9800X3D compared with other AMD CPUs at same clock speed

1

1_rick

[H]F Junkie
2FA
Joined
Feb 7, 2017
Messages
12,764
AMD's new Ryzen 7 9800X3D tested against 7800X3D, 9700X, 7700X: all locked at 4.8GHz

https://www.tweaktown.com/news/1016...d-9700x-7700x-all-locked-at-4-8ghz/index.html

ComputerBase tested the new Zen 5-based Ryzen 7 9800X3D processor to find out why the new 9800X3D is 14% faster than the 7800X3D... whether it was the new Zen 5 architecture, the CPU frequency, or the second-generation 3D V-Cache technology from AMD.

The tests found that it was more complex than just lowering the CPU clock speeds, with a bug preventing reviewers from adjusting the multiplier on the 9800X3D -- so they applied a 500MHz CPU clock offset -- achieving the effect ComputerBase intended.
Click to expand...

Also, since the 7800X3D doesn't support overclocking or multiplier adjustments, they just downclocked it 250MHz (article doesn't say how, probably BIOS?)

It doesn't say in so many words but I think this is the geomean of several games. Note: same clock speed, same RAM timings.
1731521809919.png
 
1_rick said:
AMD's new Ryzen 7 9800X3D tested against 7800X3D, 9700X, 7700X: all locked at 4.8GHz

https://www.tweaktown.com/news/1016...d-9700x-7700x-all-locked-at-4-8ghz/index.html



Also, since the 7800X3D doesn't support overclocking or multiplier adjustments, they just downclocked it 250MHz (article doesn't say how, probably BIOS?)

It doesn't say in so many words but I think this is the geomean of several games. Note: same clock speed, same RAM timings.
View attachment 691468
Click to expand...
4.8GHZ is the typical, stock boost of the 7800X3D. 5 GHZ only happens in very lightly threaded scenarios or if you setup PBO, curve optimizer, and have enough temperature headroom; it will mostly lock 5GHZ even in Cinibench.
 
That is odd.

The theory was that the reason the 9800x3d was so much faster than previous X3D variants despite Zen5 overall being a marginal improvement over Zen4 from a client load performance perspective was that putting the 3D V-Cache on the bottom allowed better heat dissipation and thus higher clocks, thus eliminating the gap in clock speed between X3D and non-x3d variants.

So, when you remove the clock speed difference and thus look solely at architectural/IPC differences a few things look really odd with this data.

1.) In launch testing the 9700x gained 2-3% on average over the 7700x if I recall. And this was after the Windows bug that was holding Zen chips back was removed.
Why is it over 10% here? Was the 9700x not boosting properly at stock settings?

2.) If we have isolated clock speed, and both chips have the X3D v-cache, how come we see almost 15% going from the 7800x3d to the 9800x3d?

Something isn't adding up here.

Unless there are some sort of actual architectural differences between X3D 9000 series and non-X3D 9000 series...
 
Zarathustra[H] said:
That is odd.

The theory was that the reason the 9800x3d was so much faster than previous X3D variants despite Zen5 overall being a marginal improvement over Zen4 from a client load performance perspective was that putting the 3D V-Cache on the bottom allowed better heat dissipation and thus higher clocks, thus eliminating the gap in clock speed between X3D and non-x3d variants.

So, when you remove the clock speed difference and thus look solely at architectural/IPC differences a few things look really odd with this data.

1.) In launch testing the 9700x gained 2-3% on average over the 7700x if I recall. And this was after the Windows bug that was holding Zen chips back was removed.
Why is it over 10% here? Was the 9700x not boosting properly at stock settings?

2.) If we have isolated clock speed, and both chips have the X3D v-cache, how come we see almost 15% going from the 7800x3d to the 9800x3d?

Something isn't adding up here.

Unless there are some sort of actual architectural differences between X3D 9000 series and non-X3D 9000 series...
Click to expand...
Zen 5 on consumer desktop is handicapped by using the same I/O die as Zen 4. As well as the same infinity fabric speed.

