Three times the charm - New AMD CPU announcement + big Navi Teaser

[leadeater]

50 minutes ago, LAwLz said:

They don't clock higher.

Most of the chips actually has a slightly lower base clock but a slightly higher boost clock. So for short burst or lightly threaded workloads we will probably see slightly more than 19% performance uplift, but in sustained workloads where all cores will be fully loaded, we will probably see less than 19% performance increase.

Lower base clock doesn't mean they clock lower, that is most likely purely for idle power management only. AMD have explicitly said they clock higher and there is zero reason to think otherwise.

[LAwLz]

12 minutes ago, porina said:

Don't know if you saw my post... somewhere, where I speculated if Zen 3 doing more work might mean it will take more power per clock, and do so at a lower clock. Having said that, 100 MHz drop on 3.5 GHz is less than 3% and the other conditions will be lower still. The IPC increase still dominates over that, and perf per watt is claimed to be improved similarly.

I got to that post after I had written that respond.

Like you said, without benchmarks we don't actually know what clock speeds will be like. My guess however is that they will be more or less the same as Zen2 products, and if that's the case the performance increase we will see will be the IPC one, which is (probably) at best 19%.

 

I feel like people are starting to grasp at straws to make the 5000 series seem better than it really is when compared to the 3000 series.

• First it was incorrect pricing for the 3000 series, statements like the pricing only being 50 dollars higher (when it was actually more like 90 to 100 dollars).
• Then they tried to argue that the 3000 series were more expensive than I was quoting because of sales, but as we can clearly see this is not the case at all.
• For a short while the argument was "if you don't like it don't buy it!" and "if you like the 3000 series so much then why don't you marry it!", but that argument didn't hold much water when they were equally guilty of pushing their agenda that "the 5000 series is great and the 3000 series is not".
• Then they switched to discussing overall budget of the PC because then you can hide the price of the processor. All of a sudden they have a clever way of reducing the % of the cost for the CPU, but they don't have to change the performance gain number! 
• Then once people realized that argument wasn't working they switched to talking about how the clock speed will be higher and therefore the performance will be more than 19%!

Soon they will come up with some other argument for why the 3000 series is bad and the 5000 is great... Because clearly just saying "the new processors are bad value" is not an option.

[Bombastinator]

5 minutes ago, LAwLz said:

That's one way of looking at it, but I think the problem of looking at it like that is that you won't actually get a flat 20% performance uplift overall.

If you are primarily running tasks that are GPU bound (like gaming) then 20% higher CPU performance might only translate to 5% higher performance in your most commonly run tasks. 

Like I wrote in my other response, every dollar spent on your CPU is a dollar you are removing from your budget on other parts, like your GPU. If you are gaming, spending 154 dollars more on your CPU might result in you getting a worse GPU, which will have a far bigger impact on gaming performance.

 

You also have to consider that the 8 core 5800X costs more than the 12 core 3900X (by 20 dollars). 

Going from the 3700X to a 5800X increases the overall PC's cost by 10.7% for ~20% higher CPU performance.

Going from the 3700X to a 3900X increases the overall PC's cost by 9.4% for ~50% higher CPU performance.

 

Although it's a bit more complicated than that because the first option is 20% higher single and multithreaded performance while the second option is primarily multithreaded performance. But if you're looking at 8 core processors then chances are you are running programs that are mostly multithreaded (or multiple single threaded applications) and can take advantage of the extra cores.

But my point is that you always get better (multithreaded at least) performance for your dollars by going with a 3000 series chip over a 5000 series chip.

And this is why independent reviews are king and why we’re all here in the first place.  Go Linus!

 

It may turn out I’m lucky and a 5600x will be enough for me to run whatever they can put on an XboxX, in which case it’s what I buy.  I don’t know yet though.  It might take a 3700(x) so I’m holding fire.  My greying 4770k warhorse still plays witcherIII good and it’s supposed to do OK with cyberpunk 2077, so I got time to find out. I will buy the cheapest machine that runs what I need it to and I don’t much care if it’s intel or AMD or whoever it is Cyrix is playing footsie with in China.

[leadeater]

52 minutes ago, LAwLz said:

I understand what you mean, but I don't think you can look at it that way.

