Windows 2003 memory capacity

Prior research shows that small correlations of residuals—indicating a decent fit of the model—might come along with indications of bad fit in terms of conventional criteria when the unique variances are small Browne et al. Standardized loadings on the general factors were high all above 0.

The model allows for correlations between corresponding factors for both paradigms. All other correlations were theoretically not expected and were fixed to zero. Models in which these fixed correlations between latent factors were estimated freely resulted in small estimates not surpassing the conventional significance criterion.

Due to the lack of latent variable correlations for the two conflict factors—which are of focal interest in their relation with WMC—and in order to keep model complexity relatively low, the relationships of cognitive-conflict factors with WMC and SM were studied in two separate structural equation models SEM , one investigating the correlation of conflict in the Simon task with WMC and SM, and the other doing the same for conflict in the Eriksen task.

Factor loadings did not notably differ from those estimated in the measurement models Model 3 and Model 4a. Model 4c had the same structure as Model 4b, except that the sub-structure of Model 4a describing individual differences in the Eriksen task was related to Model 3.

This model had the same constraints on correlations as Model 4b. Factor loadings did not notably differ from those estimated in the antecedent measurement models Model 3 and Model 4a. These positive correlations express slightly faster responses after stimulus repetition for higher levels of WM and SM the dependent variables are inverted latencies for Eriksen and Simon in all models, thus higher scores represent better performance.

The relation between WMC and fluid intelligence received considerable attention Ackerman et al. Unsworth et al. Because our indicators of WMC—with the exception of complex span—reflect primarily the capacity of working memory or PM, we understand Unsworth's two-component model as implying that SM should increment the prediction of Gf over and above WMC. The executive attention view and the binding theory both predict that WMC will show unique contributions to the explanation of Gf and both views are agnostic toward a potential increment in the prediction from SM.

We extended Model 3 with a further latent factor Gf, measured with three reasoning tests. Standardized factor loadings of Gf indicators were 0. Therefore, the small numerical difference between the freely estimated correlations is—given the statistical power in the present study—inferentially not meaningful.

To test a specific prediction derived from the binding hypothesis Oberauer et al. Shared method-specific variance was modeled in a second nested factor that we label PAsso in line with interpretation from prior models that was also orthogonal to the other factors in Model 6a. A first structural model estimating explained variance in Gf by means of binding and secondary memory indicators Model 6a. In order to achieve convergence in this model the residual variance of the numerical SM indicator was fixed to zero.

A second structural model estimating explained variance in Gf by means of binding and secondary memory indicators Model 6b. Our finding speak to three interrelated questions: What is working memory capacity? How should it best be measured? We discuss the implications of our findings for these three questions in turn.

We began by testing to what extent different working-memory task classes reflect the same WMC construct. The hypothesis that complex-span performance reflects to a large extent the ability to retrieve efficiently from SM Unsworth and Engle, a , b motivated the prediction that a Complex Span factor is more indicative of SM than the other two paradigm-specific working-memory factors, Recall-N-Back, and Updating, because the latter two paradigms arguably minimize the potential contribution of SM.

What we found in Model 1 were strong correlations of performance at the level of latent variables implying a strong overlap and congruence of the constructs measured by the four task classes. Therefore, it is unlikely that the task classes capture SM to very different degrees. Importantly, the present data support the position that Recallback tasks as a specific version of so-called n-back are valid measures of WMC.

This result reinforces prior reports Shelton et al. Our findings go beyond prior research by showing very high construct overlap of Recall-N-Back and Complex Span and a nearly perfect relationship with the factor for Updating tasks. Importantly, this conclusion—as the ones to be discussed next—is based on latent variables and not on the analysis of single tasks.

The strong construct overlap between Complex Span and Updating is of particular interest because working-memory updating has been regarded as one of the three factors of executive functions identified by Miyake et al.

The updating factor was highly correlated 0. At first glance this close relationship could be interpreted as supporting the executive-attention theory of WMC. We believe that this conclusion would be premature.

As pointed out by Ecker et al. When Ecker and colleagues separated these components, they found that only the general working-memory components were related to measures of WMC. Therefore, we interpret the close correlation between the Updating factor and the other WMC factors in our study as reflecting variance in those general working-memory components of updating tasks.

