AMD Ryzen 9 5900X

[jayjr1105]

5600X is nuts. Whipping a 10900K in most games while using HALF the power and nearly half the cost.

 

[hurakura]

The most logical explanation is that techpowerup labs are possessed by higher entities.

 

[RandallFlagg]

@W1zzard :

FWIW I think I found the reason for discrepancies in multiple review sites, which exist not only here but on other reviews as well. It's multi-faceted but it basically comes down to memory speeds. See #3 for the quick take.

#1 - Obvious one, if someone is using a 3080 or 3090 on the test bed. This doesn't really seem to be the main driver of what chip 'wins' but it does create bigger gaps between winners and losers.

#2 - Some sites are using much slower RAM on Intel platforms. Example AnandTech, who used DDR4-2933 on the 10900K but 3200 on AMD platforms. A more egreigious example is Cowcotland, who used DDR4-2666 on their 10600K and 2933 on the 10900K. This really only makes any sense if you are buying an OEM system, and if they are trying to 'simulate' an OEM buy they should probably be using all DDR4-2666 on all platforms. That would be irrelevant to 90% of the types of people who visit these sites though (as they don't buy OEM).

#3 - The big one. What I'm finding is that Zen 3 scales very well with higher speed RAM, and if it has high enough speed does indeed make a cleaner sweep in the benchmarks (even with a level playing field where Intel has the high speed RAM too). At sites like Guru3D where they used DDR4-3600 on both platforms, comparing to other reviews, it is apparent that Zen 3 *needs* fast RAM and will outperform Gen 10 in most tests if it has it. This is where you start to see the 5600X beat the 10700K and 10900K (@ stock clocks) in a lot of benchmarks even when both platforms are running DDR4-3600.

So, Zen 3 scales better with high speed RAM and 3600 seems to be the magic line.

Will be looking forward to the memory scaling article. Please do include something from Gen 10.

Edit: PCWorld also used DDR4-3600 on all platforms. In this case, Intel 10900K vs 5900X and 5950X in 5 games only won in one, Metro Exodus.

Ryzen 5000 Review: The best consumer CPU we've ever seen

AMD's historically good Ryzen 9 5900X and Ryzen 9 5950X desktop CPUs have trounced Intel's Core i9.

www.pcworld.com

AnandTech data is not matching Tec powerup.. any idea why ?

Because Anand ran their Intel chips at DDR4-2933, and their Zen 3 at DDR4-3200.

TPU ran them both at DDR4-3200.

 

[mkppo]

5600X is nuts. Whipping a 10900K in most games while using HALF the power and nearly half the cost.

It's bonkers. I bet it'll fly with a set of tuned 3800 RAM. It's looking like the CPU to have for gaming atm, and 3900 for gaming + productivity.

 

[jayjr1105]

It's bonkers. I bet it'll fly with a set of tuned 3800 RAM. It's looking like the CPU to have for gaming atm, and 3900 for gaming + productivity.

With upcoming AGESA update, 4000 might even be possible. Tuned 3800 with FCLK to 1900 will be a monster.

 

[Fasola]

@W1zzard :

FWIW I think I found the reason for discrepancies in multiple review sites, which exist not only here but on other reviews as well. It's multi-faceted but it basically comes down to memory speeds. See #3 for the quick take.

#1 - Obvious one, if someone is using a 3080 or 3090 on the test bed. This doesn't really seem to be the main driver of what chip 'wins' but it does create bigger gaps between winners and losers.

#2 - Some sites are using much slower RAM on Intel platforms. Example AnandTech, who used DDR4-2933 on the 10900K but 3200 on AMD platforms. A more egreigious example is Cowcotland, who used DDR4-2666 on their 10600K and 2933 on the 10900K. This really only makes any sense if you are buying an OEM system, and if they are trying to 'simulate' an OEM buy they should probably be using all DDR4-2666 on all platforms. That would be irrelevant to 90% of the types of people who visit these sites though (as they don't buy OEM).

#3 - The big one. What I'm finding is that Zen 3 scales very well with higher speed RAM, and if it has high enough speed does indeed make a cleaner sweep in the benchmarks (even with a level playing field where Intel has the high speed RAM too). At sites like Guru3D where they used DDR4-3600 on both platforms, comparing to other reviews, it is apparent that Zen 3 *needs* fast RAM and will outperform Gen 10 in most tests if it has it. This is where you start to see the 5600X beat the 10700K and 10900K (@ stock clocks) in a lot of benchmarks even when both platforms are running DDR4-3600.