Let's take the 3600 vs the 5600X as an example. It is roughly 50% more expensive (100 dollars) for roughly 20% higher performance. If you're building a 1000 dollar computer then sure, you could say the difference is only 10% for 20% higher performance, but that's not really the case because you are using the generalized budget for the PC, and a specific system metric when measuring performance.

A 20% better CPU doesn't perform to 20% better performance overall. Far from it.

If you are building a gaming PC, a 200 dollar CPU with a 500 dollar GPU is way better for gaming than a 300 dollar CPU with a 400 dollar GPU. That's how you need to look at things.

Every dollar spent on one component, is one dollar you are taking from another component.

$500 is the cut off point where the price increase results in a lower total system value, going from that $500 to $600 and assuming 20% performance gain is net no change in system performance per dollar. So this lower price point is worst case, so a 500 GPU would immediately mean the 5600X is the better purchase.

 

Like I also stated this applies if you intend to keep that system for a decent length of time where that extra CPU performance is going to matter, or if you upgrade the GPU years down the track and nothing else. If you buy a 3600 it will become a performance limiter, as it already can be right now, and replacing it later with a 5000 series CPU (no matter which or at what price) would be an overall worse value system than if you had just brought the 5000 series in the first place.

 

So either don't buy at all and wait, or spend as little as possible so as little is sunk in to a system with a short intended life span.

[thechinchinsong]

I guess we'll see the real value of these processors once 3rd party reviewers get their hands on them. Much like every like every processor launch since the beginning of time :).

[LAwLz]

1 minute ago, leadeater said:

Lower base clock doesn't mean they clock lower, that is most likely purely for idle power management only. AMD have explicitly said they clock higher and there is zero reason to think otherwise.

Base clock isn't for idle management. The frequency the chips can go down to is far lower than the base clock. Pretty sure Ryzen processors can go sub 1Ghz for power saving reasons.

A lower base clock is a bit worrying to me because AMD didn't just lower it for fun. I wouldn't be surprised if these new CPUs kick out a bit more heat when all cores are loaded and that's the reason why they had to lower the guaranteed frequency (base clock).

Do Ryzen chips even reach their max turbo with all cores fully loaded? From what I've heard the 3950X even with a pretty beefy cooler usually doesn't reach 4GHz when all cores are loaded. Of course, this is more of an issue with high core count parts, but the issue is still there for the lower count parts.

You shouldn't expect to hit max turbo on all cores for any kind of sustained workload. Your clock speed will be lower than that. How much lower remains to be seen.

[divito]

3 minutes ago, LAwLz said:

A lower base clock is a bit worrying to me because AMD didn't just lower it for fun.

I'd hardly call a 100MHz reduction in base across the lineup as worrying.

[leadeater]

7 minutes ago, LAwLz said:

Base clock isn't for idle management. The frequency the chips can go down to is far lower than the base clock. Pretty sure Ryzen processors can go sub 1Ghz for power saving reasons.

A lower base clock is a bit worrying to me because AMD didn't just lower it for fun. I wouldn't be surprised if these new CPUs kick out a bit more heat when all cores are loaded and that's the reason why they had to lower the guaranteed frequency (base clock).

Do Ryzen chips even reach their max turbo with all cores fully loaded? From what I've heard the 3950X even with a pretty beefy cooler usually doesn't reach 4GHz when all cores are loaded. Of course, this is more of an issue with high core count parts, but the issue is still there for the lower count parts.

You shouldn't expect to hit max turbo on all cores for any kind of sustained workload. Your clock speed will be lower than that. How much lower remains to be seen.

AMD can make the base clock almost literally anything they want, could legitimately be 1Ghz if they wanted it to be. Base clock do affect idle power usage however, there are set power states for the CPU and the lower base clock means you are shifting those power state clocks down and therefore will be using less power in lower power states. This is exactly how laptop CPUs are handled, just to a much greater extent.

 

As for clocks for Zen/Ryzen this is all directly down to the power efficiency of the core and the PPT value. Zen 3 has increase the power efficiency and is using the same 7nm TSMC process node and have stated they are using the same TDP (so unless the PPT ratio from TDP has changed) it is near impossible that Zen 3 will not clock higher under the same workloads as Zen 2. This would be directly counter to the claimed power efficiency gains AMD has made and trivial to prove.

