The Zen 5 can run at 6GHz, but be prepared to take a beating. 14 thousand for a 170W octa-core?

We had a report over the weekend that AMD processors with the Zen 5 architecture should achieve up to 40% better performance per core against Zen 4 (but not necessarily in a single-threaded application, as I discussed). There’s a lot more going around the internet now about Zen 5, so let’s take a look at some more rumors. We have another leak about the performance, but unfortunately also more negative information about the possible burning of prices and an increase in consumption – that is almost certain.

To begin with, there is information that is not of the “someone claims” type, but there is direct evidence for it. Indeed, the first traces of Granite Ridge processors have been found in public shipping logs, which is a desktop Ryzen 9000(X) for socket AM5 using the same chiplet structure and the same IO chiplet as Ryzeny 7000(X) “Raphael” (meaning it will have the same memory controller and connectivity as well as the iGPU, only the CPU cores in the CCD chiplets will change).

Granite Ridge ES samples captured

In shipping documents, two ES Ryzen samples have been labeled “Granite Ridge”, representing a six-core model and an eight-core model. Hexa-core will probably be called Ryzen 5 9600Xthe EC sample has a code 100–000001290–21 and it is revision A0 (that is, the first). An important note is that TDP is written for it in the documents 105 W (and thus the maximum consumption, the so-called PPT, would be 142 W). It is possible that a 65W model will be produced later, but there is no information about it yet.

The second model is probably the ES pattern foreshadowing Ryzen 7 9800X (according to leaker Kepler_L2 AMD will use this designation, not 9700X). The sample is labeled 100–000001404–21 and in this case it is the more mature silicon revision B0. It is an octa-core and be careful – it has a TDP 170 W (which means maximum consumption / PPT 230 W). This is a downgrade from the current generation, where the octa-core 7700X still has a 105W TDP.

It shows that Zen 5’s core consumption will also go up along with the performance increase. This 9800X model will probably not cool well enough if it consists of only one CCD chiplet, in which the resulting heat will be concentrated. It’s quite possible that in practice the consumption will be significantly lower than the 230W, because the boosting stops by reaching high temperatures before the CPU hits the consumption limits.

AMD naming the octa-core Ryzen 7 9800X, as claimed by Kepler_L2, could correspond to this increase in consumption (and hopefully performance). With this name, the higher consumption would probably be somehow more defensible than if the model labeled 9700X directly claimed the role of the successor of the 105W model 7700X.

On the other hand, the TDP of the highest Ryzen 9 9900X and 9950X models will probably not increase. They will still have a 170W TDP (230W PPT). Socket AM5 does not count on a higher CPU class. But it is possible that processors will use this allocation more consistently than today.

Ryzen 9000 significantly more expensive?

Perhaps the increase in the second digit of the model number is supposed to “cover” the increase in the price of the eight-core model at the same time. Kepler_L2 and other vague rumors are currently claiming that Zen 5 architecture processors (ie Ryzen 9000 models) will generally be more expensive than previous generations. While the Ryzen 7 7700X went on sale for $399, the Ryzen 7 9800X model could reportedly go for $449, or even $499 (if not more?). And again, the octal designation is perhaps also chosen so that the processor is nominally in a higher class and does not invite so much comparison with the cheaper Ryzen 7 7700X (although such a parallel is inevitable based on the number of cores).

This would probably also affect the six-core, of course also the higher models, it is speculated that the highest sixteen-core Ryzen 9 9950X model could be for $999. There is no concrete evidence of this yet, and it is possible that the leakers themselves are speculating. We’d love to be wrong about that message.

According to the leakers, this price increase is planned for a simple reason – Intel will not have a competitive answer. At the same time, Intel manages to hold its position on the OEM market through various inertias and hiccups, even though it has worse processors, so even in the case of a significant dominance of Zen 5, AMD will probably not significantly improve its market share (since the retail market that responds to this is only a small parts of the market). Therefore, AMD will logically want to improve at least in terms of revenue share by increasing prices (and margins). Of course, this has negative effects on us buyers.

New performance report

Alongside this speculation(?) about pricing, a new performance leak has surfaced for Zen 5. According to a Chinese leaker with the nickname Xinoassassin1, AMD now states in its non-public materials that Zen 5 will bring a single-threaded performance increase of around +20%if measured in the test Cinebench R23 1T (that is, this time really in a purely single-thread measurement). This is probably a combination of frequency and IPC effects.

So it is again a lower number than it should be in a multi-threaded load, but be careful. Already with Zen 4, Cinebench R23 (as well as others) was one of the tasks where a smaller increase in IPC was observed than in some others. While on average AMD reports an IPC increase of 13% for Zen 4 against Zen 3, only +9% in Cinebench R23.

IPC increase between Zen 3 and Zen 4 in various programs and games (official presentation, 2022)

Author: AMD

If you were to take it as a direct proportionality, the average IPC increase for Zen 5 could be +28%. But take this purely as an illustration, because Zen 5 is a different architecture and Cinebench may react to its changes completely differently than Zen 4. And of course, we can’t guarantee that the rumor from Xinoassassin1 is true.

Frequency increase, 6 GHz target broken?

There was also information among the gossips that the final production Ryzen 9000s may have slightly better frequencies than the Ryzen 7000s, so Zen 5 may not bring a regression in frequencies as previously feared. There has been a rumor that the top model could go over 6GHz, but it will probably be in a similar vein to how the Ryzen 9 5950X is able to (sometimes) go over 5GHz. So it will not be an officially stated frequency, but a “shadow turbo” (or “XFR”, as AMD stated with the first generations of Ryzens), which processors sometimes allow to be used in single-threaded applications with very good cooling or a low starting temperature, but you don’t have it guaranteed.

For example, on the Ryzen 9 5950X this unofficial band makes up to 150 MHz in addition to the official boost of 4.90 GHz (so the absolute ceiling without overclocking is 5050 MHz), on the Ryzen 9 7950X the official boost is 5.70 GHz, but the “shadow band” is again up to +150 MHz (5850 MHz). However, it seems to be used less often with the Ryzen 9 7950X.

According to this rumor, the future Ryzen 9 9950X could have an official maximum boost of either 5.80 or 5.90 GHz, but the extra shadow turbo could theoretically whip it up to 6050 MHz, but again this could be a relatively rare phenomenon.

Of course, this all depends on whether the rumor is true, so don’t take it as a guaranteed thing that is already known for sure. We are listing it more as a point of interest. However, if this is true, then the 20% improvement in Cinebench includes this factor and the IPC improvement would be slightly less than 20%. The resulting CPU performance is of course co-shaped by both of these components, it is the product of IPC and frequency.

We will know more in two months

It looks like AMD might do some Zen 5 reveal or sub-reveal at Computex (late May/late June). The question is how far in advance this sub-disclosure will be before the final release itself, when it will be possible to buy these processors. For example, Zen 2 came out shortly after (early July), Zen 4 until late September.

But whether the release will be more at the beginning of the summer, or at the end of the summer (or even in the fall), Computex will probably make it clearer what to expect from Zen 5. Depending on how far away the actual release is, this revelation will probably be more or less detailed.

Resources: https://twitter.com/x/status/1774368837137576039AnandTech Forum (1, 2, 3, 4, 5), https://twitter.com/Kepler_L2/status/1773441385925865914, https://twitter.com/momomo_us/status/1773392771149402125