So, Zen 3 scales better with high speed RAM and 3600 seems to be the magic line.

Will be looking forward to the memory scaling article. Please do include something from Gen 10.

Edit: PCWorld also used DDR4-3600 on all platforms. In this case, Intel 10900K vs 5900X and 5950X in 5 games only won in one, Metro Exodus.

Ryzen 5000 Review: The best consumer CPU we've ever seen

AMD's historically good Ryzen 9 5900X and Ryzen 9 5950X desktop CPUs have trounced Intel's Core i9.

www.pcworld.com

Because Anand ran their Intel chips at DDR4-2933, and their Zen 3 at DDR4-3200.

TPU ran them both at DDR4-3200.

GN also used 3200CL14 although 4 instead of 2 DIMMs and Zen 3 won everything they tested except RDR2 and AC: Origins.

 

[r9]

@W1zzard please fix the review so my world can make sense again.

 

[RandallFlagg]

GN also used 3200CL14 although 4 instead of 2 DIMMs and Zen 3 won everything they tested except RDR2 and AC: Origins.

What \ where are GN's test configurations?

 

[Fasola]

What \ where are GN's test configurations?

You can see it under the header and they also have a linked article with more in-depth details about their testing: https://www.gamersnexus.net/guides/3577-cpu-test-methodology-unveil-for-2020-compile-gaming-more

 

[HD64G]

No matter what (reviews' result variation), in a few weeks the best PC for gaming will have a Zen3 CPU with a Navi21 GPU (smart memory access anyone?). And that is why AMD launched their CPUs first. Great marketing tactics this time imho. And Zen3 CPUs reach higher single-threaded clocks than advertised (up to 5050Hz for 5950X).

 

[RandallFlagg]

You can see it under the header and they also have a linked article with more in-depth details about their testing: https://www.gamersnexus.net/guides/3577-cpu-test-methodology-unveil-for-2020-compile-gaming-more

View attachment 174714

OK that works.

That from GN actually validates my 3rd point from the post you responded to.

Compare that from GN where Intel won when both platforms used DDR4-3200 CL14 / RTX 3080 to this from PCWorld. PCWorld used DDR4-3600 on both platforms vs GN DDR4-3200. I'm seeing this same trend on benchmarks posted for 3dMark's site. There's a big jump in performance on Zen 3 with DDR4-3600 vs 3200. Not as much on Intel.

Something else to note, and you can really see this on Anandtech's review, GN used High / Custom settings while PCWorld used Ultra. Changing these settings often re-arranges who won - that's what you'll see at Anands if you look at the charts. I don't know why, maybe something to do with memory bandwidth to the PCI bus.

 

[agentnathan009]

On some sites it is slightly faster than Skylake, and how did AMD celebrate this? By charging $450 for 8 cores and $300 for 6 cores in 2020. The irony is very real.

Did you miss the part where the 5600X is nearly as fast as the 10900 and cheaper?

 

[RandallFlagg]

Someone with a 5950X just took #1 spot on 3D Mark Firestrike Physics - results limited to dual channel DDR4 to eliminate HEDT systems. Zen 3 still hasn't made it to top 100 on Time Spy CPU scores though.

Edit: There's a 5950X at #44 on Timespy CPU now, running DDR4-3800. The 10900K just above it is running DDR4-4700 OC to 5.6Ghz

 

[Fasola]

OK that works.

That from GN actually validates my 3rd point from the post you responded to.

Compare that from GN where Intel won when both platforms used DDR4-3200 CL14 / RTX 3080 to this from PCWorld. PCWorld used DDR4-3600 on both platforms vs GN DDR4-3200. I'm seeing this same trend on benchmarks posted for 3dMark's site. There's a big jump in performance on Zen 3 with DDR4-3600 vs 3200. Not as much on Intel.

Something else to note, and you can really see this on Anandtech's review, GN used High / Custom settings while PCWorld used Ultra. Changing these settings often re-arranges who won - that's what you'll see at Anands if you look at the charts. I don't know why, maybe something to do with memory bandwidth to the PCI bus.