Model 2 served to test the binding hypothesis of WMC Oberauer et al. In this model the general WMC factor was perfectly correlated with the Binding factor, so that the loading of Binding on WMC could be fixed to 1 without loss of fit.

These findings suggest that the common source of variance across all four task paradigms is the cognitive mechanism of building, maintaining and updating arbitrary bindings Oberauer, ; Oberauer et al. Rapid formation and updating of bindings is needed for the Updating and Recallback tasks because accurate performance in these paradigms requires memory for the relations between items and their contexts e. Bindings are also a core factor for success in Complex Span tasks, because these tasks require the recall of items in the correct serial order Schmiedek et al.

For the recall of serial positions in a list it is necessary to create firm bindings between content words, letters, length and directions of arrows and context the serial position of a word, letter, or arrow within the list.

Additionally, these bindings need to be established and maintained in the presence of an interfering secondary task. One corollary of the strong correlation between different working-memory paradigms is that all four task classes can be seen as good proxies of a general WMC factor.

A closer look at the psychometric quality and attributes of competing paradigms allows for a more refined perspective. First, concerning the magnitude of the loadings of the task-class specific factors on the second order WMC factor, the Binding and Updating factors did best, showing no task-specific variance at all, and Complex Span did comparatively less well, showing the largest amount of task-class specific variance. These differences notwithstanding, all four task classes are good indicators of WMC, because they reflect to a large extent reliable variance of the general WMC construct.

The relatively novel Recall-N-back paradigm is arguably a very efficient method for assessing WMC because it enables continuous recording of performance at a high rate. It is possible that the relevance of content variance is larger for conventional short term memory span measures, such as digit span or letter span, than for working-memory tasks Kane et al. Third, the binding tasks had shared variance with the SM task, which could be taken to compromise the clean separation of measurement of WMC and SM.

The covariance between our binding tasks and the paired-associates learning tasks that we used as indicators of SM probably reflect the shared method variance between these two task paradigms. The largely analogous methods for these two paradigms was intended to enable a direct comparison of the ability to maintain temporary bindings in working memory and the ability to acquire more long-term associations in SM as in our Models 6a and 6b.

SM could be measured in a more method-independent way through multiple indicators using different methods. Every cognitive test carries some task-specific variance that is unrelated to the construct of interest, and tests of WMC are no exception. Therefore, we generally recommend measuring WMC through a heterogeneous set of paradigms to avoid mono-operation bias Shadish et al. Often, however, only limited time and resources are available to measure WMC, such that administration of only a single test is feasible.

In light of our findings of high correlations between four different WMC paradigms, this is a defensible practice. In that case, a number of considerations can be made to choose among the available paradigms. First, the four task classes investigated here differ in their construct validity, as reflected in their loadings on the general WMC factor.

Although the differences were not large, they might weigh slightly in favor of using the Binding or the Updating task rather than Cspan or Recall-N-Back. Second, the tasks differ in their efficiency, that is, the number of independent measurements per testing time. A third consideration concerns the exhaustiveness and sufficiency of task scores.

In all complex-span tasks participants work on two tasks, a memory task and a concurrent processing task, but only their memory performance is considered in scoring. Although cut-off scores in processing-task performance are usually applied Conway et al.

This issue might be a nuisance in scientific use but is a serious problem in using Complex Span measures as a diagnostic tool in high-stakes contexts. There are also multiple ways of calculating the recall scores see the procedures discussed in Conway et al. In contrast, there is less ambiguity of scoring for the other three task classes. The results from Model 3 are consistent with previously reported correlations between WMC and SM, showing that they are separable but closely correlated constructs Unsworth et al.

The correlation reported here is stronger than that reported in previous studies. This might be the case because our WMC factor reflects the common variance of four different working-memory paradigms, whereas it was restricted to Complex Span indicators for working memory in previous work. At first glance this finding is surprising because, whereas the complex-span paradigm bears close similarity to established SM tasks such as the continuous-distractor task, the latter three paradigms were designed to minimize the potential contribution of SM: The Updating, Recall-N-Back, and Binding tasks used comparatively short retention intervals, thereby leaving little chance for encoding into SM, and they generated a high level of proactive interference, thereby minimizing the usefulness of SM representations for a similar argument regarding proactive interference see Cowan et al.