Vcache allows Zen 5's architectural changes to flex a little bit more. As cache hits effectively sidestep some of the bandwidth/latency handicaps.

If Zen 5 desktop had the new I/O die which Zen 5 server has, 9700X would have been a larger improvement over the 7700X, etc.

**IMO, AMD did this on purpose. Intel isn't pushing them. AMD can use-up the I/O dies which are already made and/or make Zen 5 more cheaply. It also allows X3D to look relatively more impressive.
And then later on, if needed, they can do a Zen5+ desktop, with the updated I/O die. Or they can wait until Zen 6 and more easily have a perceived generational improvement.
 
Last edited:
Looking at the RAM used for these tests I'd be surprised if the non 3d cache CPUs are running worth a damn. The best RAM doesn't help x3d as much but it does help non-x3d more and the slow RAM used is probably hurting the non-x3d CPUs rather badly.

This doesn't make sense of the large difference between the 7800x3d and 9800x3d. All I can guess is something fishy is going on with the testing or the specific games used.
 
SmokeRngs said:
Looking at the RAM used for these tests I'd be surprised if the non 3d cache CPUs are running worth a damn. The best RAM doesn't help x3d as much but it does help non-x3d more and the slow RAM used is probably hurting the non-x3d CPUs rather badly.

This doesn't make sense of the large difference between the 7800x3d and 9800x3d. All I can guess is something fishy is going on with the testing or the specific games used.
Click to expand...
Maybe because the 7800X3D is supposed to have locked clocks, and weird things happen when you forcefully mess with it.

They could have just... clocked all CPUs down to the level of the 7800X3D, no? Either way the original article mentions a number of bugs :https://www.computerbase.de/artikel/prozessoren/amd-ryzen-9000x3d-7000x3d-4-8-ghz-vergleich.90208/

Perhaps a test best redone at a later date, or with a different motherboard.
 
learners permit said:
Shit show test and no specifics. I see no reason to consider this valid in any way.
Click to expand...
Do not speek german but is that fair ?

https://www.computerbase.de/artikel/prozessoren/amd-ryzen-9000x3d-7000x3d-4-8-ghz-vergleich.90208/

Zarathustra[H] said:
That is odd.
Click to expand...
It is an sadly chrome translate does not help much understand the reviewer theory:
This is not because the cache, which is now located under the CPU chiplet instead of on it, has been revolutionized in other ways, but because the brakes on games that have been removed by the cache are having an even greater effect on the Zen 5 architecture.
other traductor say:
Interesting: Even in a "stock-to-stock" comparison — meaning with slightly higher CPU and memory clock speeds — the Ryzen 7 9800X3D and Ryzen 7 7800X3D show a fifteen percent difference in the reduced seven-game benchmark suite. That it isn’t more may be due to a measurement error of one or two percent, but ultimately this makes it clear once again: The new CPU gains its advantage in gaming from the cache compared to the old one, not from the higher (memory) clock speed.

This isn’t because the cache, which now sits under instead of on top of the CPU chiplet, has been revolutionized in other ways, but because the removal of bottlenecks in gaming by the cache shows an even greater effect with the Zen 5 architecture.

If the traduction is not far off, not sure they know either...

The test has baldur gates, starfield on a small 7 title, maybe it is a list of title where zen5 do particularly well ?
 
Missed the link to the original article so thanks for clarification. The tweak town reference to it was gibberish and annoying to read hence my scathing reply sorry. I've been trying to figure out why the 9800 vs 7800 has on some game engines such high % gains other than the obvious Mhz increase. When the first benchmarks were published I thought the results were odd as some games were near identical performance and others were dramatically higher than expected relatively. I'll have one next week so I can drill down into this topic a bit more and maybe gain some more insight into the differential factor.
 