View attachment 174718

It's pretty common knowledge that Zen gets more with faster memory. I was merely implying that there's more to TPUs outlier results than the lower frequency memory. We'll know once W1zzard finishes investigating and testing.

 

[dicktracy]

Did you miss the part where the 5600X is nearly as fast as the 10900 and cheaper? You guys are something else with your whining...

"AMD, please save us from Intel and Nvidia's expensive pricetags."
Yeah. The same guys that were once screaming that line in previous years are now fine with AMD's new prices. Things sure as hell didn't get cheaper with AMD, and in fact they're matching the competition in prices and sometimes even more so: $1000 GPU, $4000 HEDT CPU, and now almost $1000 for a mainstream CPU. Hypocrisy at its finest.

 

[ryzen5000]

I've been reading multiple reviews about Ryzen 5000 series, in most reputable websites Ryzen 5900X is proven to have a better overall fps result in gaming than i9-10900K.
I hope that when 5950X is reviewed here that you use a proper setup without so many bottlenecks (RAM, GPU, SSD...). to be able to stretch the potential of this CPU and get a fair test result.
Otherwise please don't test it, it would be spreading misinformation...
Ryzen 5000 is such big news for the PC market that I've registered on your forum today just to post this, taking it as my username.

 

[X71200]

3DM CPU scores... really relevant I mean for normal games when you have all cores loaded to the max getting 30-40 FPS.

Then comparing car part prices to cars, of course cars are more expensive, what else did you expect? You can compare pianos too you know. Gets more expensive easily. What even is your point?

 

[RandallFlagg]

It's pretty common knowledge that Zen gets more with faster memory. I was merely implying that there's more to TPUs outlier results than the lower frequency memory. We'll know once W1zzard finishes investigating and testing.

All platforms do, but Intel Gen 9 and Gen 10 got *a lot* more than Zen 2 from faster memory. 15% to be exact vs 8.2% on Zen 2 when you use DDR4-3200 CL 14 like TPU did.

It looks to me like that is entirely the source of the discrepancy. Look at the numbers below from the link I posted - note the 9900K is pretty close to a 10700K.

Starting from each platforms rated JEDEC standard RAM :

3900X : Going from JEDEC 3200 to 3200 CL14 gets 8.2%
3900X : Going from 3200 CL14 to 3600 CL16 gets 3%
3900X : Going from 3600 CL16 to (max stable) 3733 CL17 gets 2%

9900K : Going from JEDEC 2666 to 3200 CL14 gets 15%
9900K : Going from 3200 CL14 to 3600 CL16 gets 1%
9900K : Going from 3600 CL16 to (max stable) 4133 CL 17 gets 13%

Core i9-9900K vs. Ryzen 9 3900X mit RAM-OC in Spielen

Wie hoch ist die Spieleleistung wirklich? Core i9-9900K und Ryzen 9 3900X treten mit übertakteten Samsung-B-Dies im CPU-Limit in Spielen an.

www.computerbase.de

 

[Caring1]

I think I found the reason for discrepancies in multiple review sites, which exist not only here but on other reviews as well. It's multi-faceted but it basically comes down to memory speeds.

The points you made have been mentioned multiple times already on TPU,which the W1zz is already aware of.

 

[evernessince]

AMD has been pushing reviewers for higher memory speeds, because their product scales better with memory than their competitor's. Good for them, I'll still not use crazy fast memory in my reviews that nobody actually buys.

I'm pretty sure you are vastly under-estimating the number of people pairing their Zen CPUs with faster memory. 3600 can be had for barely anymore money than 3200 and for Zen. It's easily worth the performance bump.

 

[8BitZ80]

I've been reading multiple reviews about Ryzen 5000 series, in most reputable websites Ryzen 5900X is proven to have a better overall fps result in gaming than i9-10900K.
I hope that when 5950X is reviewed here that you use a proper setup without so many bottlenecks (RAM, GPU, SSD...). to be able to stretch the potential of this CPU and get a fair test result.
Otherwise please don't test it, it would be spreading misinformation...
Ryzen 5000 is such big news for the PC market that I've registered on your forum today just to post this, taking it as my username.

Now that the reviews are out, I've been wondering why AMD are saying the 5900X is the ultimate gaming CPU when the cheaper SKUs are just as good for gaming. I guess they make a tidy profit on it even though it's not overpriced from a per-core perspective.

