AMD Ryzen 7 9700X

[nguyen]

I had to go check some of the individual benchmarks and cross compare a little, and yeah that is exactly what is going on. Cyberpunk 2077 for example, without RT both Tom's and TPU show the 9600X performing very well - beating a 14700K by 8-13% depending which site does the test.

So they're not really in conflict, just none of them have a particularly large sample size of games and Zen 5 performance seems to vary pretty wildly from one game to the next.

HUB read our minds and posted a new video talking cherry picking games

I have always liked those 50 games benchmarks that HUB and TPU did, no room for cherry picking

 

[RandallFlagg]

Not true.

Ancient Replays and Level1Tech did test with a 7900XTX besides the 4090 and obtained different and interesting results.

Same for doing benchmarks using Linux instead of Windows.

The point is, the magical 4090 is not the end all, be all.

ComputerBase tests with a 7900 XTX and a 4090, and the results were not different in any significant way.

As far as testing with a more midrange card, well there are sites that do that too.

HotHardware uses a 4070 Super. Guess what their tests look like?

Obviously in this environment, winners and losers are more a result of chance and margin of error than actual performance advantages.

So I will say I think there is a place for this type of hardware being used in a review, however it is with budget CPUs and budget memory / motherboards.

The upper midrange CPUs are still too fast to differentiate with anything but high end GPUs.

That might change with the next gen of GPUs, as it appears they are advancing much faster than this gen of CPUs will.

 

[G777]

They have monolithic chips for laptops. Raven Ridge, Picasso, Renoir, Cezanne, Rembrandt, Phoenix, Strix Point.

I'm talking about Fire Range.

 

[phxrider]

ComputerBase tests with a 7900 XTX and a 4090, and the results were not different in any significant way.

As far as testing with a more midrange card, well there are sites that do that too.

HotHardware uses a 4070 Super. Guess what their tests look like?

Obviously in this environment, winners and losers are more a result of chance and margin of error than actual performance advantages.

So I will say I think there is a place for this type of hardware being used in a review, however it is with budget CPUs and budget memory / motherboards.

The upper midrange CPUs are still too fast to differentiate with anything but high end GPUs.

That might change with the next gen of GPUs, as it appears they are advancing much faster than this gen of CPUs will.

Thanks for the review tips, I will have to read them when I'm not at work and can spend more than a few minutes.

I wish they'd test at 1440 at the least though, I can only speak for myself, but I did not spend a grand on a video card to play at 1080, so 1080 results don't tell me anything other than theoretical differences in a lab environment, they don't tell me what the effect would be on anything I'd actually be playing.... I like that TPU even does 4K. Although, I think your statement on upper midrange CPUs is good advice.

 

[DemonicRyzen666]

ComputerBase tests with a 7900 XTX and a 4090, and the results were not different in any significant way.

As far as testing with a more midrange card, well there are sites that do that too.

HotHardware uses a 4070 Super. Guess what their tests look like?

Obviously in this environment, winners and losers are more a result of chance and margin of error than actual performance advantages.

View attachment 358588

So I will say I think there is a place for this type of hardware being used in a review, however it is with budget CPUs and budget memory / motherboards.

The upper midrange CPUs are still too fast to differentiate with anything but high end GPUs.

That might change with the next gen of GPUs, as it appears they are advancing much faster than this gen of CPUs will.

Oddly enough the notes on marketing slide point the use of not one 7900 XTX in the system but two 7900xtx in the system, which odd in self as I believe the list of games AMD they only like one or two games can even even use mutli-graphics card for things like mGPU.

 

[phxrider]

Checked out the HU's clarification video, Computerbase and HotHardware tests.

Computerbase tested at 1280x720, where the 7900xtx beat the 4090 like a red-headed stepchild, which shouldn't happen - so all that tells me is the 4090 is not well optimized for ultra-high framerates at a resolution no one ever plays it IRL.

