Ryzen 5000 still wins.

I don't trust TPU reviews anymore. Why? Look at comparison between GPU from this page. 5700 XT vs 3080

AMD Radeon RX 5700 XT Specs

AMD Navi 10, 1905 MHz, 2560 Cores, 160 TMUs, 64 ROPs, 8192 MB GDDR6, 1750 MHz, 256 bit

www.techpowerup.com

It should be 78% difference. Right? So I go to the 3080 review

NVIDIA GeForce RTX 3080 Founders Edition Review - Must-Have for 4K Gamers

NVIDIA's new GeForce RTX 3080 "Ampere" Founders Edition is a truly impressive graphics card. It not only looks fantastic, performance is also better than even the RTX 2080 Ti. In our RTX 3080 Founders Edition review, we're also taking a close look at the new cooler, which runs quietly without...

www.techpowerup.com

65% vs 100% in 1080p so only about 54%.
Ok so maybe 4 k? 50% vs 100% so 100% difference.

So let's take 2080 Super vs 3080 comparison. According to main page for 2080 Super it's a 46% difference.

NVIDIA GeForce RTX 2080 SUPER Specs

NVIDIA TU104, 1815 MHz, 3072 Cores, 192 TMUs, 64 ROPs, 8192 MB GDDR6, 1937 MHz, 256 bit

www.techpowerup.com

We go to the same 3080 review and here we are:
In 1080p 77% vs 100% so about 30%
In 4k 64% vs 100% so about 56%

Nothing is correct. So I thought maybe these are data from 1440. OK Once again.
5700 XT
57 % vs 100% so 75-76% difference
2080 Super
70% vs 100% so 43%
Still no luck.

So you look at data and wonder, where the hell do they come?

Oh, and you have remark at the bottom: "Based on TPU review data: "Performance Summary" at 1920x1080, 4K for 2080 Ti and faster."
So for both 5700XT and 2080 Super it should be 1080p data. It's clearly not.
That's why I don't trust them.

And personally I don't think you should test anything on games older than 1 year. With exception of still super popular multiplayer games.
You buy processor or GPU for future games, not past games.
And people still test processors on GTA V, a game that engine breaks at high fps as Wendell from Level1tech already proved.

 

[Daneden]

Anybody here already thought of the usage of infinty cache for future Zen IO dies?
If this really triples the available bandwith like for the radeon cards, it would just be a huge improvement.

 

[Mopetar]

Speaking of testing resolutions and all of this.

Absolute most reviewers do a very bad job of testing because:

1) Rich people/tech enthusiasts will pair high end CPUs with high end GPUs while running their games at high resolutions (1440p/4K). This is such a small minority it's just laughable - less than 7% of people. And this group is also very very vocal because most average people neither assemble their PCs, nor read reviews or participate in forums (or comments sections) like this one.

If the only people who read reviews and post on forums are primarily in the group that buys the high end products for the type of situations described in the reviews, it's hard to argue that the reviewers s are doing a bad job. They know who their readers are and what they want. If anyone wanted to read reviews of low end part combinations for the masses the internet would be full of sites doing it. I'm sure there are a few out there, but I have a feeling the readership is as big as people doing reviews of 20 year old technology for the sake of nostalgia.

Is this for real?
19% IPC with just 250 million more transistors?

I would imagine that they saved a few in moving from the 4 core CCX to the 8 core CCX design. If that also happened to be the biggest bottleneck for performance it seemed to work out rather well for them.