[Blademaster91]

59 minutes ago, Bombastinator said:

I don’t think people would have as big an issue if a precedent hadn’t been set by the zen2 stuff.  “X” meant “comes with a cooler that actually works so you don’t have to buy one”. This time “x” means “absolutely nothing.  It’s complete horseshit”. The only cpu that comes with a cooler at all is the 5600x  This sort of makes sense because it’s the only 65w chip, but what it effectively means is the earlier chips at least could run without having to drop another $30-100 on a cooler.  You could still do that, but you didn’t HAVE to.  That was actually part of what made zen2 so good at price/performance.  That seems to be gone.   There IS NO 5800x or 5900x, because Nvidia doesn’t have a cooler to tame them.  They’re 5800 and 5900 with some marketing squished on the end like a wad of discarded gum in the shape of an x.

Even though I wouldn't use it I'm disappointed only the 5600X comes with a stock cooler, at least including a cooler with the 5800X would have made it a bit better value. I'd like to know if the X sku is worth it though, I'll wait for benchmarks, kinda hoping AMD has a 5700 or 5700X slightly binned down from the 5800X, 5700X would be confusing though lol.

1 hour ago, LAwLz said:

Are you referring to me?

The MSRP of the Ryzen 3600 has always been 199. It's not a sales price. It has always been that price. In fact, for most of the time on the market has been lower than 199. There were several months where you could get it for 170 dollars.

Not just you, I see lots of comparing the 3600 to the 5xxx cpu's. The 3600X price dropped quickly after launch, down to $200, boost clocks are hardly any difference, but it comes with a better stock cooler than the 3600, so value is really subjective if you want to spend an extra few dollars on the X-sku 3xxx.

[Bombastinator]

I agree.  They’d need something harder core than a prism though.  I don’t know what the prism is “rated” at but it’s got to be under 100w and practically it’s in the low 80’s. I saw a double piped prism at one point.  Might have been “bandit castle” might have been a photoshop special.  Might have been real, but not be a big enough improvement.  Need another 30% over a prism at least I suspect.

[mrdoubtfull]

1 hour ago, LAwLz said:

 

• Then they switched to discussing overall budget of the PC because then you can hide the price of the processor. All of a sudden they have a clever way of reducing the % of the cost for the CPU, but they don't have to change the performance gain number! 

I was just trying to show another way to look at it..

 

If we get a decent performance boost with the entire system and using most applications/gaming, then I don't think you can ignore the total system cost for those building new. 

 

I think we need to wait for benchmarks before making ANY assumptions, honestly.. I think the pricing is a little high, yes.. But I don't think it's the end of the world. I'm sure they'll offer more skus later (if not sooner).. 

 

I guess we just have to wait and see how this is going to play out instead of arguing about it.. Because we see the good that's doing lol. 

[LAwLz]

9 hours ago, leadeater said:

AMD can make the base clock almost literally anything they want, could legitimately be 1Ghz if they wanted it to be. Base clock do affect idle power usage however, there are set power states for the CPU and the lower base clock means you are shifting those power state clocks down and therefore will be using less power in lower power states. This is exactly how laptop CPUs are handled, just to a much greater extent.

 

As for clocks for Zen/Ryzen this is all directly down to the power efficiency of the core and the PPT value. Zen 3 has increase the power efficiency and is using the same 7nm TSMC process node and have stated they are using the same TDP (so unless the PPT ratio from TDP has changed) it is near impossible that Zen 3 will not clock higher under the same workloads as Zen 2. This would be directly counter to the claimed power efficiency gains AMD has made and trivial to prove.

A 100Mhz lower base clock and a 200Mhz higher boost clock means to me that they are more or less the same frequency than the Ryzen 3000 series. That was my point.

Yes they could make the base clock whatever they want, but clearly they don't. If it was for power savings reasons they wouldn't just lower it 100Mhz. They would lower it far more, because Ryzen chips already go into far more aggressive power saving states where they go below 1GHz. The base clock isn't for power saving reasons. It's the lower guaranteed frequency you get out of the chips. If they lower the guaranteed speed then that to me indicates that they might not be able to keep as high frequency with all cores running.

They didn't lower it for fun.