HotHardware's test with the 4070S showed pretty much what I thought and why I suggested at least testing a few games with a lower CPU - their test showed if you have a midrange GPU, you really shouldn't be worried about your CPU and should save your money for a GPU if it's not fast enough for you.

HU gave props to TPU for a good selection of titles.

Still nothing tells me how CPU performance affects my 7900xtx at resolutions and settings people who spend a grand on a GPU actually play games at.

 

[RandallFlagg]

Checked out the HU's clarification video, Computerbase and HotHardware tests.

Computerbase tested at 1280x720, where the 7900xtx beat the 4090 like a red-headed stepchild, which shouldn't happen - so all that tells me is the 4090 is not well optimized for ultra-high framerates at a resolution no one ever plays it IRL.

HotHardware's test with the 4070S showed pretty much what I thought and why I suggested at least testing a few games with a lower CPU - their test showed if you have a midrange GPU, you really shouldn't be worried about your CPU and should save your money for a GPU if it's not fast enough for you.

HU gave props to TPU for a good selection of titles.

Still nothing tells me how CPU performance affects my 7900xtx at resolutions and settings people who spend a grand on a GPU actually play games at.

It is a bit annoying that almost everyone uses a 4090 for testing. You'd think some of these smaller sites would try to grab a niche, do something a little different for their testing.

Techspot did tests that're a bit old now, wish they would update this and make it not just about AMD. For the time, some important things could be gleaned.

4 Years of Ryzen 5, CPU & GPU Scaling Benchmark

We want to see how the Ryzen 5 1600X, 2600X, 3600X and 5600X compare in half a dozen games using the GeForce RTX 3090, RTX 3070, Radeon...

www.techspot.com

But then there's this :

Ryzen 5 5600X vs. Ryzen 7 5800X vs. Ryzen 9 5900X vs. 5950X: GPU Scaling Benchmark

In this review we want to see how the Ryzen 5 5600X, Ryzen 7 5800X, Ryzen 9 5900X and 5950X compare in gaming performance using the RTX...

www.techspot.com

 

[SL2]

Here's a Ryzen 3000 retest by GN from last month.

 

[Zach_01]

While I'm on telling you how to do your job (I'm kidding and I'm sure you welcome suggestions), it would be AWESOME if someone benched the games with something lower than a 4090 so all of us who don't have a $2000 GPU can see how much CPU choice actually affects us. Something like a 4070ti/7900xt that's closer to what most people have, that might still show a difference... Or even something in the $500 range. Just a thought - it drives me nuts that no one does this. The 4090 tests are great for showing the absolute theoretical differences, but I suspect that doesn't affect most gamers in the same way as the charts indicate.

Sorry if I sound critical, Techpowerup's reviews are awesome and are truly the first I look for when a new component comes out.

You are right, 99% of users do not have a 4090.
But think of it like this...

A review IMHO should always have at least 3 resolutions... 1080p, 1440p, 2160p and I will explain why this has value beyond of the preferred or available res each of us has.
You will notice that as the resolution increases and more load is dumped on the GPU the differences in FPS drop between almost all the CPUs.
So by increasing the resolution you are virtually making the GPU smaller and smaller.

So can you imagine what will happen if you use a real smaller GPU against the 4090 lets say at 1080p?

 

[Neo_Morpheus]

ComputerBase tests with a 7900 XTX and a 4090, and the results were not different in any significant way.

As far as testing with a more midrange card, well there are sites that do that too.

HotHardware uses a 4070 Super. Guess what their tests look like?

Obviously in this environment, winners and losers are more a result of chance and margin of error than actual performance advantages.

View attachment 358588

So I will say I think there is a place for this type of hardware being used in a review, however it is with budget CPUs and budget memory / motherboards.

The upper midrange CPUs are still too fast to differentiate with anything but high end GPUs.

That might change with the next gen of GPUs, as it appears they are advancing much faster than this gen of CPUs will.

So two reviewers I mentioned are wrong because the ones that you mentioned obtained different results that does satisfy your narrative?

ok…

Still nothing tells me how CPU performance affects my 7900xtx at resolutions and settings people who spend a grand on a GPU actually play games at.