LukeTbk said:
Do not speek german but is that fair ?

https://www.computerbase.de/artikel/prozessoren/amd-ryzen-9000x3d-7000x3d-4-8-ghz-vergleich.90208/


It is an sadly chrome translate does not help much understand the reviewer theory:
This is not because the cache, which is now located under the CPU chiplet instead of on it, has been revolutionized in other ways, but because the brakes on games that have been removed by the cache are having an even greater effect on the Zen 5 architecture.
other traductor say:
Interesting: Even in a "stock-to-stock" comparison — meaning with slightly higher CPU and memory clock speeds — the Ryzen 7 9800X3D and Ryzen 7 7800X3D show a fifteen percent difference in the reduced seven-game benchmark suite. That it isn’t more may be due to a measurement error of one or two percent, but ultimately this makes it clear once again: The new CPU gains its advantage in gaming from the cache compared to the old one, not from the higher (memory) clock speed.

This isn’t because the cache, which now sits under instead of on top of the CPU chiplet, has been revolutionized in other ways, but because the removal of bottlenecks in gaming by the cache shows an even greater effect with the Zen 5 architecture.

If the traduction is not far off, not sure they know either...

The test has baldur gates, starfield on a small 7 title, maybe it is a list of title where zen5 do particularly well ?
Click to expand...


I took 6 years of German from 1993-1999, followed by a 6 month exchange in Hessen, and none of that is helping me much here.

Maybe if I slow down and read it more carefully I'll be able to figure it out. No time for that right now though.
 
I wrote a big thing here on Zen5 and it’s memory controller being far better tuned for lower latencies at lower clocks but then I stumbled across this article from Chips and Cheese that did it far better.

https://chipsandcheese.com/p/zen-5-variants-and-more-clock-for-clock

It looks like the changes AMD made to the new memory controller, paired with the design changes for improving AVX512, and topped off with the new cache location allows for Zen5 to perform far better at lower clocks.

It also appears as if Zen5 has far better branch prediction which aids in memory usage which allows it to make better use of the cache as it has to spend less time fetching data due to its own mistakes.
 
Lakados said:
I wrote a big thing here on Zen5 and it’s memory controller being far better tuned for lower latencies at lower clocks but then I stumbled across this article from Chips and Cheese that did it far better.

https://chipsandcheese.com/p/zen-5-variants-and-more-clock-for-clock

It looks like the changes AMD made to the new memory controller, paired with the design changes for improving AVX512, and topped off with the new cache location allows for Zen5 to perform far better at lower clocks.

It also appears as if Zen5 has far better branch prediction which aids in memory usage which allows it to make better use of the cache as it has to spend less time fetching data due to its own mistakes.
Click to expand...

That makes sense.

They use the same base architecture for client (mobile & desktop), workstation, and server.

They are optimizing for server/enterprise, which makes sense because that is where the money is.

Server chips usually have more cores, and lower clocks.

So Zen5 is really a server first arch, that also works for client.
 
Zarathustra[H] said:
That makes sense.

They use the same base architecture for client (mobile & desktop), workstation, and server.

They are optimizing for server/enterprise, which makes sense because that is where the money is.

Server chips usually have more cores, and lower clocks.

So Zen5 is really a server first arch, that also works for client.
Click to expand...
If they have a benefit to the consumer parts and a big gain in the server space, which Intel still owns, this seems like the right strategy.
 
Not sure what people mean by new extension, but the benchmark here is playing video game....

Baldurs gates 3 it using something like AVX2 a little be that would be 10 years old...

Those game are made with running on console CPU also in mind who are quite limited new extension wise I think.
 
chameleoneel said:
**IMO, AMD did this on purpose. Intel isn't pushing them. AMD can use-up the I/O dies which are already made and/or make Zen 5 more cheaply. It also allows X3D to look relatively more impressive.
And then later on, if needed, they can do a Zen5+ desktop, with the updated I/O die. Or they can wait until Zen 6 and more easily have a perceived generational improvement.
Click to expand...
That's exactly what they will do. AMD sometimes cheapens out at odd times, but Zen 5+ should be pretty good.
 
You must log in or register to reply here.
Back
Top