 

9 hours ago, leadeater said:

As for clocks for Zen/Ryzen this is all directly down to the power efficiency of the core and the PPT value. Zen 3 has increase the power efficiency and is using the same 7nm TSMC process node and have stated they are using the same TDP (so unless the PPT ratio from TDP has changed) it is near impossible that Zen 3 will not clock higher under the same workloads as Zen 2. This would be directly counter to the claimed power efficiency gains AMD has made and trivial to prove.

I think that remains to be seen. In any case, I don't think Zen3 will clock that much higher than Zen2 so I think bringing that up as some reason why the Ryzen 5000 isn't worse value than the 3000 is silly.

[LAwLz]

7 hours ago, mrdoubtfull said:

If we get a decent performance boost with the entire system and using most applications/gaming, then I don't think you can ignore the total system cost for those building new. 

That's the thing though, 20% higher CPU performance does not translate to 20% higher performance for "the entire system". That's why you can't say it's ~10% more money for 20% higher performance.

It might be 10% more money for 5% higher performance in the tasks most people do, and it might be 5% that nobody would even notice because those tasks are already executed fast enough.

 

And like I said, even in your example I still think buying the 12 core Ryzen 3000 is the better deal. You get more performance for less money.

[porina]

10 hours ago, leadeater said:

AMD can make the base clock almost literally anything they want, could legitimately be 1Ghz if they wanted it to be. Base clock do affect idle power usage however, there are set power states for the CPU and the lower base clock means you are shifting those power state clocks down and therefore will be using less power in lower power states. This is exactly how laptop CPUs are handled, just to a much greater extent.

I don't know the exact wording AMD use to define base clock, but if we take the Intel definition, it is the minimum guaranteed LOAD clock when the CPU is fitted with a cooler rated equal to the TDP. Idle is unrelated to base clock. I think AMD made the point that Zen 2 cores can turn off when not in use. So zero clock on idle regardless what the base clock is.

 

10 hours ago, leadeater said:

As for clocks for Zen/Ryzen this is all directly down to the power efficiency of the core and the PPT value. Zen 3 has increase the power efficiency and is using the same 7nm TSMC process node and have stated they are using the same TDP (so unless the PPT ratio from TDP has changed) it is near impossible that Zen 3 will not clock higher under the same workloads as Zen 2. This would be directly counter to the claimed power efficiency gains AMD has made and trivial to prove.

Agree it would be easy for someone with such a CPU in their hands to test. My gut feeling is there could be a slight reduction in heavy load clocks, but more than made up for in the increased IPC and potentially higher boost in some cases.

 

That reminds me of a comment someone made about their Cinebench score reveal, I'll get back on that once I grab a coffee as we might be able to tease out some info from that.

[LAwLz]

29 minutes ago, porina said:

Agree it would be easy for someone with such a CPU in their hands to test. My gut feeling is there could be a slight reduction in heavy load clocks, but more than made up for in the increased IPC and potentially higher boost in some cases.

This is what I am thinking too. It would explain why they lowered the base clock as well (which again, doesn't have anything to do with power savings).

If Ryzen 5000 always runs at a higher frequency than Ryzen 3000 then I don't get why AMD would lower their base clock. Sure you might say base clock is a meaningless number but the fact remains that AMD made the decision to lower it for some reason. They didn't do it just for fun or because they felt like it.

[porina]

12 minutes ago, porina said:

That reminds me of a comment someone made about their Cinebench score reveal, I'll get back on that once I grab a coffee as we might be able to tease out some info from that.

On the AMD launch stream just before 17 minutes, they show the single thread CB20 score of 631 for the 5900X. Let's assume it is running at the max single core turbo of 4.8 GHz. The multicore score ratio is just about viewable, and I think it says 12.91x (someone with a 4k display check it please?). This suggests an all core score 8146. They blurred it out the in the display so we can't see the actual number.

 

Next part I have to make an educated guess. How much benefit does CB20 get with SMT? We don't know, and I don't even have a number for Zen 2. From my previous testing I have a value of 30% for Zen+, and 33% for Coffee Lake. Their Comet Lake comparison CPU scored 544 at 5.3 GHz assumed single core turbo. Because we know architecturally Comet Lake is the same back through to Skylake, that gives a check point. My Coffee Lake CPU gave a score of 103.2 per core, per GHz single thread. AMD's Intel CPU score works out 102.6 per core per GHz. Close enough. Do the same with Zen 3, we get 131.5 per core per GHz. Using AMD's numbers that's 28% higher single thread IPC in CB20 than Skylake family, and using my numbers 32% faster than Zen+.