I hate that rarely you will see those results on the initial CPU review, but as explained by many the reason is to isolate and concentrate on the CPU since at high resolutions, we are bottlenecked by the gpu.

It is a bit annoying that almost everyone uses a 4090 for testing.

Todays bribed influencers, formerly known as tech reviewers ( with some exceptions) need to comply with ngreedia demands of only showing their free 4090s or they will not get more free halo gpus.

 

[phxrider]

I hate that rarely you will see those results on the initial CPU review, but as explained by many the reason is to isolate and concentrate on the CPU since at high resolutions, we are bottlenecked by the gpu.

As I said previously, I understand why they do this and it is proper scientific method for comparing CPUs, but unfortunately those results don't tell most people if buying a new CPU (or CPU/MB/RAM combo) will help their system run games faster or if it would just be a waste of a lot of money.

Todays bribed influencers, formerly known as tech reviewers ( with some exceptions) need to comply with ngreedia demands of only showing their free 4090s or they will not get more free halo gpus.

I don't think TPU is bribed by Nvidia. They are simply using the fastest GPU to isolate CPU differences. Whether or not you agree with that methodology is another topic, but I don't think there's any nefarious reasoning behind it.

You are right, 99% of users do not have a 4090.
But think of it like this...

A review IMHO should always have at least 3 resolutions... 1080p, 1440p, 2160p and I will explain why this has value beyond of the preferred or available res each of us has.
You will notice that as the resolution increases and more load is dumped on the GPU the differences in FPS drop between almost all the CPUs.
So by increasing the resolution you are virtually making the GPU smaller and smaller.

Sure, but people don't just read reviews out of scientific curiosity about which CPU is fastest in an ideal environment - they read them to decide if spending the money on an upgrade would help their computer run stuff faster, "stuff" being whatever apps they care about whether that's games or MS Outlook or video encoding, etc. While the results are certainly valid for a pure CPU comparison, they don't provide context relevant to the average reader.

So what I can gather from the results is if I have a 4090, I can up my gaming FPS by going to 7800x3d (for example), and if I have a 5600x (also for example), I'm probably wasting my money on a 4090 unless I want to drop a 5800x3d in or get a whole new computer with a current gen CPU. However, if I had, say, a 4060, I would know nothing about whether a CPU upgrade would speed up my games based on what's on the review.

So can you imagine what will happen if you use a real smaller GPU against the 4090 lets say at 1080p?

Oddly, it looks like the 7900xtx spanks it at super low resolutions like 1280x720 (look at the Computerbase review). This is probably driver related IMO, as I doubt Nvidia spends a lot of time optimizing performance for resolutions that low.

View attachment 358674

I'm not sure why they did this test - most games on Zen 3 "X" CPUs run within a FPS or 3 of each other between CPUs. And Cyberpunk is particularly bad among games for showing CPU differences.

 

[SIGSEGV]

Thanks for the reviews and I will wait for the 9950X review.
I see a massive improvement over the predecessor compared with all Zen4 CPUs in non-gaming workloads.
Now I have a better picture of what's next for my 2700X replacement. I am surely going to pick non-3DX to support my work.

 

[Neo_Morpheus]

As I said previously, I understand why they do this and it is proper scientific method for comparing CPUs, but unfortunately those results don't tell most people if buying a new CPU (or CPU/MB/RAM combo) will help their system run games faster or if it would just be a waste of a lot of money.

Bro, did you read what i wrote?

First round of reviews rarely do that.

Usually comes after on following reviews.

I don't think TPU is bribed by Nvidia. They are simply using the fastest GPU to isolate CPU differences. Whether or not you agree with that methodology is another topic, but I don't think there's any nefarious reasoning behind it.

Did I specifically said that TPU is doing that?

Worse, do you really think that there aren't biased “reviewers” out there?

Because if do, then I do have a nice bridge to sell you.