 

[Mike2Fr]

Now that the reviews are out, I've been wondering why AMD are saying the 5900X is the ultimate gaming CPU when the cheaper SKUs are just as good for gaming. I guess they make a tidy profit on it even though it's not overpriced from a per-core perspective.

The extra cache compared to the 5800x and under might be useful with the 6000 series graphic cards. We will find out pretty soon.

 

[Xuper]

Multi-threaded energy efficiency of the Ryzen 9 5900X is now twice as good as the Core i9-10900K.

Holy Molly !

 

[Spanners]

Intel used to release new substantially faster CPU architectures without doing this: Sandy Bridge, Haswell, Sky Lake were all a lot faster than previous generation CPUs without price hikes and in certain cases even cost substantially less than their predecessors, e.g. the Intel Core i5-2500K was released for $216 while the Intel Core i7-920 cost $305.

What's bad for Intel (charging top dollar) is absolutely OK for AMD because ... because AMD. We get it. Let's see if AMD keeps raising prices for the next Ryzen series (Zen 4/DDR5) and what new crappy excuses AMD fans will come up with.

Your comment fails to acknowledge that Intel already has been forced to change strategy and pricing due to competition from AMD. Zen Gen1 offered worse IPC but previously unheard of core-counts for it's mainstream platform and price. It's not impossible that Intel would still be selling a 4/8 CPU as an i7 at the top of their range like the 7700K if AMD had gone bankrupt and wasn't actively competing. Now that AMD has the IPC advantage and a massive efficiency lead (I'm not hearing much talk about space heaters and power bills anymore) you expect them to continue to offer this at a discount to Intel? That's just completely ludicrous.

Of course Intel could offer a faster product at a discount and still make a larger profit. They were using process node advancements to make smaller dies but not increase core counts. They could get away with this due to AMDs lack of competitive products. The i7 920 was a 263 mm² die but the 2500k was 216 mm². AMD is using the same node here as the previous generation. They also used more artificial product segmentation (overclocking, hyper-threading, memory speeds) which is not so prevalent with current AMD lineups. In the end this is a business so maybe the people who were making those arguments about Intel previously were naive but that doesn't make your argument any more rational.

P.S. Please don't infer that people who don't agree with you are "fans" of AMD. I'm too old for that kind of malarkey and I'm a long term Intel shareholder. Thanks for the review btw W1zz, long term reader first time commenter.

 

[Deleted member 24505]

Your comment fails to acknowledge that Intel already has been forced to change strategy and pricing due to competition from AMD. Zen Gen1 offered worse IPC but previously unheard of core-counts for it's mainstream platform and price. It's not impossible that Intel would still be selling a 4/8 CPU as an i7 at the top of their range like the 7700K if AMD had gone bankrupt and wasn't actively competing. Now that AMD has the IPC advantage and a massive efficiency lead (I'm not hearing much talk about space heaters and power bills anymore) you expect them to continue to offer this at a discount to Intel? That's just completely ludicrous.

Of course Intel could offer a faster product at a discount and still make a larger profit. They were using process node advancements to make smaller dies but not increase core counts. They could get away with this due to AMDs lack of competitive products. The i7 920 was 263 mm² die but the 2500k was 216 mm². AMD is using the same node here as the previous generation. They also used more artificial product segmentation (overclocking, hyper-threading, memory speeds) which is not so prevalent with current AMD lineups. In the end this is a business so maybe the people who were making those arguments about Intel previously were naive but that doesn't make your argument any more rational.

P.S. Please don't infer that people who don't agree with you are "fans" of AMD. I'm too old for that kind of malarkey and I'm a long term Intel shareholder. Thanks for the review btw W1zz, long term reader first time commenter.

The point is though, AMD may have raised core counts, but for mainly gaming, they are irrelevant. Imo games still do not need more than a 6/8 core cpu at most. Everyone raved when Ryzen came out, oo ooo but it has 12 cores(well 6 with hyper) but games didn't use them anyway. Imo the one advantage of ryzen coming out was the big kick up the behind it gave Intel. They may have been caught with their pants down, but do not write them off(as every Ryzen fan seems to be doing) they have the budget, and experience to recover from this, even though it may take them a while after been caught snoozing.