 

Now the difficult part. We know they improved both single thread and multithread performance. We just don't know the ratio between them for CB20, which would be required to estimate the running clock for the multithread run. Actually, forget that. The calculated all core score sounds a bit low. The ratio at close to 12 suggests that the IPC increase from SMT is largely offset by lower all core clock. It is more likely that they didn't have a serious run of multithread so the ratio is not showing best case. For comparison, on a quick trip to hwbot, it looks like 3900X submissions with a score in that ball park were around 4.5 GHz. The clocks and IPC don't add up with that score.

 

I think I'll abort here. There's not enough information and I'm just digging a hole at this point.

[LAwLz]

3 hours ago, porina said:

On the AMD launch stream just before 17 minutes, they show the single thread CB20 score of 631 for the 5900X. Let's assume it is running at the max single core turbo of 4.8 GHz. The multicore score ratio is just about viewable, and I think it says 12.91x (someone with a 4k display check it please?). This suggests an all core score 8146. They blurred it out the in the display so we can't see the actual number.

 

-snip-

Speaking of which, just saw an article from Guru3D that someone has uploaded a bunch of Ryzen 5000 benchmarks to Cinebench.

I don't know if they are legitimate, and it's always best to wait for official benchmarks, but these seems pretty legitimate.

 

https://www.guru3d.com/news-story/zen-3-lineup-cinebench-r20-scores-pop-up-in-database.html

 

 

 

 

 

These scores would make the 5600X 20.5% faster than the 3600X in single core, and 15% faster in multicore.

That lines up very well with what I am expecting.

 

43% higher price for less than 15-20% higher performance... Not a good deal.

 

 

 

We can also compare the 5800X to the 3900X since they are roughly the same price (with the 3900X being slightly cheaper).

The 3900X has 17% slower single core performance but 25% higher multithreaded performance. That seems like a good tradeoff to me although not as clear cut as the 3600(X) vs 5600X.

 

The new CPUs gets really good scores (judging by these leaked benchmarks that we shouldn't trust 100%), but their pricing is quite frankly terrible.

[Stahlmann]

24 minutes ago, LAwLz said:

The new CPUs gets really good scores (judging by these leaked benchmarks that we shouldn't trust 100%), but their pricing is quite frankly terrible.

I guess AMD really thinks when they have the performance crown, they can increase their prices over the competition aswell. Problem is, i think they forgot that they still fight an uphill battle against Intel as a brand. And most people who don't get much into this still think "Intel is better than AMD". I think they make a huge mistake with their new pricing. For example at 300$ many people will get a 10600K over a 5600X now and AMD is losing out on sales again. Basically they strenghten the perceived value of the Intel CPUs IMO.

[AlwaysFSX]

12 minutes ago, Stahlmann said:

I guess AMD really thinks when they have the performance crown, they can increase their prices over the competition aswell. Problem is, i think they forgot that they still fight an uphill battle against Intel as a brand. And most people who don't get much into this still think "Intel is better than AMD". I think they make a huge mistake with their new pricing. For example at 300$ many people will get a 10600K over a 5600X now and AMD is losing out on sales again. Basically they strenghten the perceived value of the Intel CPUs IMO.

Unless my math is wrong, the 10600k is worse for the money.

 

5600X $300 (+9%)
604 | 4,312
+20% | +19%
10600k $275
503 | 3,605

[leadeater]

5 hours ago, porina said:

I don't know the exact wording AMD use to define base clock, but if we take the Intel definition, it is the minimum guaranteed LOAD clock when the CPU is fitted with a cooler rated equal to the TDP. Idle is unrelated to base clock. I think AMD made the point that Zen 2 cores can turn off when not in use. So zero clock on idle regardless what the base clock is.

It's not quite unrelated, also idle I was more meaning conditions not under load that would be bringing the processor to P0 state or boost state, P0 being maximum Power and Frequency. What's not clear to me however is if P0 is base clock or boost state as both Intel and AMD have different clocks at maximum load depending on number of active cores. Fortunately for this it doesn't really matter.