 

[Zach_01]

As I said previously, I understand why they do this and it is proper scientific method for comparing CPUs, but unfortunately those results don't tell most people if buying a new CPU (or CPU/MB/RAM combo) will help their system run games faster or if it would just be a waste of a lot of money.


I don't think TPU is bribed by Nvidia. They are simply using the fastest GPU to isolate CPU differences. Whether or not you agree with that methodology is another topic, but I don't think there's any nefarious reasoning behind it.


Sure, but people don't just read reviews out of scientific curiosity about which CPU is fastest in an ideal environment - they read them to decide if spending the money on an upgrade would help their computer run stuff faster, "stuff" being whatever apps they care about whether that's games or MS Outlook or video encoding, etc. While the results are certainly valid for a pure CPU comparison, they don't provide context relevant to the average reader.

So what I can gather from the results is if I have a 4090, I can up my gaming FPS by going to 7800x3d (for example), and if I have a 5600x (also for example), I'm probably wasting my money on a 4090 unless I want to drop a 5800x3d in or get a whole new computer with a current gen CPU. However, if I had, say, a 4060, I would know nothing about whether a CPU upgrade would speed up my games based on what's on the review.


Oddly, it looks like the 7900xtx spanks it at super low resolutions like 1280x720 (look at the Computerbase review). This is probably driver related IMO, as I doubt Nvidia spends a lot of time optimizing performance for resolutions that low.


I'm not sure why they did this test - most games on Zen 3 "X" CPUs run within a FPS or 3 of each other between CPUs. And Cyberpunk is particularly bad among games for showing CPU differences.

Ok I'll break it down. I was going for making people think it for a moment but failed apparently

What I meant to say with this is...
As you are increasing resolution and keeping the same CPU/GPU it gets less and less significant what CPU you are using because the GPU is the bottleneck in this situation (keep these bold words on the side for later)

Example:

720p
See the difference?
The 7800X3D performs more than 2.0x against the last CPU on chart
More than double... its huge


4K
Notice how smaller the difference is
Now the 7800X3D is about 1.2x better than the last one
+20%


------------------------------------------------

Now lets go with different GPUs High to low end

When you decreasing the power of the GPU by testing a mid-range one and then a low end one... it gets less and less significant what CPU you are using because the GPU is the bottleneck
Reminds you anything?
Its exactly the same thing as increasing resolution on the same GPU
Thats why almost no one tests different GPUs on a CPU review. It has less value.

Small example
See how differences between CPUs get diminished as the GPU power drops
And mind you this is on 1080p. At 1440p differences are even smaller and so on


And to be honest there is some frenzy-ness going on about what CPU everyone should get for gaming especially for higher res.
Not to be mistaken... I'm not saying go get the cheaper (mind the 1% low too), but certainly you dont need the absolute top one(s) to enjoy gaming at high-er res.
Not even out of the top 10 CPUs.

 

[Wasteland]

As I said previously, I understand why they do this and it is proper scientific method for comparing CPUs, but unfortunately those results don't tell most people if buying a new CPU (or CPU/MB/RAM combo) will help their system run games faster or if it would just be a waste of a lot of money.

CPU reviews tell you whether a GPU upgrade will offer your system a worthwhile uplift. GPU reviews tell you whether a CPU upgrade will offer a worthwhile uplift. If your CPU can only hit, say, 80 FPS in your favorite game, then you won't benefit from a GPU that can output 160. And vice-versa. In short, each piece of hardware has a maximum throughput figure in a given context. That's what reviews are meant to isolate and exhibit--the subject hardware's potential, which incidentally also determines its useful life span. Thus, one must compare results for each major component to get the whole picture.

There's no point in reviewers trying to offer "realistic" CPU/GPU combos in their benchmarks, because everyone's configuration and use case differs, but if you really want to see GPU-bound results in a CPU review, you can often find them under the 4k results at e.g. TPU. W1zzard really does go above and beyond.