Therefore, it appears implausible that variance in the efficient use of SM plays a major role in determining performance in those working-memory paradigms. Our finding becomes less surprising when we consider the reverse direction of causality: According to the binding hypothesis, high WMC reflects the ability to establish robust bindings in working memory, which in turn support encoding of those bindings into SM.

At the same time, there is a substantial proportion of unique variance in both SM and the second order WMC factor. This conclusion is reinforced by the finding that Binding provides an independent contribution besides SM to predicting fluid intelligence Models 6a and 6b.

One possible interpretation of this finding is that fluid intelligence reflects on the one hand the ability to maintain and update temporary bindings in working memory Oberauer at al. We tested the relative contribution of bindings in working memory and association-learning in SM to predicting fluid intelligence in two models focusing on Binding and SM as predictors of Gf.

Despite the strong collinearity between SM and the Binding factor it seems as if the Binding factor was slightly more important for the prediction of fluid intelligence: Once Binding performance was statistically controlled for, SM was no longer significantly related to fluid intelligence.

Although the present results require replication we conclude that the results for Model 3 and the models derived from it are in line with the binding hypothesis of WMC Oberauer et al. In Models 4a through 4c we tested the relation between the efficiency of cognitive control and the other factors.

The two factors representing the efficiency of cognitive control—conflict costs in the Simon task and in the Eriksen flanker task—were unrelated with each other. This is a replication of previous findings showing that conflict related slow-down factors don't generalize across these two paradigms Keye et al.

If both paradigms capture similar conflict costs they should share a substantial amount of variance, and hence should be correlated at least moderately positive with each other, contrary to our finding. We conclude that either at least one of the two conflict cost factors is not a valid measure of cognitive control in response-conflict situations, or that individual differences in cognitive control in response-conflict tasks are entirely task specific.

From the perspective of the theory of Executive Attention this results is discouraging because it suggests that measures taken to reflect the ability to cope with interference and distraction essentially capture task-specific variance in performance.

We also observed no correlation between the conflict-related slow down factors in either the Simon or the Eriksen task and any of the WM and SM factors. These findings replicate and extend previous reports on the correlation between WMC and factors of conflict costs after removing variance due to individual differences in overall speed Keye et al.

Other research on this issue is somewhat inconclusive: Redick and Engle and Heitz and Engle found that high WM span participants were faster at minimizing the distracting impact of incompatible flankers as compared to low WM span participants—a result that was not found for the compatible trials Heitz and Engle, These results arose from extreme-group comparisons, which are regarded as problematic in individual-differences research see Preacher et al.

Other studies that did not rely on extreme-group comparison reported no correlation between WMC and performance on the flanker task Friedman and Miyake, The present methodological approach relies on latent variable modeling of factors reflecting cognitive control, partialling out baseline reaction time along with other potentially confounded response components such as repetition priming.

In addition, we used a broad battery of working memory measures to measure WMC on a high level of generality. We found that any relation of both the Simon and the Eriksen paradigm with WMC is entirely due to individual differences in overall choice reaction time.

This relation is probably best understood as an instance of the correlation between WMC and general cognitive speed in choice tasks Schmiedek et al. To conclude, we obtained strong evidence for the binding hypothesis of WMC: This hypothesis correctly predicted that complex-span tasks, updating tasks, and binding tasks all shared a large proportion of variance, which reflects a broad general WMC construct and strongly predicts fluid intelligence.

We also obtained some evidence for the two-component hypothesis of Unsworth and Engle a , b. This hypothesis correctly predicted that a measure of SM is closely related to complex span and to fluid intelligence. This hypothesis did not predict, but is at least compatible with the finding that SM was equally strongly correlated with our updating and binding tests. Finally, our results are not well explained by the executive-attention theory of WMC, which erroneously predicts a close correlation between measures of inhibition one the one hand, measures of WMC and fluid intelligence on the other hand.

Correspondence and requests for reprints should be addressed to Oliver Wilhelm, Department of Psychology, University Ulm, e-mail: ed.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Miyake and Friedman recently provided an update on their model, which has better psychometric properties. Person parameters are compromised if there are individual differences in response to task order variations.

Therefore we cannot rule out effects of the specific task order. Any such effects, however, should inflate correlation estimates in the same way for all correlations, without distorting the pattern of correlations that is critical for our structural analyses. The multiple imputations of datasets were realized with the R package mice by van Buuren and Groothuis-Oudshoorn using a routine implemented by Robitzsch Estimating all models with these imputed data had only small effects on correlation and regression coefficients and had no impact on the conclusions drawn from the models.