 

The thing with Intel TDP is it's not bound to the clocks as much as you'd think, the listed TDP is for the cooler design yes and is a guarantee that the base clocks can be achieved with a cooler of the specified capability however the specified TDP can, and has been for many CPUs, be above that which is actually required. Intel has for a long time gone with a set TDP across a product range/set of products based off the highest one in that category so lesser CPUs SKUs below it get given the same TDP spec but are in fact lower in power usage so don't actually need a cooler of that spec. This is where it is important that TDP is not a direct indicator of power draw.

 

Having a lower base clock on the same node using the same architecture will result in lower power usage at the same power state (this is an ACPI spec thing) and all the lower power states, Pn with higher numbers, have historically been scaled off base clock. So if you lower the base clock you have lowered the clocks at each power state so you will be using less power at idle.

 

All things being equal a lower base clock will result in lower power usage, actually how much I don't know. Knowing what power state a CPU is at at any point in time is hard and bringing a CPU to a specific power state is also hard so actually going out and trying to measure it I wish the best of luck to anyone that tries.

 

If we start talking specifically about AMD then it's a lot more complicated as AMD uses Precision Boost 2 (for Zen 2) which is WAY more granular than ACPI P-States/Intel Power States. The same theory still applies however, lowering the base clock lowers the power usage at all equivalent power states below it compared to an equivalent CPU featuring a higher base clock because as far as I know Precision Boost 2 only applies to above base clock (as the name suggests) and it is ACPI P-States at and below base clock.

 

The only reason I can see AMD lowering the base clock for Zen 3 is for power management and no other reason.

 

Edit:

Also another reason I just thought of to lower the base clock is so there is a larger frequency range under Precision Boost 2(3?) control which is granular to 25MHz increments, unless Zen 3 features better than this, so you would be getting better frequency and thermal controls under boost by lowering the base clocks. So there is a second reason to lower base clocks.

 

Situation where I can see this being necessary is when a single core, possibly two, are boosting to highest frequency possible which uses a lot of power eating in to PPT and the rest of the cores need to clock lower but must be above base clock or all boosting stops. So I expect this to be more related to uneven core loads than even or well multi-threaded workloads. Power required for a given frequency is very steep on the upper end, on Zen 2 and I don't see that changing for Zen 3.

 

So I've changed my own mind now, I suspect the lower base clocks are to support the higher one/two core boost clocks that are above Zen 2 maximum clocks.

[leadeater]

4 hours ago, porina said:

My gut feeling is there could be a slight reduction in heavy load clocks, but more than made up for in the increased IPC and potentially higher boost in some cases.

That would almost go against everything we know about Precision Boost 2(3?) and that Zen 3 has been shown to clock higher already, at least on a single core however re: Precision Boost 2(3?). There is the possibility of it being true since the specific wording was increased performance per watt which is a factor of both frequency and IPC.

 

So I guess my question would have to be has the increase in IPC caused an increase in power at the same clock?

 

It's rather late so either I'm being stupid or is this 20% performance per watt gain over Zen 2 being claimed above? And with a claimed average IPC gain of 19% I can only see two realistic possibilities with all the information we currently know, exactly the same clock under the same multi-thread workloads or higher than Zen 2.

 

[spartaman64]

1 hour ago, Stahlmann said:

I guess AMD really thinks when they have the performance crown, they can increase their prices over the competition aswell. Problem is, i think they forgot that they still fight an uphill battle against Intel as a brand. And most people who don't get much into this still think "Intel is better than AMD". I think they make a huge mistake with their new pricing. For example at 300$ many people will get a 10600K over a 5600X now and AMD is losing out on sales again. Basically they strenghten the perceived value of the Intel CPUs IMO.

thats not how perceived value works. if they really want to increase their perceived value then they would price their cpus higher than the intel cpus. why do you think stuff like gucci have a high perceived value because they are good quality? because they are cheap? no because they are fucking expensive 

[LAwLz]

1 hour ago, AlwaysFSX said:

Unless my math is wrong, the 10600k is worse for the money.