It is sometimes easy for us tech enthusiasts to forget that CPU reviews are intended first and foremost for people who are already looking to buy; they're not intended to convince you to replace your perfectly good CPU with one that performs 15% better, or w/e. If you have to choose between two CPUs that are currently on offer at a given price tier, then why wouldn't you want the one with better max throughput in your preferred use case? It will last longer. It will accommodate more/better GPU upgrades. Sure, maybe the slower CPU has other advantages--maybe it uses less power, or costs slightly less, etc--but you can't weigh those properly unless you also know the max throughput.

On a more general note, the era of assuming that gamers are GPU-bound is basically over. Ten or fifteen years ago, when everyone targeted 60 Hz (and when CPU tech was stagnant) it might have made sense to complain about "misleading" CPU-isolated benchmarks. Now it really doesn't. Even single-player games can often be heavily CPU bound, either because of a quirk in the engine (see Starfield) or because the player wants to push three-digit FPS to take full advantage of his fancy high-refresh monitor (and may be willing to dial down visuals to get there). The widespread use of upscaling tech (DLSS/FSR) also tends to push the burden back towards the CPU. Then there's a whole crew of competitive multiplayer gamers, who push absurd frame rates and mostly don't care about graphical fidelity. (These people existed ten years ago, of course, but they were stuck with a decade of what you might call "Sandy Bridge Plus" Intel quad cores, and didn't have fancy high-refresh rate monitors to reduce tearing when pushing stratospheric frames.)

Then there are CPU-intensive strategy games. Lately I've been playing something of a hybrid space-sim/4x called X4 Foundations. That game can look fantastic without stressing most any modern GPU really at all, but good luck finding a CPU that will run at 60 FPS stable once the universe gets cluttered. You don't need a 4090 to find CPU-bound scenarios.

 

[phxrider]

You guys all make some good points. Good discussion!

Did I specifically said that TPU is doing that?

No you did not, but since we are commenting on a review with a 4090, it was easy to assume it might be implied. I guess this is why they say we shouldn't ass/u/me, right? Sorry, my bad.

Worse, do you really think that there aren't biased “reviewers” out there?

Of course there are. But most of the big, well known ones go out of their way to be sure they aren't.

 

[doc7000]

A lot of reviewers have made a lot of good points however some reviewers have missed the plot when it comes to power efficiency, When both the 7700X and 9700X are run at 140 watts the 9700X beats it by a good margin in many workloads.

While yes at 70 watts the zen 4 and zen 5 are similar when it comes to power efficiency however a lot of cpu architectures are pretty efficient at lower power levels. If I am going to run a CPU at 140 watts then that is really where I would want the gains.

 

[RandallFlagg]

A lot of reviewers have made a lot of good points however some reviewers have missed the plot when it comes to power efficiency, When both the 7700X and 9700X are run at 140 watts the 9700X beats it by a good margin in many workloads.

While yes at 70 watts the zen 4 and zen 5 are similar when it comes to power efficiency however a lot of cpu architectures are pretty efficient at lower power levels. If I am going to run a CPU at 140 watts then that is really where I would want the gains.

Yeah, it's not like even an Intel user is running their CPU at 200W 24/7. The high power draw is only for relatively short bursts, and it doesn't add up to much.

And to that point, virtually every power efficiency test out there is hopelessly flawed.

The reason is straightforward - they ignore what PCs do 90+ % of their time. And that is - nothing.

They are at, or near, idle. Whether you're browsing here on this forum, or watching a YouTube video, on Facebook / Twitter / Instagram etc - your CPU is mostly idling.

Real workflow analysis of different chips, including the idle times, gives an entirely different perspective on 'efficiency' and 'power use'.

In short, in the real world, Zen is and always has been horribly inefficient due to very high idle power draws.

 

[THU31]

@RandallFlagg

I've seen that before and it didn't make sense to me. If Ryzen chips are consuming so much at idle, how is TPU getting sub-20 W consumption in certain single-threaded applications? They measure the power physically from the 12 V connector. Some CPUs even go below 10 W.

Either way, personally I only care about efficiency in terms of heat output. Idle consumption is irrelevant to me, and I'm happy if the CPU is peaking under 100 W in gaming (so i9 chips are automatically disqualified).