National Center for Biotechnology Information , U. Front Psychol. Published online Jul Prepublished online May 2. Author information Article notes Copyright and License information Disclaimer. Reviewed by: Colin G. Received Mar 18; Accepted Jun This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.

This article has been cited by other articles in PMC. Abstract A latent variable study examined whether different classes of working-memory tasks measure the same general construct of working-memory capacity WMC. Keywords: working memory capacity, fluid intelligence, secondary memory, cognitive control, binding.

Open in a separate window. Figure 1. WMC as primary and secondary memory Building on traditional dual-store models, Unsworth and Engle a proposed that performance in complex-span tasks draws on two sources, a limited capacity component that maintains information over brief periods of time, and a more durable component that stores information over longer time periods.

The binding hypothesis of WMC In our own view, working memory is a system for building, maintaining and rapidly updating arbitrary bindings.

Aims and predictions for the present study The present study has three interlinked aims. Procedure Trained research assistants tested up to 9 participants simultaneously. Measures, scores and estimates of reliability The measures are conceptualized as indicators for four task classes.

Figure 2. Complex span tasks Cspan In the reading span task , adapted from Kane et al. Updating tasks updating e. Recall 1-back RNb Dobbs and Rule, Three 1-back tasks were designed to measure recall of continuously updated elements. Figure 3. Binding tasks binding The binding tests relied on pairing stimuli from different content domains.

Reasoning tasks Gf In line with the variation of content domains in the WMC measures, assessment of reasoning ability was based upon three tests varying in content. Tasks measuring response inhibition inhibition Stimuli for the Eriksen Flanker E task were five left or right pointing arrows presented in a row in the center of the screen.

Data treatment Both test sessions were attended by participants. Results We tested a series of latent variable models to address the research questions outlined above. Figure 4. Working memory capacity and secondary memory We next address the relation between the general WMC factor established in Model 2 and SM. Figure 5. Working memory capacity and controlled attention We next explored associations of WMC with measures of cognitive control, in particular response inhibition in the Eriksen flanker and the Simon paradigm.

Measurement model of response inhibition To investigate whether the two paradigms Simon and Eriksen flanker are measuring the same underlying construct of response inhibition we specified a measurement model relating latent factors of performance on the Simon task and the Eriksen Flanker task. Figure 6. Figure 7. Skip to main content.

This browser is no longer supported. Download Microsoft Edge More info. Contents Exit focus mode. Is this page helpful?

Please rate your experience Yes No. Any additional feedback? There it suggests itself to buy more memory in next hardware onlineshop or local computer store and spend the operating system 4GB, 8GB or even soon it will sound ridiculous … 16 GB RAM.

In the theory this may be correct regardless of the larger load for the voltage regulators, however in practice it looks unfortunately somewhat different. Because operating systems like e. In the following table some operating systems are specified with the possible memory addressing. With 64 bit operating systems there are some other conditions for the maximum memory amount, like e.

Therefore here is again a short explanation about 4 gigabyte memory in Windows 64 bit and 32 bit operating systems from the previous memory reviews.

In the meanwhile the RAM size is fortunately much higher, because with bit operating systems like e. The following table compares the number of processors and the amount of physical RAM that are supported by the xbased versions of Windows Server and by Windows XP Professional x64 Edition to those that are supported by the bit versions. Operating system. Number of processors. Physical RAM. Microsoft Windows Server , Standard Edition.

Microsoft Windows Server , Standard x64 Edition. Microsoft Windows Server , Enterprise Edition. Microsoft Windows Server , Enterprise x64 Edition.

Microsoft Windows Server , Datacenter Edition. Microsoft Windows Server , Datacenter x64 Edition. Microsoft Windows XP Professional. Memory allocation settings. The following table compares the memory allocation settings that are supported by the xbased versions of Windows Server and Windows XP Professional x64 Edition to those that are supported by the bit versions.

Collapse this tableExpand this table. Total amount of virtual address space. Amount of virtual address space per bit process. Amount of virtual address space for the bit processes. Not applicable. Amount of paged pool memory. Amount of non-paged pool memory. Size of system cache.