 

5600X $300 (+9%)
604 | 4,312
+20% | +19%
10600k $275
503 | 3,605

Yep, even the 5600X offers better price to performance than Intel's offerings, despite being bad value in my eyes.

At this point AMD is basically only competing with itself, and if I were AMD and my only interest was to make as much money as possible, I'd be kicking myself for releasing a product that was too good last gen (the 3000 series).

I wouldn't be surprised if AMD will stop manufacturing the 3000 series to get it off the market as quickly as possible, so they don't risk cannibalizing sales of the (probably) higher margin 5000 series.

[spartaman64]

1 hour ago, LAwLz said:

Speaking of which, just saw an article from Guru3D that someone has uploaded a bunch of Ryzen 5000 benchmarks to Cinebench.

I don't know if they are legitimate, and it's always best to wait for official benchmarks, but these seems pretty legitimate.

 

https://www.guru3d.com/news-story/zen-3-lineup-cinebench-r20-scores-pop-up-in-database.html

 

These scores would make the 5600X 20.5% faster than the 3600X in single core, and 15% faster in multicore.

That lines up very well with what I am expecting.

 

43% higher price for less than 15-20% higher performance... Not a good deal.

 

We can also compare the 5800X to the 3900X since they are roughly the same price (with the 3900X being slightly cheaper).

The 3900X has 17% slower single core performance but 25% higher multithreaded performance. That seems like a good tradeoff to me although not as clear cut as the 3600(X) vs 5600X.

 

The new CPUs gets really good scores (judging by these leaked benchmarks that we shouldn't trust 100%), but their pricing is quite frankly terrible.

thats not how you should compare performance to price you need to take the cost of the entire system. can you place a 3600 on a table and say to it ok run cinebench? 

 

if you say its not worth to upgrade to the 5000 series if you have a 3000 series cpu then id agree with you but when is it ever worth it to upgrade your cpu in one generation? heck i dont even think its worth it to upgrade your gpu every generation even though gpus have a much higher increase in performance

[LAwLz]

15 minutes ago, leadeater said:

That would almost go against everything we know about Precision Boost 2(3?) and that Zen 3 has been shown to clock higher already, at least on a single core however re: Precision Boost 2(3?). There is the possibility of it being true since the specific wording was increased performance per watt which is a factor of both frequency and IPC.

 

So I guess my question would have to be has the increase in IPC caused an increase in power at the same clock?

 

It's rather late so either I'm being stupid or is this 20% performance per watt gain over Zen 2 being claimed above? And with a claimed average IPC gain of 19% I can only see two realistic possibilities with all the information we currently know, exactly the same clock under the same multi-thread workloads or higher than Zen 2.

I think that's hard to tell without having chips in hand and testing them.

Something work remembering is that those are first party benchmarks and those are almost always cherry picked to show as big of an improvement as possible.

 

It's also worth remembering that we have no idea how those numbers were measured. I see a little asterisk there but can't make out what it says. Performance per watt is not a static number and it changes depending on which test and architecture you run. One architecture might have terrible performance per watt compared to another in some specific tests, but in others the reverse might be true. We saw this very clearly when everyone and their dog was buying GPUs to mine on. Despite Nvidia offering much better performance per watt in games, AMD pulled ahead in performance per watt for mining.

 

We don't know what performance level they were targeting either for those tests. Performance per watt usually goes down the more you push a chip. The 3900XT might be extremely efficient when it runs at 3GHz, but to get it to run at 4GHz (33% higher performance) might require 50% more power, making it far less efficient. The 3900XT might simply be further away from the performance per watt sweet spot than the unnamed Zen 3 processor was in that demo. Since they don't mention (from what I can tell) what the performance was like in the tests, their "behind the scenes" numbers might look like this:

Zen2 at 100 watts = 100 FPS

Zen3 at 30 watts = 36 FPS

Therefore, Zen3 gets 20% higher performance per watts.

But if we scaled Zen3 up to 100 watts then it might only get 105 FPS, making the performance per watt difference a mere 5% increase.

 

I am not saying that's what AMD did, but what I am saying is that without any knowledge about how AMD arrived at those numbers, they are pretty much meaningless. AMD could essentially put in any number they want in those graphs and they could tweak the three different test benches to get the result they desire. It's just that their lab results won't really match any real world use case.