 

[RandallFlagg]

I'm not really going to argue a whole lot on this, it's all been done before.

Testing a PC under load and deriving real world efficiency from it is like testing a car's efficiency while doing 1/4 mile wide open throttle runs. Nobody does that in the automotive space because it is, frankly, stupid.

When sites like Consumer Reports or Motor Trend test a car's MPG, they run it through a mixed use route or in CR's case, a track they made specifically for that. It mimics various real-world driving, the vast majority of which is not done at full throttle.

That is what the video reviewer above calls a 'work flow'. His argument is pretty iron clad, especially when you consider that 'work flow' is exactly what happens when measuring efficiency / reliability of anything, ever, whether it's a transformer for a power substation or a car or a commercial truck. For some reason, this common sense methodology has never made it into the PC world. Too much dependence on clicks, too many false narratives.

 

[DemonicRyzen666]

I keep talking about the Zen5 x86 core. It's safe to say that Zen4 is an improvement over Zen3. But not about Zen5. Zen5's x86 core is a new microarchitecture design from the ground up.

View attachment 357876




View attachment 357877


View attachment 357878

the Data caches are 48Kbytes on Zen 5. & I believe they are 12 way?

 

[THU31]

I'm not really going to argue a whole lot on this, it's all been done before.

It's not my intention to argue, I'd just like to understand it.

So what about this graph? Where does the 20-25 W in those 3 apps come from?
Maybe that high whole system idle power usage is because of the motherboard/chipset, not the CPU?
People often say it's the I/O die that's wasting power, but wouldn't that show up in the EPS connector measurement? Unless the CPU also draws power from the 24-pin connector. The 8500G also has pretty high total system usage, but it doesn't have an I/O die.

 

[nexxusty]

9700X is just a server CPU in Desktop CPU. First AMD CPU with fully native 512bit.

Zen5's AVX512 Teardown + More...

Wow. What an article. Thank for posting this.

Completely different outlook on Zen 5 after reading. That's what raw data will do.

Just the fact that if you can cool Zen 5 enough, it won't throttle at all in AVX-512.... Awesome. Absolutely awesome.

 

[RandallFlagg]

It's not my intention to argue, I'd just like to understand it.

So what about this graph? Where does the 20-25 W in those 3 apps come from?
Maybe that high whole system idle power usage is because of the motherboard/chipset, not the CPU?
People often say it's the I/O die that's wasting power, but wouldn't that show up in the EPS connector measurement? Unless the CPU also draws power from the 24-pin connector. The 8500G also has pretty high total system usage, but it doesn't have an I/O die.

Every one of those benchmarks is essentially a measure of 'wide open throttle' for its respective task.

But is that how it happens in the real world? No, of course not.

Think about a software developer, and note the compile line on your chart shows ~90W. That chart suggests that anytime I'm coding (the workflow) I'm using 90W? Of course not. Most of 'coding' is reading, writing in a text editor, and looking up 'how' to code something on a browser. That's the *workflow*. It would only use 90W during the code compiles, which is like 1% of the time spent during development (if that).

So none of that represents a real 'workflow'. It represents max performance under a workload for each of those apps, which is typically something that only happens for a small fraction of the time spent on each of those.

 

[THU31]

You didn't even read my post. I CLEARLY asked about the THREE apps that use 20-25 W.
That's so much lower than the total system idle consumption of 80 W (I know that's before power conversion where some loss occurs). If the CPU is using 20-25 W, what is using the remaining 55-60 W? Is it all the other components in the system?

If so, then I don't really see how a difference of 22 W is relevant. I counted the cost of electricity in my country (where the average income is 3x lower compared to the west). At 12 hours every single day it comes to about $2 per month. In the grand scheme of the total power consumption in my house, it's basically nothing.

The chart from the video shows energy consumption of ~60 kWh per year. That's $15 over here, so even cutting it in half saves just $7.5 per year. I mean sure, one extra pizza per year sounds nice.